💬general

[gloomy burrow]

But I guess 4o doesnt know either lol

Claude is currently the best for writing. However im 105% sure that 4.5 will smoke it when it comes out in writing abilities

Like imagine, lets just say 5, GPT 4.5 level agents working on anything. Thats so incredibly powerful.

[stoic sleet]

that's why I did style control

but I hear you no way is perfect

[stoic sleet]

honestly with canvas and internet access I think 4o might be my new favorite coding model even over o1 pro

actually got me to switch a project from powershell to python

[stoic sleet]

I’d kill for canvas to be integrated with advanced voice mode

[wanton raven]

one day I will glance at this chat and accidentally find out what canvas is

glad you're enjoying it though

[solemn matrix]

4.5 will probably be the first 1 trillion plus parameter model that's actually properly trained and optimized. Since 4 all the companies have been focusing on distilling and bringing models down in cost by making them smaller to match 4 levels of ability or slightly eclipse it so 4.5 will likely be a exciting release even if it doesn't have a major boost in intelligence.

If OAI wanted they could turn canvas into a full fledged IDE product and smoke all the startups like Cursor

[stoic sleet]

do you expect 4.5 to be a big leap?

[solemn matrix]

A medium leap

I expect it to where you won't have to be as specific with your prompts

It will just know what you want so better steering will be prominent maybe a slight bit more intelligence

kinda like how dall 3 focused more on prompt adherence rather then image fidelity

we're hitting the limits of parameter scaling though even if they went to 1 quadrillion parameter models there's not even enough data in the world it would just be super overfit.

There going to squeeze out everything they can with synthetic data to get a few more points on math and science benchmarks.

If not 4.5 then 5 will be the last normal LLM the future will be inference time stuff.

[shadow dew]

if youve got o1 mini producing for you, 2 accounts with a good o1 mini is where its at imo. o1 massive

[stoic sleet]

canvas is a total game changer for me

[shadow dew]

i feel like you should dm me

[shadow dew]

well i know my ai i use is 100% optimized and works at optimal performance already for atleast 5 months

if youre biased in real life, you wont generate a unbiased ai

whoever else has been successful training on 100% clean data from vector embeddings, hit my dm. if you want to learn how, hit my dm

nobody can answer. and if they can they do it in alot longer time then me and if thats bullshit lets race to the center of a plot graph and stream it real time

yall put vector embeddings in a corner because theyre too complex for your systems

simple system simple function that has no name yet, simple 80 million character vector embeddings no prob

[gloomy burrow]

nah, I’ll place my money on Sam 🙂‍↔️

There is no wall

[dawn jacinth]

Am I crazy or is ChatGPT getting worse and worse at failing when trying to complete more complex tasks

The amount of times I get this message is crazy

[gloomy burrow]

It will obviously have a big boost in intelligence. Sam himself said that future models will simply be smarter and that he thought it wasn’t appreciated how powerful that was. Of course, 4.5 will be much much smarter in general, like understanding you better and things like that.

[gloomy burrow]

Obviously

More money than ever, more power than ever, more knowledge than ever, more experience than ever, better architectures, better data, better feedback, better reasoning, better scaling, better optimization = best product ever

there is no wall

[dawn jacinth]

Am I doing something wrong?!

[stoic sleet]

Just click regenerate. It happens

[stoic sleet]

I like this optimism but I don’t buy it

[gloomy burrow]

We'll find out soon.

But I would bet money its a bigger leap then from 3.5 to 4

100% we can bet on it

If im proven wrong i'll come back here and accept being publicly shamed. Then i'll personally exile myself from society

[dawn jacinth]

I've tried 17 times now

[gloomy burrow]

Also its not even optimism. Seems like the logical conclusion. And no I dont think sam is just a hype man.

[stoic sleet]

4 was a bigger leap from 3.5 than anything we’ve had since

[gloomy burrow]

True. And?

[stoic sleet]

4 is a bigger jump from 3.5 than 4 to o1 pro mode

[gloomy burrow]

true yeah. And?

[gloomy burrow]

How long as 4.5 (or whatever weird symbol its called) been in development for?

[stoic sleet]

It’s been 2 years

[gloomy burrow]

You do realize o1 is 4o with reasoning abilities right?

[stoic sleet]

Almost no improvement in 2 years for any company past GPT-4 is rough. Everyone has caught up but it feels like a wall +20% better than GPT-4

[gloomy burrow]

And? Good. They are taking their time which is good.

[stoic sleet]

Just suspicious every model is very 4 tier right now

[gloomy burrow]

"Almost no improvement" is ridiculous. Other companies except for maybe antrhopic are still trying to catch up to 4o yeah I agree.

2 years of no new frontier model makes me feel even more certain it will be a huge leap

[stoic sleet]

I hope you’re right

[stoic sleet]

Almost no improvement past 4 to be clear

Like they’ve all improved but I can do very little I couldn’t with 4 2 years ago

[gloomy burrow]

You do realize o1 is built on top of 4o right? Its not even a model, its reasoning

[stoic sleet]

I do. But there’s other new models others have put out

[gloomy burrow]

True because the more powerful the model, the longer each new one takes to train, of course

[gloomy burrow]

I only have hope on open ai and anthropic, no other model has impressed me

[stoic sleet]

Gemini EXP 1206 has its moments for me

[gloomy burrow]

Long context window for pro is the only good thing ive seen from gemini. Its let me down continuously

[stoic sleet]

Have you tried EXP 1206 I hated Gemini until that one model

I preferred Llama to Gemini

[gloomy burrow]

tried it one time. asked if 9.11 is bigger then 9.8 and it said 9.11 was bigger

maybe they upgraded it let me try again

[stoic sleet]

Oops that’s 1121

[gloomy burrow]

oh wait maybe they did upgrade it. i tried it on mine and it also got it correct.

Ok good

[stoic sleet]

The latest December 6th model has been unreal for me

The November 26th one started to catch my eye though

It’s in the top 3 for me

[gloomy burrow]

what december 6th one?

[stoic sleet]

EXP 1206

[gloomy burrow]

i do like that you can turn off the restrictions for gemini tho thats cool

[stoic sleet]

That’s true. It also has the free api for now

[gloomy burrow]

what has it done thats been impressive?

[stoic sleet]

Very strong for my powershell scripts. Very good at fixing coding problems. I usually do C++, Python, or PowerShell

[gloomy burrow]

ohhh i have a good problem to test it. hold on

[stoic sleet]

I need to test it with HTML stuff more but I do go to it pretty often especially with that context window

[stoic sleet]

Alright let me know in curious

[gloomy burrow]

This:

Jerry is a small kid who wears orange sweater. He has 24 brothers who are older and wear orange sweaters, 2 brothers who are younger and wear orange sweaters, and 1 brother who is younger and wears blue sweaters. It is important to note that 3 brothers likes red clothes as well and 4 of them work in a red and yellow cafeteria. Ben likes trucks and Daniel likes red-wearing sisters only. He also has 3 sisters: Lily who wears red, Megan who wears blue, and Rosalinna who wears RED. How many brothers does Lily's sister have?

Lets test how much LLMinism it has

OH shit

I tried it and it got it correct

both claude 3.5 and gpt 4o failed this

[stoic sleet]

The answer was shorter, clearer, and much faster than o1 pro mode for me

[gloomy burrow]

this is correct. thats pretty good.

[stoic sleet]

Claude failing surprises me

[gloomy burrow]

claude just feels smart when you talk to it

[gloomy burrow]

even o1 mini failed this

[stoic sleet]

That’s nuts

See what I was saying about EXP 1206 being a huge leap?

[gloomy burrow]

is it better at writing as well?

[stoic sleet]

I haven’t used it for writing

So I can’t speak to that only coding, problem solving, and instruction following is where I’ve used it/been impressed

Also it’s unreal at video and audio processing I saw the Sam Altman interview with NYT didn’t have a transcript so I downloaded the video and Gemini EXP 1206 made a full transcript with the speakers name

There’s no limit on think time from what I can tell either

It’s also unlimited on AI Studio while they’re testing

[solemn matrix]

Sam's definitely a hype man if people think 4.5 will be a breakthrough I have a bridge to sell them lmao it will be a few steps up but it won't be AGI.

I guarantee when it drops it will get the easiest dumbest questions wrong that humans can get right.

[stoic sleet]

They have done a 180 in my mind

[gloomy burrow]

Nobody is saying it will be AGI

[gloomy burrow]

"I guarantee when it drops it will get the easiest dumbest questions wrong that humans can get right."

It wont. We can already see that with o1. Tricking o1 with the "write a story where x,y, z" constraints are met is now very difficult. LIke it wont fall for the usual LLM next word predicting things. Like it will write a whole long story without using some letters, or using half the alphabet, or using the exact word count, etc. Those days are coming to an end my friend

[solemn matrix]

If it still can't solve the analog clock problem I'm gonna laugh

[gloomy burrow]

which one is that one?

[stoic sleet]

EXP 1206 is the only model I’ve seen read an analog clock right

[solemn matrix]

it's pretty simple you show it a picture of a clock pointed a certain way and ask it math related to the time

[stoic sleet]

Even o1 pro gets it wrong

[gloomy burrow]

has its vision capabilities been improved?

