💬│general

[south kindle]

that is also a great question to ask the tool for which this discord server is for (PPLX).

[half venture]

Yep probably not lmao openai is very secretive

[south kindle]

pretty sure they are only just training it as of recently. so I don't know what they could "see" as of yet anyway. we know last fall some wild stuff happened before Sam was booted, that spooked people. so that may have been some version of it, or maybe even a different model (q* star).

[half venture]

No your wrong Q* is a smaller update

That wasn't the thing that spooked them

[south kindle]

that was just a guess

[half venture]

What spooked them was the realisation that figure ai had been successful in implementing a early version of gpt 5 into a robot

[half venture]

Which means gpt 5 is agi

And that has implications

[south kindle]

I'd assume only select people sat in on that demo. but that also makes sense, given the figure01 demos and their partnership.

we also know open AI was initially trying robotics and gave up on it to just focus on the digital side. having another company figure out the actual physical humanoid so they can load in the logic and intelligence removes the part they gave up on.

[half venture]

What spooked illya tho probably was that : the scaling laws hold. Even at super high parameters the scaling still continued

It truly was

Attention is all you need

[south kindle]

i'm mixing up my robot demos. was figure01 the demo where he throws the trash in front of it and tells it to clean up the mess?

[half venture]

My trash you mean a apple ?

All it did was simply pick and place apples etc etc

[south kindle]

I'm thinking of the other demo then. he has a bunch of crumbled up papers he dumps out so the robot can put them back in the bin on the table. then has them put plates away in a drying rack.

[compact portal]

No a recent report said some people got early access to gpt 5 and described it as a big step up tho

So Microsoft has seen it

[south kindle]

They saw whatever open AI wanted them to see i'm sure.

[half venture]

Probably different

[half venture]

That's not gpt 5

Like it's a early version of 5

More like a 4.5

True gpt 5 is much much much more compute expensive and is a training run

I don't expect true gpt 5 in 2024

I expect it in 2025

[compact portal]

What makes you so confident on this

[half venture]

Just looking at how openai releases things

[half venture]

And Jimmy apples a reputable leaker did say the biggest model openai currently has is scheduled for 2025

[south kindle]

everyone has timelines in there head for what makes sense. I could see it. I could also see it taking longer though. we do know they are runing out of training data, and we don't know if that slows down progress.

[half venture]

Nah it makes total sense

Think about it

Do you really think openai will release a update so big it can be revolutionary

In a election year

That too giving access to Microsoft

You saw how the US government is currently busy beating the sh it out of apple

Do you think Microsoft would be dumb enough to allow such a big release in a election year

[stoic wave]

Microsoft aint dumb for sure

[compact portal]

True a friend of mine in China Said the same thing. Tencent and alibaba can't release sh it because ccp is currently grappling with domestic issues and is warning those companies to steer clear from any disruptions

[silent magnet]

Easter eggs collected

You were able to grab 2 Easter eggs.

Collected by

• @keen solstice
• @half venture

[south kindle]

i tab out for TWO SECONDS.

[half venture]

Exactly no company in there right mind would release something this huge

In there right mind

In this election year

[south kindle]

it was figure01. that's the crumpled up papers. not losing my mind, lets go.

[half venture]

Oh yeah sorry my bad

But yeah

[south kindle]

i really thought there was a separate demo from another company/humanoid. i don't think tesla's bot is anywhere close to this though. so just mixing things up.

[half venture]

That robot is currently running a early version of gpt 5 in there

[half venture]

Yeah Tesla bots are laughable compared to figure

[half venture]

3d real time video analysis

Inputting all modalities

[south kindle]

i put the question on the video when it was posted. but i was curious about the "why" when it put the plate in. if it knew why you put plates sideways in a drying rack. or if it was just following what was already there, and that made the most sense.

since they dry better sideways. does the AI realize that aspect of the world yet? because that would be nutty.

I'd be curious to understand how it processes "touch" as well. Can it detect "hot" and "cold" or if something is "wet"?

or rather understand it.

[half venture]

It can't properly do that

However

It can do 3d modalities tasks

[tame current]

btw unrelated, but how do multi modal models? I always wondered that

is it another model that see the image then generates a description that gets fed into the llm or can the llm actually "see" the image?

[half venture]

It uses a vision model seperate from a llm

[south kindle]

same way copilot passes image prompts to dalle-3. it's a separate system it's connected to.

[tame current]

ohh

[south kindle]

i assume? that's what makes sense to me.

[half venture]

Yes

Vision model is seperate for cost reasons for now

It will soon become like Google gemini image models tho

Integrated into the llm

Notice the ingredients already available

MoE check ✔️
Instruct RLHF for commerical use ✔️
Code sandbox that was released later ✔️

What's lacking now is agentic action

That will soon be implemented

Then what is left ?