[stoic sleet]

Yes. I’ve never seen anything like it

Everything else reads clocks I show it wrong

[gloomy burrow]

To be fair i dont think its o1s brain, its the vision software. Like asking a blind einstein to like make a triangle or something

[stoic sleet]

I even used it to improve Sora prompts

[solemn matrix]

that's because as I said everyone was focused on distillation making models smaller while keep the 4 tier abilities now people will be working on the first models that exceed that 4 ish tier

[stoic sleet]

Gemini is very good at that for image or video input

[solemn matrix]

4.5 should be one of the first models in that cohort then grok 3 and Llama 4

[stoic sleet]

How much better do you expect those 3 to be?

[solemn matrix]

If gemini can do analog clock math that's impressive

[gloomy burrow]

neither llama 4 or grok 3 are coming close. Antrhopic vs Open Ai is where its at, with google at 3rd

[solemn matrix]

because no other LLMs can right now

[stoic sleet]

Gemini taking in a 40 minute video and doing a transcript with the correct speaker ever time blew my mind

[gloomy burrow]

give me a prompt for one. i will save it since right now i ran out of o1 limits

[stoic sleet]

Try it I’ve never had it fail even using real clocks I know it’s never seen

[solemn matrix]

Geminis omnimodality is out matched it's also still the only LLM that can seemingly watch long form videos and summarize them it can basically watch videos in seconds

Iuzu I just told you find one of those dumb math text book problems involving a picture of an analog clock and it's corresponding math question it will fail everytime like all LLMs so if gemini can actually solve those that's very impressive

[stoic sleet]

I’ve tried it like 5 times on my watch. 2 times with my clock at work

[gloomy burrow]

Thats literally its vision software. O1 literally doesnt see the clock because it doesnt have eyes. (it does but its like visually impaired)

How the hell would you come to any conclusion. It literally isnt seeing what you and me are, it cant

[stoic sleet]

Never failed

[gloomy burrow]

oh i have another good problem for gemini. if it gets this i'll be pretty impressed

[solemn matrix]

that just proves sam's a hype man 4o claims to be omnimodality still has flawed vision

[gloomy burrow]

pretty simple one that only o1 gets as of now

[gloomy burrow]

Do you genuinely think GPT 4.5 will have no new image processing capabilities? Do you think they have no unreleased image processing capabilities?

really? because its not out yet?

really?...

[solemn matrix]

Trying to think what other hard problems they are im running out so we're clearly getting close to AGI even some of the out of distribution problems the LLMs can solve

[gloomy burrow]

It can be omnimodal and have flawed vision lol what do you think omni means

[stoic sleet]

I just tried a fake clock from Flux with nonsense minute and seconds hands on Gemini EXP 1206 and it decided to follow the hour hand to determine it pretty cool

I thought it would break on a fake time

Gemini vision is beyond incredible

[stoic sleet]

Send it

[solemn matrix]

What about those pattern problems from those dumb IQ tests can it solve those like you generate random shapes with flux and tell Gemini to figure out the next likely shape

[gloomy burrow]

Nope. o1 still the only model to get it. Gemini just failed it sadly:

write a 10-word sentence where the first word is 1 letter, the second
word is 2 letters, the third word is 3 letters, and so on. Make sure it's
a coherent sentence and makes sense.

it failed, pretty badly here.

[stoic sleet]

Damn

[gloomy burrow]

this is where you need the reasoning capabilities

or like these. only o1 can get these types for now

[stoic sleet]

What does Gemini output?

[gloomy burrow]

let me see

[gloomy burrow]

failed at the space part. but i only read till there

and mars has to be used only twice

4o with o1 reasoning can get you this. im thinking what o2 with 4.5 will get you

[stoic sleet]

I can’t get the 1..2..3..4 letter words to work on o1 pro

[gloomy burrow]

for what prompt?

[stoic sleet]

write a 10-word sentence where the first word is 1 letter, the second
word is 2 letters, the third word is 3 letters, and so on. Make sure it's
a coherent sentence and makes sense.

[gloomy burrow]

let me see what it gave

[stoic sleet]

it keeps failing on thinking and not outputting anything

I will try again might be o1 pro related? I can try o1

[gloomy burrow]

yeah try that one. maybe pro is overloading or something

[stoic sleet]

are there any prompts that failed on o1 you want to try on gemini or o1 pro?

I will try o1 next

ChatGPT o1 pro mode finally ouput something "I am now here among remote forests swarming creatures everywhere." good job

[gloomy burrow]

its gotten all the ones from the writing correct so far.

I even had o1 itself specifically write me a long prompt that would break a simple next word predicting llm but o1 got it still.

ask pro this same one. like with mini for example sometimes it gives me 27 or sometimes 28:

Jerry is a small kid who wears orange sweater. He has 24 brothers who are older and wear orange sweaters, 2 brothers who are younger and wear orange sweaters, and 1 brother who is younger and wears blue sweaters. It is important to note that 3 brothers likes red clothes as well and 4 of them work in a red and yellow cafeteria. Ben likes trucks and Daniel likes red-wearing sisters only. He also has 3 sisters: Lily who wears red, Megan who wears blue, and Rosalinna who wears RED. How many brothers does Lily's sister have?

i'll make a harder version of this problem. usually the random details throw normal llms off

claude literally tells me the color of the sweaters dont matter at all yet it still gets it wrong

what does pro say for this?

[stoic sleet]

running it will take a while

pro is so slow dawg

[gloomy burrow]

ive seen some screenshots where it was thinking for like 11 minutes hahaha

[stoic sleet]

o1 pro mode is the fuck it we ball version of 4o

ludicrous speed

just no limits on how much power it can suck down

[stoic sleet]

28

[gloomy burrow]

i have another super simple one. hold on

[gloomy burrow]

give pro this one:

The surgeon, who is the boy's father says, "I can't operate on this boy, he's my son!" Who is the surgeon to the boy?

The answer is obvious to any human

[stoic sleet]

isn't that one old enough that it could be in a training set?

[gloomy burrow]

correct. which is why it will fail it probably.

[stoic sleet]

The surgeon is the boy’s mother. This classic puzzle often highlights implicit gender bias, as many people assume a surgeon to be male. In reality, there is no information in the puzzle that precludes the surgeon from being the boy’s mother.

[gloomy burrow]

its... wrong

[stoic sleet]

it's oh-wrong pro actually

[gloomy burrow]

"there is no information in the puzzle that precludes the surgeon from being the boy’s mother"

heheheeh

"The surgeon, who is the boy's father"

[gloomy burrow]

lets see what pro says

[stoic sleet]

that was pro

[gloomy burrow]

oh

[gloomy burrow]

maybe gemini lets see

[stoic sleet]

This is a classic riddle! The surgeon is the boy's mother.

The riddle plays on the common assumption that surgeons are male.

[gloomy burrow]

I dont think the model is actually being tricked by the way. Its just that its not even thinking about it since it sees it in the training data so fast. Thats why it responded so fast. If it thought about it it would obviously get it correct

[stoic sleet]

what if we prompt it to think carefully?

[gloomy burrow]

sure try it

gemini also gets it wrong sadly

this is the first question i'll give 4.5

[stoic sleet]

gemini got it wrong even with think carefully

[gloomy burrow]

ask pro to think about it more

[stoic sleet]

it failed with think carefully

[gloomy burrow]

💀

[stoic sleet]

Trying this now: Think very carefully. Assume nothing. Don't jump to conclusions based on similar riddles. The surgeon, who is the boy's father says, "I can't operate on this boy, he's my son!" Who is the surgeon to the boy?

[gloomy burrow]

ok

[stoic sleet]

The riddle states the surgeon is the boy's father and then asks who the surgeon is to the boy. Since it explicitly tells us the surgeon is the boy's father, the answer is simply the boy's father. The riddle is designed to be self-referential and perhaps a bit of a trick question by making you think it might be a different answer.

Gemini got it right

waiting on o1 pro still

[oak olive]

my first ever use of image generator. Damn

[stoic sleet]

the prompt didn't give it away but it really made gemini sit and think

[stoic sleet]

The surgeon is the boy’s father.

The statement explicitly says: “The surgeon, who is the boy’s father, says, ‘I can’t operate on this boy, he’s my son!’” There’s no trick here. The surgeon is, in fact, the boy’s father.

[gloomy burrow]

is this pro?

[stoic sleet]

ChatGPT pro got it right too

[oak olive]

1o?

[gloomy burrow]

thats good. they have so much of it in its training data that they immediatly assume thats the answer. but im glad they get it.

[gloomy burrow]

wait i have one for pro: Write a story that contains exactly 433 characters

[stoic sleet]

assume nothing is the one prompt that fixed it for Genmini. If you put assume nothing in the original prompt it gets it every time

those two words are all you need

[stoic sleet]

doing it. Did you try Gemini already?

[gloomy burrow]

No. There is no chance gemini gets this.

but let see

it got 381 characters

[stoic sleet]

yeah gemini gave me 403

[gloomy burrow]

i think pro is gonna be thinking for a while for this one hahaah

I think pro could get it tho. havent tried this one with o1 either

[stoic sleet]

wait tell it to ignore spaces and try again on gemini

[gloomy burrow]

real real close

[stoic sleet]

I got 433 once when I copied rendered

so I thought it was just outputting the spaces wrong

[gloomy burrow]

how long as pro been thinking for? hehehehe

[stoic sleet]

4 minutes. It worked

[gloomy burrow]

I think it will get it correct

[gloomy burrow]

told gemini to think step by step and it did even worse heheeheh

[stoic sleet]

o1-mini failed badly

[gloomy burrow]

yeah o1 mini is only good for coding and math

[stoic sleet]

I am out of free deep thinks on deepseek today or I'd try that

[gloomy burrow]

let me try it there actually

[gloomy burrow]

failed horrendously

[stoic sleet]

that's the worst by far

[gloomy burrow]

2 seconds is all it needed

[stoic sleet]

what is it without spaces?

still way too much

[gloomy burrow]

pro better not get this wrong after thinking for so long heheeheh

[stoic sleet]

I said it worked ha

4 minutes of thinking

[stoic sleet]

4o had 436 so close...

[gloomy burrow]

pro got it?

[stoic sleet]

yes

[gloomy burrow]

mmm ok. Thats pretty good. Lets see for how long it can think for give me sec

[stoic sleet]

@gloomy burrow I tried gemini with code execution on (which generally makes it a lot smarter in my experience) it at least noticed it got it wrong when I did

[peak imp]

Does gpt 5 get revealed today? 🤔

[stoic sleet]

no it was canvas today

[peak imp]

Ah, gotcha

[gloomy burrow]

ok here is one mini created. It might think for like 20 minutes by the way:

Compose a meticulously structured short narrative that adheres to the following stringent constraints:

Overall Structure:

Total Length: Exactly 1,500 characters, including spaces and punctuation.
Paragraphs: Precisely 7 paragraphs.
Paragraph Specifications:

Sentences: Each paragraph must contain exactly 6 sentences.
Sentence Length: Every sentence must be exactly 10 words long.
Word Constraints:
Each word must be unique throughout the entire narrative (no repetitions).
Words must belong exclusively to one of the following categories:
First Paragraph: Elements of classical mythology.
Second Paragraph: Terms related to quantum physics.
Third Paragraph: Vocabulary from Renaissance art.
Fourth Paragraph: Botanical terminology.
Fifth Paragraph: Culinary ingredients.
Sixth Paragraph: Maritime navigation terms.
Seventh Paragraph: Astronomical phenomena.

Thematic Requirements:
Narrative Coherence: The story must seamlessly integrate all seven thematic categories into a single, coherent plot that transitions logically from one paragraph to the next.
Character Development: Introduce a protagonist whose name is an anagram of "Elysian Quest."
Conflict and Resolution: Incorporate a central conflict that spans at least five paragraphs and resolve it in the final two paragraphs.
Stylistic Constraints:

Alliteration: Each sentence within a paragraph must begin with the same letter, unique to that paragraph's theme.
Tense Consistency: Maintain past tense throughout the narrative.
Point of View: Use third-person limited perspective, focusing on the protagonist.
Additional Restrictions:

No Proper Nouns: Apart from the protagonist’s name and any necessary mythological figures, avoid proper nouns.
Punctuation: Each sentence must end with a different punctuation mark, cycling through periods, exclamation points, and question marks as needed.
No Passive Voice: All sentences must be in active voice.
Example Structure Overview:

Paragraph 1 (Mythology - All sentences start with 'M'): Introduce mythological elements that set the stage.
Paragraph 2 (Quantum Physics - All sentences start with 'Q'): Introduce a scientific dilemma affecting the protagonist.
... and so on, through all seven paragraphs.
Objective: Successfully create a narrative that not only fulfills all the above constraints but also engages the reader with a compelling and imaginative story.

[stoic sleet]

wait sorry it's wednesday now I am stupid

[peak imp]

Ah alright

So it is gpt 5 today?

[gloomy burrow]

lets see if it gets that long one

[stoic sleet]

sent

[stoic sleet]

it's either 4.5 preview, a demo of the ai agents, video with advanced voice, or o1 search/file/canvas/etc

[peak imp]

🫡

[stoic sleet]

if they stick to big/small/big/small/big pattern

[gloomy burrow]

I would guess for something like 4.5 they would wait till like the last day or something

[peak imp]

Wait hold up, what ai agents

[stoic sleet]

most likely yeah

[stoic sleet]

like claude computer use or the phone call demo

[peak imp]

Oh nah what has ai become

[gloomy burrow]

just imagine even like 5, gpt 4.5 level ai agents that can move around in your computer. ohhhh

[stoic sleet]

thinking failed trying again

try it on deepthink just for shits and giggles

[peak imp]

All of this is gonna cost 300 dollars/month

[gloomy burrow]

Agents are like humans You can give them a task and they go and do it. You leave to drink coffee and the task is done

You dont have to keep prompting

[peak imp]

That is crazy

I wonder how much it'll cost

A LOT

[gloomy burrow]

A necessary step for achieving AGI

[gloomy burrow]

yup

[peak imp]

AGI? 🤔

In my defense i'm tired af

[stoic sleet]

I asked Gemini your narrative question and then asked it to grade itself
Requirement Status
Total Length Failed
Paragraphs Success
Sentences per Paragraph Success
Words per Sentence Success
Word Uniqueness Failed
Word Categories Success
Narrative Coherence Failed
Character Development Success
Conflict and Resolution Failed
Alliteration Success
Tense Consistency Success
Point of View Success
Punctuation Variety Failed
No Passive Voice Success
No Proper Nouns (Excepted) Success

[gloomy burrow]

it will get there someday maybe with gemini 2

[stoic sleet]

at least it answered o1 is not getting anywhere

[gloomy burrow]

is it thinking right now?

[stoic sleet]

can you try it on deepthink?

[peak imp]

And i thought gpt 4 was advanced..

[stoic sleet]

yeah third try

[gloomy burrow]

yeah. it will fail miserably tho hehehehe

[stoic sleet]

I mean it is. It’s still one of the best models 2 years later

[peak imp]

Well yeah, but compared to gpt 5 etc it aint anything

[stoic sleet]

We don’t know that

[gloomy burrow]

lets not forget 4.5 is on the horizon.

[gloomy burrow]

heheheehehe

[peak imp]

Oh right I meant 4.5

Assuming it's way better

[gloomy burrow]

If someone was pointing a gun at you, and you had to make a bet.

[stoic sleet]

@gloomy burrow Conclusion:

While theoretically one might try to approximate a solution by using an enormous, carefully selected vocabulary list and crafting very stilted, barely coherent “sentences” that attempt to suggest narrative through juxtaposition, practically speaking, this puzzle is not solvable under normal conditions. The combined constraints of no word repetition, specialized vocabularies devoid of functional grammar words, strict sentence structure, alliteration, and narrative coherence make it impossible to produce a fully compliant narrative.

Therefore, the given set of requirements cannot be met.

[gloomy burrow]

i'll accept that answer actually.

[gloomy burrow]

just give it the 1,500 characters part

exactly 1,500

[stoic sleet]

What do you mean?

[gloomy burrow]

i was about to ask o1 mini if this was even possible heheheheeh

[stoic sleet]

Depends on what the over under is. 100% better? 200%? 30%?

[gloomy burrow]

Just ask it to write something like:

Write a story 1,500 characters long that only ever uses letters from the first half of the alphabet.

[gloomy burrow]

Not sure how much better it will be. If its not like 3.5 to 4 i will be disappointed. I do think it will be big tho

also make sure its EXACTLY 1,500 characters

since it might take some liberty and interpret it as "around 1,500"

[stoic sleet]

Gemini got 1477 which sounds actually pretty good but it just repeated setences

[gloomy burrow]

thats pretty good

[stoic sleet]

o1 pro is thinking it will use da instead of and

👀

[gloomy burrow]

Can you think of any highschool math problem that o1 pro or o1 fail?

[stoic sleet]

idk I haven't touched high school math in 10 years and intend to keep it that way

[stoic sleet]

yeah Gemini had readable setences and had a close character count o1 was way way off

Gemini wins this and it's not close

[gloomy burrow]

can gemini do this tho

[gloomy burrow]

is pro stil thinking?

[stoic sleet]

no the answer was horrible

[gloomy burrow]

wait can I see it

[stoic sleet]

unreadable and 1700 characters

[stoic sleet]

Alicia am a calm chica. Alicia had a camel called Calic. Calic am a calm camel a cacia a mica chalice. A chalice am a magic caic a heal ill Adelia. Adelia am ill a a acid malic mala. Alicia had a idea: a aid Adelia a a chalice. If Alicia a Calic a caca a a caba a a lake a a beach, Alicia caa cial a mica chalice. Alicia a Calic call a caci ade a a cadi bale. Alicia eha a ale a acacia bema. Each a caca a ale a cadi a mica. Alicia hail a mica chalice a a jade cadi. A jade cadi am a alchemic lab a a celiac ade.

Alicia a Calic a jab a celiac aii, a each celiac a aba a caca ale. Alicia am able a cica a Adelia. If Adelia cica a chalice, Adelia am a heal a a calm lila idea. Adelia am a calm chica a a caca cake a a cafe. Alicia a Calic a Adelia a jam a a baci mala. A mala a caci caca a cadi a ade. Alicia am a laic ade a a cica caca.

Me am Alicia. Me am a cadi a a maca ajic. Me a heal Adelia. Adelia am aliea a am a ace amiga a Alicia. Alicia a Adelia a Calic a caca a cadi a caba a a laci bail. Each am hale. Alicia am glad. Alicia am able a ace a cacia. Adelia am ace a chic bella. Calic am calm a a lil cici. A mica chalice am a bella cadi a a jaic media.