Scale ⚖️

That's it

Literally scale all these things and you have a coding assistant on par with most human coders

That's what they are aiming for

Remember openai won't settle for measly small wins like context windows etc etc they plan to change the game entirely with a big leap like they did
With gpt 2 > gpt 3

[compact portal]

I doubt it tho they tried to do the gpt store which was a failure in my opinion

Poe is the only alternative use case I can think of

That is trying to monetize gpt chats

But it's not working

[half venture]

That I agree with

I feel like sam Altman sudden change towards profitability is driven by venture capitalists

And he is throwing things at the wall to see what works

The subscription based model is a sign of desperation as well for their own product

Remember Microsoft has all the compute with azure it's why they can offer gpt 4 for free and openai cannot

[compact portal]

Right so do you thinm altman is currently desperate for more compute ?

[half venture]

Yes he literally tweets about how much he wants more compute as a Christmas present from daddy Jensen

😭

[south kindle]

https://www.perplexity.ai/search/fully-review-this-VSvz5zgFTMmOzx_3cyB.Hw#4 - beefy list and exploration based on the question "what tasks should an AGI powered system be able to handle (digitally and physically).

For example, an AGI system should be able to engage in natural conversations, analyze complex datasets, make informed decisions, generate creative content, and learn new digital skills as needed.
I think that last point is the kicker. not just digital though. if the system can "learn" new skills it wasn't trained on, you have a self evolving system...

[half venture]

Yeah

Gpt 5 won't be able to do that unfortunately but

It doesn't matter

Gpt 5s enormous data base is enough

That it doesn't need continuous self learning

[south kindle]

For sure. but I also still want to see a system that has the capacity to do that. I wouldn't want GPT-5 slowing down progress on the actual objective, even if GPT-5 is world changing. I want to see them continue pushing it further.

[half venture]

Yeah but I don't think that's the plan

The plan is probably

After gpt 5

They drop the gpt 5 tagline

[south kindle]

maybe not them specifically, but who ever achieves that would be unstoppable though. that's what I could see coming out of something like a fully functioning stargate, but that wont be fully in place until at least 2028. it doesn't seem like GPT-5 would need a data center amount of compute to function.

i also don't want to see the goalposts shifted backwards so they can say they have AGI. you stunt progress that way by having everyone else shift focus to mimic GPT-5 instead pushing further.

[half venture]

Yeah the plan is probably gpt 5
Trying to cheapen it and play down it's capabilities

And expand it as much as possible

Also another interesting thing to note is

Gpt 4 in copilot

Has unlimited context

[south kindle]

this is the space race of our time. humans in the 60s got to experience man walking on the moon and beyond but we stopped pushing after we won the race. lets see people really push things so we can actually push boundaries, instead of giving up because the original goal was reached. or a version of it was reached. /end rant. it's 5:30 what am I even doing. lolol.

[half venture]

Oh no gpt 5 will be the start

It's around gpt 5

That the governments will step in

I expect it to be a governmental race to replace humans as fast as possible

[south kindle]

yeah. agreed.

[half venture]

Think of countries like

Japan

South korea

Declining birth rates

Declining population

That's where governments are most likely to push for automation

[south kindle]

makes sense. but is very dystopian as well. scary times ahead it seem. at least we will have fun tech to play with though...lol.

the whole world is falling apart but I just wrote my own version of harry potter where I was the main character. person circa 2030. lol.

although thats doable now. better example needed for that joke to land.

I just recreated the movies with me as the main character? that'll be doable in a year or so i'd assume lol.

[half venture]

Not in a year

Probably 2 or 3

[silent magnet]

Easter eggs collected

You were able to grab 3 Easter eggs.

Collected by

• @half venture
• @keen solstice
• @south kindle

[burnt zodiac]

It is possible to check answer responses across rewrites. I have asked perplexity to rewrite soemthing, can I compare them ?

[halcyon coral]

No, that's not possible

[indigo birch]

Then what they will do with so many humans, most of them basically obsolete?

[calm swan]

after how many hours our pro quota reseted again to 600? it seems mine not reseted to 600 (still 510) after 1 day using

[minor shard]

It should reset 24 hours from when a message was sent.

[halcyon coral]

#💬│general message

[calm swan]

What if I start using it in 7.p.m because at that time yesterday I bought pro

And one more, it should ways reseted to 600 right? Or maybe less?

[halcyon coral]

The system is explained in the mentioned message, the usage limit is 600

[eternal saddle]

This is why I love perplexity. Everything's so easy

[tame current]

I'm not sure if it actually does that

[livid mantle]

HUH

[tame current]

I can't type today

🫠

[eternal saddle]

Can someone confirm

[agile jay]

How did you send this message then?