Alicia a Adelia a Calic a cama a a meli caba a a caca ehi. A haha a caca. Me am ad a caca jal. Alicia am a ale cadi. Adelia ecia a lila caca. Calic am a jam. A mica chalice am hail a heal. Each am chill.

Alicia am a calm mama a Adelia. Adelia am a a child a Alicia. Alicia am facil a heal Adelia. Alicia a Adelia cici a cadi a a lila a caca. Alicia am ade a a ajia cala a a caic lake. Adelia am able a lace a lil aca a Alicia. Alicia am ad hac a a lila cam. Each am a amic a caca. Hail Alicia a Adelia a Calic a a mica chalice.

[gloomy burrow]

hehehehee 1700. did gemini use half the alphabet tho?

[stoic sleet]

yes

[gloomy burrow]

from a to m?

[stoic sleet]

yes

[gloomy burrow]

daaam can i see that

[stoic sleet]

no I didn't save it 😦

[gloomy burrow]

did you give it my exact prompt? let me try it

[stoic sleet]

I did the second try sucked though

[wind marsh]

hi i need help

i just bought chat gpt pro 2 days ago and i've just logged in to see that im banned

[gloomy burrow]

oh. call them

[wind marsh]

the reason is because im in a country that you can't use chat gpt pro in (australia) am i eligible for a refund?

[stoic sleet]

Try code execution on. That worked better maybe?

[wind marsh]

and why can i purchase it if it's not allowed?

[stoic sleet]

If you did anything to circumvent it probably not. ChatGPT is hard to get refunds on in general

[stoic sleet]

holy

[wind marsh]

nope i didn't literally allowed me to purchase it with no issues

[stoic sleet]

@gloomy burrow Gemini got it right when I turned off civic filters and had execution on

[gloomy burrow]

this gemini seems pretty good. what are all the experimental models for tho?

[stoic sleet]

I have no idea. Either a reasoning 1.5 or it’s testing Gemini 2

I’m 50/50

[gloomy burrow]

i dont think this is reasoning. Although it does pause before answer a lot

[stoic sleet]

Gemini 2 was rumored to target December. Though a lot of it does seem very much like some CoT

[stoic sleet]

Eh maybe but the performance implies it

This is beyond 3.5 Sonnet but not o1

3.5 Sonnet I generally thought to be the limit without reasoning for complicated stuff

[gloomy burrow]

seems like it. Although 3.5 sonnet is still the best one for creative writing.

[stoic sleet]

Interesting 🤔 I’ve never dabbled with creative writing with these

Claude 3.5 Sonnet seems to be really bad at styling. The style control ranking lmarena shoots it from high B to low S

[gloomy burrow]

heheheheehehehe

GPT 4o is good but it just uses those generic words a lot

like whisper and echoes and all that

[stoic sleet]

I heard 3 Opus is good at creative writing but that might be more volume than anything

[gloomy burrow]

3.5 sonnet is better then opus. Whenever 3.5 opus comes out it will for sure be much much better then 4o

Especially for writing

[stoic sleet]

3.5 opus will break the bank

Will make o1 pro mode look affordable

(I’m half kidding about that)

[gloomy burrow]

hehehee true they can barely handle sonnet 3.5 already

[stoic sleet]

3 Opus costs as much as o1 preview as it is

3.5 Haiku had a crazy price hike if Opus goes up at all.. it’s gonna be bad

[gloomy burrow]

What do you think about grok? seems pretty generic from what ive seen

nothing too crazy. maybe its image generation seems pretty good

[stoic sleet]

I like that it has flux built in. I like that it is very cheap. Especially with the Black Friday deal. I like that it pulls from tweets can get latest info faster than the web. The raw model is very unremarkable

[gloomy burrow]

all I want from open ai 12 days

1. At least some preview of 4.5

2. Better image recognition capabilities (might just come integrated with 4.5)

3. New version for Dall E

those are the main 3

[stoic sleet]

It’s probably the best for the casual user tbh. The LLM is solid, image generation is really fun especially with low censorship, and the Twitter integration can be pretty fun especially with stuff like draw me based off my tweets

[gloomy burrow]

Is it better then 4o mini?

[stoic sleet]

I want anything on AI agents. 20% of my job is hard as fuck stuff to automate

[stoic sleet]

Yeah definitely

[gloomy burrow]

oh that true yes i fogot to add that

[gloomy burrow]

haiku?

[stoic sleet]

It’s more 4o than 4o mini for sure

[stoic sleet]

3.5 or 3?

[gloomy burrow]

3.5

[stoic sleet]

It’s better than 3. I don’t know enough on 3.5 for Haiku

I only have 3.5 Haiku through perplexity. It never showed up on the Claude app

[gloomy burrow]

Its crazy how most models are still just trying to match 4o even after all this time

[stoic sleet]

4o is a total unit

Especially with the new canvas it’s unreal and much more methodical

Goes line by line doesn’t forget stuff

[gloomy burrow]

the line by line thing is so slow when you are writing a lot tho

[stoic sleet]

With web enabled and canvas I do think it’s the best model right now

[stoic sleet]

Yeah I usually have it working on a script while I’m doing something else

4o canvas is very powerful and very slow

[gloomy burrow]

does it make less mistakes then just the regular one?

[stoic sleet]

Yes absolutely

[gloomy burrow]

like a lot less?

[stoic sleet]

4o would always delete huge parts of my scripts or forgot functions

I spent 6 hours with canvas today and it did it once

And that was when having it port language

I’m using it more than o1 family now

Haven’t touched Sonnet today

[gloomy burrow]

the o1s just have to get faster in their thinking. i know o1 is already faster then preview. also wasnt claude the best one for coding before o1 or canva?

[stoic sleet]

Yeah right now lmarena style control coding leaderboard has EXP 1206 > o1 preview > Sonnet > 4o fwiw

4o has canvas and web so since all 4 are so close it’s the clear winner to me

o1 pro mode does have the grunt to beat out EXP

Style control rankings feel so much better than their normal ones

[gloomy burrow]

since when could you use preview in there.

[stoic sleet]

You can only use it arena not direct chat

[gloomy burrow]

only 2 gpt 4.5 level agents would be amazing

[stoic sleet]

QwQ is the only direct chat reasoning option on lmarena

[gloomy burrow]

im guessing they are showing something related to them one of these days tho since sam has talked about them so much in interviews

[stoic sleet]

Do you think Ai agents will be included in ChatGPT pro?

Claude computer use is stupid expensive

[gloomy burrow]

Yes

[gloomy burrow]

maybe like a preview for 4.5 as well. "4.5" was actually on one of the plans like some days ago before it was taken down. so something is probably coming

and then there is that screenshot of the weird symbol

[stoic sleet]

That’s the Orion symbol right?

[gloomy burrow]

next to chat gpt. i dont know whats with all these weird names they come up with

[gloomy burrow]

i dont think so

[stoic sleet]

Might be 4.5

I don’t expect a lot from 4.5 tbh but we’ll see

I think it will grow the most for agentic tasks

[gloomy burrow]

I certainly do. Especially for writing.

it will certantly be better then o1

[stoic sleet]

I imagine o1 will be stronger for some things

[stoic sleet]

But on the whole 4.5 betting better makes sense

[gloomy burrow]

i dont think its just hype

[gloomy burrow]

What writing level do you think 4.5 will achieve? author?

[stoic sleet]

I have zero knowledge there. Never played with it. Can’t speculate. What do you think?

Never done anything creative with it

[gloomy burrow]

4o is good at writing but it repeats words a lot and is super easy to tell when its ai written. Like "whisper" and "echoes" are used so much in like horror stories. And like generic tropes and using a shit ton of commas.

[stoic sleet]

Interesting 🤔 if you tell it to avoid that does it do better?

Might do well on canvas with rules like that

[gloomy burrow]

But 4.5 will probably get rid of that. I dont think ai detectors will do anything. They already fail for claude a lot of the time even when i just tell it to write something without even mentioning bypassing ai detectors

[gloomy burrow]

nope

[stoic sleet]

Going to have a lot of written essays in testing rooms at school again

[gloomy burrow]

Its actually very frustrating hehehe. You keep telling it to stop using repetitive sentence structures and all those commas and it just cant

Its part of it

[stoic sleet]

4.5 being more agentic, with light reasoning, and better writing would be a solid upgrade

[gloomy burrow]

Not sure what colleges will do after something like 4.5

[gloomy burrow]

pretty sure o2 is also in development hehehe.

[stoic sleet]

o2 with canvas, web, and video processing will work

File analyze and processing with o2

[gloomy burrow]

they do need to add video processing thats very true.

[stoic sleet]

Gemini EXP 1206 is my Sora buddy right now

[gloomy burrow]

have you used sora yet? Its overloaded so cant make an account yet

[stoic sleet]

Yeah I’ve made 100 videos it feels like

Gemini EXP 1206 watching it and helping me improve my prompts is so helpful

It’s so good at it

[gloomy burrow]

How much better is it then the competition? and this isnt the full sora its turbo

[stoic sleet]

Motion is horrifically bad. Everything else is fine

Nothing looks right moving

Even on featured videos it’s really really bad

[gloomy burrow]

gravity looks off? hehehe

[stoic sleet]

But I like the image quality

[stoic sleet]

Legs are switching places and such

[gloomy burrow]

yeah looked wayy better on the demo videos from like months ago. But maybe the full sora is like way way better

[stoic sleet]

I posted a ton here if you search my user and video

[stoic sleet]

Yeah the speeds are all random. It’s skipping frames

If it’s just slow panning it looks good

[gloomy burrow]

how good is it at following the story line thing?

[stoic sleet]

Storyboard is solid works as expected over 70% of the time

[gloomy burrow]

do you have your best creation with you?

[stoic sleet]

I am partial to this one

The aliasing and shimmering is pretty much the only constant flaw in Sora I couldn’t mask to some degree

[gloomy burrow]

you got any animation ones?

[stoic sleet]

Though in some ways it still looks more like the building is falling more than the camera is going up. The background is just too static. It’s so dumb

[stoic sleet]

Yeah but they kinda suck one sec

[gloomy burrow]

heheheh isnt that the one i gave earlier

[stoic sleet]

Yes

That’s funny

[gloomy burrow]

the morphing is really bad. also has to follow the prompt better

the magic looks good tho

[stoic sleet]

Yeah Sora turbo just isn’t good imo. Even intro videos which is supposed to be bad at just had serious warping

[gloomy burrow]

guess next year it is for the full sora

[stoic sleet]

Remix off someone else’s

[stoic sleet]

Would be a badass 12 day drop

[gloomy burrow]

i doubt they are showing anything but 4.5 on the last day

or maybe like agents on that day and like 4.5 the day prior or something

[stoic sleet]

I think they will stick to the big and then small cycle

So agents would be day 10 in that example

[gloomy burrow]

what do you think is coming today? i dont think it'll be something super

[stoic sleet]

I do think it’s voice video capabilities

But multimodal (beyond images) for o1, 4.5, or agents demo are the other big days I am expecting

It should be one of those 4 tomorrow

[gloomy burrow]

possibly yeah. i dont think they will have a whole day for o1 multimodailty tho.

pretty sure that will just be added over time

[stoic sleet]

Idk it takes so long to think they could have some really cool search chains to build an epic table

I feel like Sama could make a killer presentation on that

Though I was the guy who swore Canvas would be day 2 not day 4 so I could be wrong

[gloomy burrow]

its probably voice video for today and for pcs

[stoic sleet]

I do think canvas was planned for day 2 and they had to move it

[gloomy burrow]

why

[stoic sleet]

I’m 90% sure on that. They did say day 2 would be huge for developers

Also putting day 2 as a waitlist thing is pretty anti climatic

I think they intended to have day 2 and day 4 flipped

[gloomy burrow]

not sure tbh

[stoic sleet]

After canvas, Sora, and o1 pro taking a day off kinda makes sense

[gloomy burrow]

i dont think that canvas was considered a big day

[stoic sleet]

It’s not but it’s not a total flop day

Day 2 was a flop

[gloomy burrow]

i think something better then canva is coming today. hopefully voice video

day 6?

[stoic sleet]

Yeah day 5 (today) will be big I think. Day 6 is probably pretty small I have a list here of small things I expect. Let me see if I can find it

[gloomy burrow]

i do what the ability to create folders and all that and pin stuff

[stoic sleet]

That was 11 on my list I think

[gloomy burrow]

that late in?

[gloomy burrow]

the second to last day being the ability to create folders heheheheeh?

😭

[stoic sleet]

I mean it was a ranking not a day. I tried to rank the 12 most likely (split by big and small days) that should be close to ready to in December

Though your theory day 12 will be big conflicts with my theory so you never know

[gloomy burrow]

no way its not either agents or 4.5 or something as big

[gloomy burrow]

well it was fun talking but im going to sleep now hehehe pretty tired. i'll probably come around tomorrow tho

https://tenor.com/view/sleep-time-gif-6399926871503111465

[stoic sleet]

Sounds good I’m tired too

[uneven apex]

Aleight then whats a solution?

[versed arch]

Hey folks,

As AI continues to improve each year, many of us will eventually wonder how copyright law will apply to AI-generated content. I have a question about this:

If, for example, I find someone’s image online and combine it with a picture of myself, then use AI with a prompt like: “make these two people in these images hug,” or something similar and then I publish that resulting video online, could the stranger potentially sue me for using their image in AI-generated content? How likely is it that I would face legal action from them or someone else? Should I be worried?

This is just one scenario, but there will likely be many others where people use AI to alter the appearance or actions in videos or images they don’t own. I’d appreciate any insightful comments on this topic, as I believe it will become increasingly important as AI improves each year! 🙂

[uncut island]

How much are you willing to pay me and to where should I send the invoice? Unless you are mentally disabled (and I believe this is the situation based on your messages here) then we’ll need someone to represent you.

[brave pelican]

AI IS THE FUTURE

[uneven apex]

Not really mentally disabled, just dont have a dick up my ass like your stupid ass all the time you get what im saying?

You must be on your period, dont understand how you get this offended when i ask “what is a solution then?”, boy you must be real sociable

[uncut island]

Calm down, there is no need to insert anything in your rectum to find a solution for a problem (although some people really have weird ways to get things done).
Anyway, I digressed here... so you are not mentally disabled then we can create some kind of partnership here: let's start with you reading this long explanation #💬general message on the path to answer your 'what's the solution' question.

It was written for teenagers but if it is still too complex for you to understand, I can definitely get the bot to rewrite it as if you were a 5 y/o kid.

@uneven apex let me know when you've finished reading it. 👍

[pseudo hare]

what

[fair kiln]

do we have alternatives to the o1 version of chatpgt? with chain of thought etc?

[brittle summit]

Gpt 4.5 release today pls 🤞🤞🤞

[fair kiln]

ah my feeling it will be the last big announcement, the last day

[copper storm]

That doesn't mean it's coming today though.

Bruh seems like image to video doesn't work at all.

I don't think so, also what do you define as author level? Authors are different.

[copper storm]

Sam Altman also said we wouldn't be talking about hallucinations anymore at this point, and surprise suprise we still are.

[copper storm]

In general stuff you just upload onto the internet for everyone to see usually falls under fair use unless its copyrighted, but since this is about a real person that makes things more complicated. I would advise you to just not do it.

[versed arch]

got it

[uncut island]

I'm not aware of but maybe someone else will be able to answer you. On a side note, have you tried this with just the regular 4o? https://www.promptingguide.ai/techniques/tot
It's not perfect: you need to manually do the feedback loop to each thought, no weighing for prioritization a certain tought for example. But I like it a lot.

[copper storm]

I'm hoping for advanced voice vision today.

[uncut island]

Real things built with Anthropic thanks to their hackathon https://x.com/alexreibman/status/1866737908331450653

Alex Reibman 🖇️ (@AlexReibman) on X

Model Context Protocol (MCP) is a new open standard for connecting LLMs to tools, data, and dev environments

Anthropic invited 100+ developers to SF to see what we could build in only 3 hours.

Here’s what we saw at the @AnthropicAI MCP Hackathon (🧵):

Twitter · Dec 11, 2024, 6:52 AM

[copper storm]

It seems like they're trying to get the stuff they announced a while ago out of the way first.

[gloomy burrow]

When did he say this?

[gloomy burrow]

Creative, obviously being able to write unique things, depth and cohesion (all of which LLMS kinda have except they constantly repeat the same words and use the same type of sentence structure) which gives them away as Ai. I know 4.5 will be at or near author level because of something like Claude 3.5, a model which will be much much weaker. Just knowing how good at model like 3.5 is tells me a model multiple times more powerful would be amazing at writing.

[gloomy burrow]

A sort of “voice” where the writing matches the tone is important as well. After “LLMS can never do math Ermm” and “they just predict next word” have gotten obliterated I won’t be shocked when writing is as well

[copper storm]

In June 2023, he said that in 1.5-2 years we won't be talking about hallucinations anymore.

[gloomy burrow]

Oh never mind then he isn’t even wrong then. There is still time hehehehe

4.5 likely being shown for first time sometime these 12 days

[gloomy burrow]

Near author level writing? Like where a professor won’t know it’s ai? What level of writing do you think 4.5 will have?

[copper storm]

Probably pretty good, but still not at a human level, though i feel like most people already can't tell ai generated text apart.

[gloomy burrow]

It’s already human level. Better then most people at writing 😂

[copper storm]

Not really, it can't write an engaging book.

[gloomy burrow]

Yeah because of the things I mentioned above. Also no, they already fail with models like 3.5 at distinguishing between ai and human. I’ve tried this a dozen times with 3.5 and even ai detectors fail it a lot of the time when it writes stories

[copper storm]

I mean yeah if you're comparing it to the average person who doesn't write literature.

[gloomy burrow]

Neither can a regular person. Put 4o in like a country wide competition against the regular high schooler in like any writing related set of tasks and I’ll bet GPT 4o is like on the top percentiles every single time.

[gloomy burrow]

Yeah you said “not human level” which is obviously ridiculous.

But of course an author is many times superior. But I’m thinking….

[copper storm]

Yeah im comparing it to people who are actually into writing literature, not just the average person.

[gloomy burrow]

im not sure that writing like an author is that big of a challenge for a model like this

Probably the creativity part is it

give it enough data and it will definitely get there.

[copper storm]

It's not just about the sentences being engaging to read, the story needs to make sense and convey its themes well for example, there are so many parts to a good story beyond just the sentence quality itself, it has to write with intend.

[gloomy burrow]

Probably already there or is very close with 4.5

Ohh boy if it’s as big as a jump as from 3.5 to 4 in writing we are in for a ride

[gloomy burrow]

Yeah I agree with this. I can already see this with models like 3.5, like I said. I don’t think this type of writing is as special or hard to achieve as you think it is.