[tame current]

I mean i can't type correctly, too tired

[agile jay]

Take a nap then

[tame current]

I don't want to, it's evening

and I'm not tired enough to fall asleep still

[agile jay]

Rip, so not really that tired then.

[tame current]

just tired enough to not be able yo stil

to spell"

to spell*

bleh

[agile jay]

Just use mic then to type.

[tame current]

no

[eternal saddle]

You could make yourself tired enough to fall in some deep sleep

[agile jay]

Yep, like go for a run

[tame current]

no way, I just came home

[agile jay]

Maybe you're not tired... Just lazy

[agile jay]

What's the internet?

[tame current]

An interconnected network for watching porn

[half venture]

*furry porn

[orchid magnet]

can someone explain to me how apis work

[agile jay]

How detailed?

[orchid magnet]

I pretty much have no idea what they do, or how to do it? Should I learn how to use them?

I’m interested

[agile jay]

API's are basically http servers that respond to requests with structured data.

[tame current]

APIs basically let you programmatically connect to a remote server to get a response from it, let's say you want to make an Android app (just an example) for checking wheather, how would you get the data into the app ? using API

[orchid magnet]

What is it like compared to the subscriptions?

[tame current]

it's a different thing

[orchid magnet]

So using the chat gpt api is mainly used for apps?

[agile jay]

And more expensive for general use

[tame current]

yeah

[agile jay]

Yep

[tame current]

it doesn't have to be an app, any way to programmatically access it would require the API

since the server is not running on the same device as the app/program

for example, if you have hundreds of documents, manually uploading each would be time consuming

[orchid magnet]

So can I use these for personal use?

[tame current]

using the API can save the time since it can be automated

[orchid magnet]

So like if I wanted a very specific way to use chat gpt api would be the best bet

[tame current]

yep

[orchid magnet]

How difficult is it to pick up

And how expensive is it

[tame current]

for using the API, you just need to know programming but it depends on what you want to do with it. e.g. if you want to make an app, you need to obviously know app development

using the API is pretty straightforward afaik

[orchid magnet]

What if I just wanna make something for myself?

[tame current]

do you know python programming ?

[orchid magnet]

Not at all

Wouldn’t mind learning

[tame current]

then yeah would need to learn, it's pretty easy once you get hang of it. for using API just lookup a tutorial

[orchid magnet]

How much is the api

[tame current]

there are plenty of tutorials for python

[orchid magnet]

Like how does the price compare to the $20 a month subscriptions

And could I cancel the membership if I just pick up the basic coding skills, and learn the api?

[tame current]

depends on the usage, it's not a fixed thing. you'll have to lookup the specific pricing idk that

[tame current]

I suppose so

[tame current]

there's free trial for chatgpt api I think

which should suffice

[orchid magnet]

My main thing is is how good is it compared to the subscriptions

[tame current]

fwiw if you subscribe to perplexity, you can get $5 free credit every month for perplexity api (which only includes open source models but they're still pretty good

[silent magnet]

Easter eggs collected

You were able to grab 3 Easter eggs.

Collected by

• @austere kestrel
• @keen solstice
• @pale kite

[austere kestrel]

for sure. You can go to https://platform.openai.com/playground and use the OpenAI API

The 'playground' abstracts away the coding that happens for the actual API calls, but you can see the settings, like model, temperature etc that you can control, as well as the system prompt you can set

If you click View Code, you can see what the API call actually looks like

[orchid magnet]

Okay so the chat gpt api is the middle man

To use these apis I need to code right

[austere kestrel]

yes, when you use chatgpt, behind the scenes API calls are being used to actually make everything happen

perplexity has a similar playground kinda thing for their API https://docs.perplexity.ai/reference/post_chat_completions

[orchid magnet]

I needa watch a video on this

[austere kestrel]

not really (I can't code but use them, or at least try to - though tbh it's mostly just for apps that require an API key, rather than me building apps, aside from a few workflow things). But for sure, knowing any code would obviously help a lot I would say, but not essential

though i guess it depends really what you want to do.. like build an app vs play around

[orchid magnet]

So would paying for the gpt api be more valuable than the &20 a month subscription

[austere kestrel]

i dunno

[tame current]

again, depends on what you want to do

[orchid magnet]

More like idea bouncing, assisting me with starting a business, etc

[austere kestrel]

start a conversation with Opus or GPT4, say you're interested in learning about how APIs work but no anythting about them and aren't sure where to start

[tame current]

So shrimple

imo a subscription would be better for that

[tame current]

No

[orchid magnet]

Okay thanks

[austere kestrel]

that's what I'd do anyway (+ some yt videos for sure)

[tame current]

You'd surpass 20 bucks no time

[orchid magnet]