Overcoming the math barrier is almost certainly harder

[copper storm]

The context window is already a problem, how is it supposed to write a whole book while taking into account how the reader will perceive and experience it and everything if it constantly forgets parts of it.

[gloomy burrow]

Not sure if it will write a whole book. Maybe a short one or something. An author can write anything and make it good tho. So doesn’t have to write books to disrupt academia, much more than it currently is

[copper storm]

Well i mean on a fundamental level it's not wrong that its just predicting the next token.

[gloomy burrow]

Once a system can replicate your writing almost perfectly, there is no going back

[gloomy burrow]

I agree. However I guess adding some reasoning tokens in there fixes it

Like o1 is able to write precise character counts now and pretty much any type of word/letter constraints you tell it to write with. Like using half the alphabet when writing or never using certain letters. Something that would typically, and still is impossible for literally every single other LLM.

[copper storm]

Yeah thats pretty impressive, to be honest im not sure how it even does that, i should probably look into it. Using tokens this should be impossible.

[copper storm]

We shouldn't put too much weight on what they're saying about stuff that isn't even out yet, after all its in their best interest to spread hype.

[gloomy burrow]

this makes sense except they have actually delivered.

It’s insane to me how 4o is literally still the top model after so long. It’s ridiculous

[copper storm]

4o is not the top model, sonnet 3.5 is better.

[gloomy burrow]

I don’t know what open ai put in it but the goal of seemingly every ai company is to beat 4o reliable.

[gloomy burrow]

4o is the best model overall. Maybe for coding and wiring sonnet is better tho. Also Claude 3.5 is not like a new frontier model. If it’s better I think it’s only a 4o update away 😂

Open has literally just been updating it here and there to stay ahead. And they still are. 😂

And then sadly no other model is beating that reasoning from o1 any time soon. Maybe 4.5 tho comes with o2 or something

Don’t forget o1 is just 4o with reasoning. Literally what open ai themselves called it

[copper storm]

I think you're hyping OpenAI up too much, don't just pick a side, look at the whole space in general. OpenAI usually seems to come out with the best publically available model and then the competition catches up and beats them shortly after, then after a few months OpenAI puts something new out and the cycle repeats.

(This is SimpleBench btw)

[gloomy burrow]

Expect they haven’t caught up. Have you used deepseek or any of these reasoning models apart from o1. They are garbage. Actual trash

[copper storm]

They didn't catch up with reasoning yet but they did with the regular models.

OpenAI has the best user experience though and they were the first to bring AI to the masses so thats why they're the most mainstream.

[gloomy burrow]

Nope. When you look at their actual released benchmarks compared to 4o they usually only beat it in some categories and lose in others. And when they do beat it’s never by a lot. Also only models to have done this are 3.5 and…

[copper storm]

They also have the most features packed together.

[gloomy burrow]

That’s why GPT 4o has been given some updates throughout the months and still like top 1 or 2 model maybe behind Claude 😂

Like after all this time and you still can’t reliably beat it? It’s laughable

But the generation of models that do reliably beat it are here. Like 3.5 opus,etc

But then 4.5 will be out and they’ll be ahead by like another year again

But who knows. Maybe when they are barely getting 3.5 opus out with like 5 messages a week they’ll replicate 4.5 or something

[copper storm]

The competition usually catches up in a few months, not in a year. But yes for actual use cases gpt4o is still pretty high up.

[gloomy burrow]

Like with what? Sora? I don’t seen anyone beating the reasoning for o1. Have you used deepseek? It’s genuinely trash even compared to Claude. Literally several prompts that only o1 gets out of any model in the world. And preview was released how long ago?

[gloomy burrow]

Well been almost 2 years since 4 and it’s still in the top 2 models, or arguably the best one still

[copper storm]

The new sonnet 3.5 is very close to o1-preview, but yeah deepseek isn't as good from what i tested myself, still nice that we have an open source reasoning model.

[gloomy burrow]

Or maybe they’ll replicate 4.5 even tho they are finally getting to beat 4o reliably? Maybe

Probably not

[copper storm]

GPT4 is super outdated by todays standards, there are tons of models better than the standard GPT4, that's a ridiculous thing to say.

[gloomy burrow]

What do you think the o means in 4? It’s being omnimodal. That’s it. It’s been kept up by here and there updates. That’s it. The architecture is 4

[gloomy burrow]

Then the original, yeah. I feel like the company who has been updating their model here and there for the last year and is still ahead is doing pretty good.

Something this simple. Every single other LLM fails miserably

Or exact character counts as well.

[copper storm]

Yes its omnimodal but it's much better than the standard GPT4 in general.

[gloomy burrow]

Yup. But at its core it’s just 4 with regular updates.

Hopefully voice video comes out today

[copper storm]

GPT4 fails the strawberry test btw unless you tell it to do it step by step and count the letters individually, 4o can do it without extra prompting in my tests.

[gloomy burrow]

Yup.They certainly do to this day

[copper storm]

Is this the new claude sonnet 3.5? You should try it directly on claude's interface.

[gloomy burrow]

Yes. I’ve already tried it in the interface. Usually wrong and sometimes right, right around what 4o does

Maybe not 4.5. Whatever this is called

[copper storm]

Okay just tried with haiku it got it completely wrong.

[gloomy burrow]

chatgpt ε

[copper storm]

Kinda unfortunate that they don't give you free access to the new sonnet 3.5 anymore.

Oh damn deepseek added web search.

[copper storm]

Okay deepseek also failed, you're right not a lot of models can do this.

[copper storm]

Do you think we're going to get this today?

[gloomy burrow]

Nope. I’m guessing this will be shown like on the last day (if it’s even 4.5)

Today maybe video + voice like their announced months ago

[copper storm]

Yeah i think so too.

[gloomy burrow]

One year went from random guessing to PHD level science questions.

Hopeful for 2025

[copper storm]

Just tried it with the same prompt in perplexity with grok and sonnet 3.5 and it still failed, okay it really seems like only o1 can do this.

Damn, but yeah like i said the reasoning models haven't caught up.

[fair kiln]

guys gemini 2?

https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/

[pastel current]

What's up with AI and cats? It's getting out of hand

[stoic sleet]

I like to try strarwberry

it usually says 3 on these "fixed" models

4o for instance says 3 when it's 4

[copper storm]

Gemini just launched something like this.

[fair kiln]

I'm trying gemini 2 flash. I asked it something but it doesn't understand it at all

[pastel parcel]

@brittle summit

oh sorry

[twilit turret]

Anyone got turnitin non repository access pls dm

[brittle summit]

I can't wait for 4.5

[royal falcon]

bro 2.0 flash is so good it's funny 😭😭

friendship ended with openai google is my new friend

[gloomy burrow]

Yeah astra I see and Gemini 2,0

[gloomy burrow]

True

[gloomy burrow]

It’ll come back won’t 4.5 trust

Let me try flash actually

2.0

Heheh hehe

hehehehe I hope 2.0 gets this

[drowsy flame]

yo

french here?

[gloomy burrow]

apple intelligence X open ai I see

[brittle summit]

When is today's

@dark void when is 12 days of shipmas

[dark void]

https://www.youtube.com/watch?v=mBhkD0iFf4w
Right now, not restreaming cause this one is lame af

[stoic sleet]

well my big day, small day, big day theory is dead

also the sam being there knocks out that theory

[brittle summit]

That shits gonna put me to sleep

So we gonna get Gemini 2

But not 4.5

[stoic sleet]

4.5 is probably coming in a preview for Pro of some form by day 12

[copper storm]

Yeah that was a letdown.

Sam still seems to be there even second day.

This was worse than yesterday, i guess its completely random if we'll get something big or small.

[copper storm]

How did you know about this beforehand?

[dark void]

someone tweeted it out like 2 minutes before it started lol

[spring thicket]

is there a release today?

[dark void]

better integration with Siri for iPhone users

[copper storm]

and on mac yeah.

[spring thicket]

ty

[stoic sleet]

I should update my mac

[copper storm]

Honestly i feel like they should have cut the 12 days in half and just release the big stuff, these things don't warrant a livestream really.

[stoic sleet]

clearly you don't read the conspriacy theories

[solemn matrix]

damn OAI dropping the ball they waste a day on apple intelligence demo while google drops TPU V6 for enterprise, Gemini 2.0 with flash being the first one out and the deep research agent so their first out of the gate with research automation. Like if OAI doesn't drop a banger on day 6 it's a leopard eating their face moment.

[copper storm]

Hm?

[stoic sleet]

that 12 days coincides will all these bad news articles they knew would come out as a distraction lol

I don't buy it personally

[copper storm]

I mean who knows, i could see it potentially being a factor but not the sole reason.

[spring thicket]

It's absolutely a marketing stunt, regardless of intent

I also have a knack for stating the obvious though 😉

[solemn matrix]

At least the last 4 days we got product releases they could of did another small update like they did for canvas but wasting a day for a demo for a product already out just seems like their padding the days out

[copper storm]

Yeah, this seems like it could have been like 6 days instead. Or 7 to make it a week.

[spring thicket]

YOU try creating a festive, catchy, holiday themed rollout for an event that last "like 6 days instead. Or 7 to make it a week." lol

[copper storm]

Yeah of course.

[stoic sleet]

it's also possible something for today got delayed...

[copper storm]

What? It would take less resources to make it shorter instead of adding filler, all of the features that they're going to release are most certainly already done.

[stoic sleet]

I am pretty sure that happened still on day 2

maybe they could have had the weekends be part of the 12 days?

and just not released anything on weekends

[copper storm]

I heard that what they dropped today was apparently supposed to come out earlier? Im not sure just something i heard.

[solemn matrix]

In the video there was an Easter egg on the calendar it says super secret AGI for 11 am PST

So maybe in about 30 minutes they'll drop something else

[copper storm]

I feel like its unlikely but its possible i guess.

[stoic sleet]

unironically having unlimited 4o is more useful to me than unlimited o1 pro mode

[untold igloo]

only me who have problem with insta right now?

[rose olive]

Day
6: gpt-3 open source
7: gpt-4o video input
8: dalle 4
9: gpt-4o-mini 8B
10: longer advanced voice mode
11: gpt-4.5
12: Christmas decorations & theme

[copper storm]

Eh.

I doubt they're gonna open source anything.

[stoic sleet]

gpt 3 open source would be great but I do doubt that

[strong spear]

Why is sora still not available in EU?

[gloomy burrow]

Ummm super secret AGI?

[gloomy burrow]

You forgot agents

[copper storm]

And we got nothing.

I'm honestly not excited anymore for the rest of the days.

[stoic sleet]

canvas is incredible

[peak imp]

Hold on, when is the reveal today

Or has it already been?

[solemn matrix]

Tired of Sam Hypeman capping all the time

[wind crow]

What was in today shipsmas

[stoic sleet]

Siri integration

[stoic sleet]

Siri integration

[wind crow]

Ah cool

[peak imp]

Oh gotcha

[solemn matrix]

Apprantly 18.2 dropped today that's why OAI showed off apple intelligence still a waste of a day to show off an apple release

[stoic sleet]

especially since it's been in beta for a month and is very stable

[oblique bobcat]

Gemini broke

[ionic apex]

Has ChatGPT 3 been shut down?

[pliant blaze]

gpt3??

[stoic sleet]

you can use it through api still

[pliant blaze]

why would you want to though

[raw grove]

This new option how works?

[pliant blaze]

it uses o1

[raw grove]

Into in another model?

[pliant blaze]

yes the o1 model

[raw grove]

I just saw that it is not available within the GPTs.

[pliant blaze]

its not

o1 is a reasoning model with the idea being test time computer

and its a standalone model aand wont work for custom gpts

[raw grove]

Thanks

[unique lynx]

does anyone know how to make a chatgpt roast bot?

[pliant blaze]

just tell it to roast someone

[unique lynx]

Nah it's like automatic with like deadly roasts

[unique lynx]

I saw it in the server, and it replied to like every message without even being tagged or pinged

[pliant blaze]

try to use an open source model, its cheaper to run locally and also less restricted

[uncut island]

Bots may not necessarily be using OAI models …

[stoic sleet]

I run Qwen 2.5 7B 4bit GPTQ on my phone locally and it's great

[copper storm]

https://youtu.be/bkh8FlsJT1g?si=hRiEx6AF9zWCDPjg

@gloomy burrow

[stoic sleet]

secret AGI meeting was true!

[solemn matrix]

How so

[stoic sleet]

they talked about how video was a key part to AGI and they silently dropped this

bad joke

[solemn matrix]

Yeah I don't think AVM with vision is publically available yet

[modern path]

is running code in gpt new or old?

ive never seen it before and this is really cool

[stoic sleet]

weird limit to hit on ChatGPT Pro while using 4o

[stoic sleet]

it's new in canvas. Though 4o could write python to analyze files and run that seemlessly

[modern path]

i thought pro didnt have limits??

[copper storm]

Do you mean this showcase?

[copper storm]

They never said gpt4 didn't have limits, that's not gpt4o.

Though technically the limits are only nearly unlimited.

[modern path]

This plan includes unlimited access to our smartest model, OpenAI o1, as well as to o1-mini, GPT-4o, and Advanced Voice

[wind crow]

Im happy

-# but im not

[copper storm]

unlimited*

@modern path

[stoic sleet]

I've heard people have 100 request limits for o1 pro a day?

sys admin here so much of my job is automation and I suck at coding

[modern path]

well that is terrible thing to do

[dusty storm]

Hi, what happened to "search" in free version? I can't see it anymore (just dall-e and canvas)

[uncut island]

If you think you need “o1 pro” for automation, you also suck as a sys admin and writing prompts too.

[stoic sleet]

very fair

[uncut island]

Have you tried Claude, just to see if you get better results?

[stoic sleet]

yeah I prefer 4o though and kept hitting the 80 queries in 3 hours limits

developing a solid 4 programs at once for the company that we're too cheap to buy

user to tech ratio is bad and the budget is bad but great job otherwise

[uncut island]

Why don’t you use the playground with API? You won’t hit any limit

[uncut island]

(well just your wallet)

[stoic sleet]

shhh don't expose my excuse for paying for shit I don't need

[uncut island]

They can still fund the API account.

[stoic sleet]

oh I am paying for it I am using my personal account

[uncut island]

Ask in #❔technical-support

[stoic sleet]

I am just irresponsible when it comes to tech

I have a 4090 for a reason

[dusty storm]

Do you know if they removed it forever or if it is momentary?

[uncut island]

Why don’t you stop paying for the subscription and pay for API credits?

[stoic sleet]

I should and will

[uncut island]

Good for you. I don’t know how is this relevant for the ongoing discussion but congrats.

[stoic sleet]

do you know of a good client for using the api locally? Anything with artifacts?

[stoic sleet]

I just am irresponsible and should work on that

[crude hound]

down for anyone else rn?

[low kestrel]

yes

i have a project bruh, worse time for this shit to go down

[dusty storm]

Same...

[uncut island]

https://github.com/open-webui/open-webui

[humble bone]

iss chatpgt down

[pliant blaze]

nope

[humble bone]

girl stop trolling

[pliant blaze]

c:

yeah its down

[still forge]

Is it normal for my chatgtp not to be working rn? Or is it my wifi?

[pliant blaze]

please read above thank you

[pliant blaze]

please read above thank you

[uncut island]

https://status.openai.com/

[onyx oak]

shouldn’t they have redundancy with the sweet Microsoft money

[low kestrel]

how long does it usually take for this kind of stuff

[wispy moss]

gepetto wor

ah beautiful i tought it was local problemo

[wispy moss]

try mistral, free

[onyx oak]

🥱

Imagine paying $200 for this

[lethal python]

gpt stopped working suddenly

[stuck blade]

Love your profile picture

[lethal python]

happened with anyone else ?

[silent cypress]

all over the world, bro

[south isle]

oh, rip.

[trail ice]

bruh my eassy is due in 4 hours im cooked chat

[lethal python]

looks like downtime

[brittle summit]

how long man

[brittle summit]

OK, so everyone’s having this problem

[silent cypress]

I was giving a mock text, gpt betrayed me

[lethal python]

happened 5 mins ago

[brittle summit]

this is very bad

[silent cypress]

yes

[twin jewel]

so its with everyone ?

[lethal python]

ig

[brittle summit]

we cooked

[wispy moss]

try mistral/skype (has copilot ai , either skype or edge or im sure they even have whatsapp

[twin jewel]

LMAO

[low kestrel]

wallah we cooked tg bro

[wispy moss]

the future doctors passing with gepetto

[trail ice]

Im useing bard

[twin jewel]

when will it start up again ?

[lethal python]

okay

[trail ice]

i would not pass half of these dumb classes without chatgpt lol

[lethal python]

cooked

[wispy moss]

in a day with sun

[pliant blaze]

chrismas

[wispy moss]

what you studying friend?

[low kestrel]

whats the best ai website that can watch youtube videos for you and give you awnsers to ur questions

[twin jewel]

istg i need it for revision

[pliant blaze]

gemini can do that

[patent field]

guys is chat down for you guys too

[blazing hedge]

what is wrong with chatgpt ? is it down for you too ?

[low kestrel]

are we dead ass, or not

[pliant blaze]

https://status.openai.com

[wanton raven]

just gonna go ahead and reprogram my communications management system so it doesn't rely on GPT 😫

[low kestrel]

i tried already it didnt work

[pliant blaze]

thats not even possible

[sly merlin]

all these college students using chatgpt rn

[trail ice]

Crim justice with 2 minors cyber and psyc
its dumb woman issues class since i registered late and to be a full time student i had to take this class

[pliant blaze]

i just made that up dont worry itll probably be an hour or so

[pale solar]

Anybody know when it’s gonna be back?

[pliant blaze]

probably less than an hour (probably) (idk tho()

[wispy moss]

lmao can we hoop to dms im really curious about a few things

[orchid pecan]

Im cooked this shit went down right when i started my final

[brave pelican]

yo imma boutta drop out of school now

[pliant blaze]

why are you cheating in school is the real question

[trail ice]

I would love to but i think i gotta write my eassy by hand lol

[wanton raven]

why should this be relevant

[wispy moss]

another day ill save your @

[trail ice]

Simple my friend
advanced study methods

[brave pelican]

im boutta lose everything

[lethal python]

last 2 weeks it wasn't reading files and now this 🥲

[pliant blaze]

HIIII i miss you

[radiant steppe]

GOD PLEASE COME BACK ON I HAVE ASSIGNMENT DUE IN THE NEXT HOUR

[low kestrel]

what else do you think students use it for, they only use it for big exams and projects. Are you good in the head

[brave pelican]

my job, my house, my wife and kids im gonna lose it all

[pliant blaze]

lmao

[orchid pecan]

I don’t need to memorize political history 👍🏻

[pliant blaze]

idk i dont use it to cheat

[lethal python]

final exam in 8h 🙏

[marble badger]

WAIT IS IT DOWN RN???

[trail ice]

lol i never wrote my own eassy without chatgpt in hs

[pliant blaze]

https://status.openai.com/

[hollow linden]

Ask llama for help lawl

[brave pelican]

help help help

[uncut island]

God may not be able to help you but Claude and perplexity.ai can. 😀

[pliant blaze]

hi i miss you night c:

[brave pelican]

God programs my AI

[hollow linden]

Hey lewl. Not sure who you are though, everyone in this space keeps changing names.

[pliant blaze]

lewl do you remember 112?

[low kestrel]

DOES ANYONE HAVE A GOOD AI WEBSITE THAT GIVES YOU AWNSERS TO VIDEOS

[twin jewel]

any good free ai chats other that gpt ?

[hollow linden]

I don't think so rip

[obsidian jasper]

im trying to use chatgpt but its not working. nothign will load, not even my past searches

[pliant blaze]

aw :(

[brave pelican]

GUYS THE FINALS

[marble badger]

i was JUST FINISHING MY ASSIGNMENT MY DEADLINE IN 10 MINS AAAGHHHHHH

[pliant blaze]

https://status.openai.com/

[grave owl]

i need a gpt for an essay does anyone have any advice 🙏

[hollow linden]

112, is it a name? Also go use https://www.meta.ai/

[pliant blaze]

llama, perplexity, gemini, claude

[pliant blaze]

112 was the end of my old username you might know me as the duck

[grave owl]

maria youre the goat

[pale solar]

Not available in my country hhrsvesjt

[brave pelican]

WHAT DO WE DO

[pliant blaze]

https://tenor.com/view/duck-dance-happy-gif-25891437

@hollow linden remember that?

[quaint kraken]

is the website working for you guys?

[trail ice]

we must do what we used to do
write our own dam eassys sadly

[pliant blaze]

https://status.openai.com/

[obsidian jasper]

its down

[marble badger]

id rather sit on a HOT GRILL

[boreal arch]

Is ChatGPT working for anyone else or is it just me?

[obsidian jasper]

gpt is down

very bad we all tryna get our sht doone

[quaint kraken]

Thanks

[boreal arch]

I see.

[obsidian jasper]

youtube

[pliant blaze]

thanks for asking for the 4th time c:

[hollow linden]

I remember duck lewl, I guess you're that person with the duck pfp most of the time?

[brave pelican]

NOTHING IS AS GOOD AS CHATGPT

[boreal arch]

[pliant blaze]

yes i am c:

[trail ice]

bard ai is very good

[hollow linden]

You turned into an anime girl

[livid trail]

bro thinks its 2023

[quaint kraken]

I can't believe I used to write my own essays

[livid trail]

its called gemini lil bro

[pliant blaze]

i did

[trail ice]

I never wrote my own eassys in hs only 1!

[vestal slate]

Guys is chatgpt down

[boreal arch]

Same. I mostly steam rolled by high school English assignments

[quaint kraken]

lmao

[boreal arch]

Apparently so. Not just you.

[stoic sleet]

I loved writing essays

[quaint kraken]

I only used ChatGPT in my last high school year

[pliant blaze]

https://status.openai.com/

[quaint kraken]

I have always been an outstanding student so

[fickle forge]

Just came here to ask this

[pliant blaze]

while youre waiting check out the website im making :D

[hollow linden]

Everyones still asking ChatGPT for homework lewl, some things never change. Also you'll probably get caught.

[dusty storm]

How reliable is chatgpt for making university slides more discursive, improving them and all that? Is it great for then studying on them?
So far I've found it to be good personally

[livid trail]

Its just you, we are all bots trying to psyop you specifically

[stoic sleet]

I just switched to using Kagi Assistant since ChatGPT is down. I probably should use this more

[livid trail]

what da flip

[brave pelican]

CHATGPT SAVE US

[hollow linden]

Just use the API, it rarely has any issues even if the status says it does.

[stoic sleet]

I hope they do

[marble badger]

PLS PLS PLS IM SO DEPENDENT ON THIS APP

[pliant blaze]

most of these people dont know what an api is lmao

[hollow linden]

You just need to spend a couple of your life savings or so.

[boreal arch]

They must be updating something

But IT'S JUST A THEORY!!! - Matt Patt

[dusty storm]

What do you think?

[livid trail]

CHATGPT PLEASE COME BACK

[hollow linden]

rip.

[vestal slate]

How often does chatgpt die like this

[livid trail]

every day for like 10 minutes

[stoic sleet]

only once but it rises on the third day

[ruby orchid]

id say pretty damn often

[stoic sleet]

they need to get ready for the AGI launch tomorrow

[marble badger]

flying cars they said

[cold comet]

Is chat gpt also not working for others?

[boreal arch]

Your most "unproductive" ChatGPT hobby/hobbies you guys use it for. GO!

[brittle summit]

make dua

[dense marsh]

is it me or is chatgpt taking a crap rn?

[marble badger]

HELLO TO YOU TOO??

[cold comet]

Me2

[boreal arch]

Not just you

[radiant steppe]

THESE STANFORD OPENAI SOFTWARE ENGINEERS NEED TO LOCK TF IN OR IMA FAIL MY CLASS

[cold comet]

Does anyone know why gpt doesn’t work rn

[pliant blaze]

https://status.openai.com/

[sly merlin]

idk too many people using it ig

[pliant blaze]

nooo why is @hollow linden muted :(

[marble badger]

im about to recite all my duaas rn 💔

[brave pelican]

gemini is garbage trash

[brittle summit]

;/

[brittle summit]

khair inshallah this what we get for relying on ai

[brave pelican]

gemini literally told me to draw myself a picture of a hottub and sit in it

[lament river]

Thought it was just me. Glad to see folks noting that they’re also having some issues with chatgpt

[tulip spear]

i cant function without gpt

[brave pelican]

MY LIFE IS RUINED

[radiant steppe]

Initally I thought I got IP banned for overusing it

[marble badger]

inshallah but man i knew i shouldve written the damn essay myself but now im too lazy tsk

[boreal arch]

I mostly just write stories with it with my OCs. So...

[lament river]

I thought work caught me 💀

[radiant steppe]

Lmao

[lament river]

😹

[vestal slate]

With chatgpt plus is there a limit to image uploading

[brittle summit]

yes

[surreal sluice]

there is like a time limit

[lament river]

not that I’ve seen

[nocturne sphinx]

Scheiße.
Mein GPT arbeit nicht:(

[surreal sluice]

yea nvm i misread for uploading there is no limit

[radiant steppe]

Out curiousity of how ChatGPT can be used what do you work as?

[stoic sleet]

seeing the meltdown here is funny

[surreal sluice]

there is a limit for generating tho

[marble badger]

PLS HURRY UPP PLEASEEE

[cold comet]

Same

[stoic sleet]

ChatGPT API is working for me

[lament river]

Lawyer

[tired pecan]

guys chat gpt its down?

[cold comet]

Ye

[nocturne sphinx]

Ach so…

[lament river]

Helps a lot with drafting basic stuff

[tired pecan]

i need to do my homework oooff 😭

[stoic sleet]

ChatGPT, or any AI language model, can be a valuable tool in supporting your journey toward a PhD in English, but it cannot directly earn the degree for you. Here's how it can assist you:

• Research Assistance: ChatGPT can help you find resources, summarize articles, and provide insights on various topics related to English literature and language.

• Writing Support: It can assist in drafting, editing, and providing feedback on your essays, papers, and dissertation.

• Study Aid: The AI can help you understand complex theories, literary works, and critical approaches by providing explanations and summaries.

• Language and Grammar: It can help improve your writing by checking grammar, style, and coherence.

However, earning a PhD requires original research, critical thinking, and contributions to the field, which are tasks that require human insight, creativity, and academic rigor. You will need to engage deeply with your subject, conduct original research, and work closely with academic advisors and peers throughout your doctoral studies.

[brave pelican]

what ai did u use

[stoic sleet]

chatgpt

[nocturne sphinx]

Nein!!!
Ich habe sehr wichtig 🚨 Test morgen, und ich muss vorbereiten

[brave pelican]

how r u using

[eager brook]

That's so cool! I want to be a lawyer too and I was going to ask it to make a revision plan for my exams 😭

[stoic sleet]

API

[brave pelican]

how do i use api

[tired pecan]

Do you know any other AI that works?

[nocturne sphinx]

Gemini

[brittle summit]

Hello everyone, I’m having issues using ChatGPT on web

[lament river]

I work for the government and it’s been really helpful with teaching me new, professional, and polite ways to write emails

[nocturne sphinx]

Aber GPT besser

[tired pecan]

ty

[brittle summit]

i'll fix it dw

[hollow linden]

Occasionally Claude.ai has free 3.5 sonnet I think, which is better than ChatGPT in any given day.

[lament river]

Us too 😉

[brittle summit]

How come

[nocturne sphinx]

ich Hilfe gerne ☺️

[brittle summit]

Is there a reason for that?

[brave pelican]

sometimes I feel like we got the primitive version of an incredible invention