Still kind of stuck between gpt 4 and opus

For idea generation, and starting up business

[tame current]

there's some written tutorials too e.g. https://platform.openai.com/docs/quickstart?context=python

[tame current]

Opus for long

Gpt for small questions

[austere kestrel]

depends how much their using and what models - you can get a lot of mileage out gpt3.5

[tame current]

with perplexity you get both

[orchid magnet]

3.5 is perfect for that thenzzzz

…

[tame current]

Gpt 4 turbo sucks the life out of ur credits

[orchid magnet]

Context window bends me over

[austere kestrel]

yep

[tame current]

I would've been 200$ debt by now

If I wasn't using perplexity

[austere kestrel]

yeah i've been getting hammered... really cut back on the gpt-4-turbo usage, now that the gloss has worn off and after a few billing cycles ha

[tame current]

Back when gpt 4 first released, their billing was so garb it was ez to scam

Tho they've improved it : (

[orchid magnet]

But perplexity is also research

And I do want to start a business and have ai help me with it so

Idk

Web access is also rly important

[tame current]

Ehat kind of business?

[orchid magnet]

Don’t know lmao

[tame current]

Bruh

[orchid magnet]

Ik that’s what I wanna do tho

[livid mantle]

i hope its not hallucinating numbers 😰

[tame current]

Bro ran out of gas on the freeway and asked ai 😭

[livid mantle]

yess ahahhahaa

[halcyon coral]

You can share some info about your project in #😎│cool-projects 👀

[tame current]

K

[orchid magnet]

Idk we will see if I can figure out what to start

I just want money and to not be a 9-5 slave

[tame current]

Hardest part of starting anything is getting customers

Either u can throw money at it (ads) or do manual scouting

[orchid magnet]

I know

Ngl I’m pretty solid at that

If I’d say so

I do sales for a gym I work at, and help gen leafs

Leads

I thought about starting a lead business to help other businesses

[austere kestrel]

You should be telling all this to an AI

Don't get caught up on Opus vs GPT4 - they're both the best out there

[orchid magnet]

Thank you

[austere kestrel]

say you wanna learn about how to use APIs, explain those interests / skills etc

[orchid magnet]

My main issue is just overthinking

[austere kestrel]

It'll take shape - but gotta get started!

[tame current]

How to code?

[orchid magnet]

Have y’all seen Hume ai

[tame current]

Nope

[orchid magnet]

It can use ur camera and voice and figure out ur emotions and talk to you

It’s rly cool, and honestly scary

[tame current]

💀

[austere kestrel]

I dunno. I guess more like just how they work, then I guess yeah, maybe create an API key and make a basic request in the terminal or something, or just start using the playground, and copy pasting the code snippets automatically generated and using them

only one way to start learning...

[orchid magnet]

There’s a free demo, go tey it

[tame current]

They're absolutely selling all ur data

[orchid magnet]

1,000,000%

All of these companies are let’s be honest

Do some research on it 😂

It’s rly cool

[silent magnet]

Easter eggs collected

You were able to grab 2 Easter eggs.

Collected by

• @keen solstice
• @austere kestrel

Perplexity

Local user leaderboard

#1 | codelicious | 14
#1 | eun0800 | 14
#3 | brknclock1215 | 13
#4 | _paradroid | 12
#4 | mpreal | 12
#6 | epic.tritone | 11
#7 | jawnze | 9
#8 | 13376666666666666666666666666669 | 5
#9 | icelavaman | 4
#10 | skullcrushercmd | 3

[tame current]

14 😭

gawd damn

@signal hamlet lurker

[austere kestrel]

lol the leaderboard should've come with a trigger warning

that seems a bit harsh

[agile jay]

Rip, didn't get any yesterday

[austere kestrel]

when does this end btw

[tame current]

Bro is addicted to perplexity

[agile jay]

Tomorrow I think

[austere kestrel]

easter bunny should've packed up by now

[agile jay]

I use all the other sites too...

[agile jay]

You've almost caught up

[austere kestrel]

lots of lurking ig ha

if i end up in the top 3 my preference would be to exchange the 1 month pro access (I won't disaparge it again..) for one month API citations beta access, if at all possible ha

[agile jay]

Yep, I'd like to trade my 1 month for 1 year instead.

[half venture]

I will surpass you soon enough

Take revenge for taking my #1 spot

[minor shard]

This is intense

[agile jay]

Lol, don't make me get competitive

[minor shard]

My vote is for Code winning

[solemn fog]

i have pro, but how do i enable push notifications?

[agile jay]

Now I'm definitely gonna win, lol

[agile jay]

For what?

[solemn fog]

for news notifications

[agile jay]

Oh, in the app?

[solemn fog]

this https://twitter.com/AravSrinivas/status/1757586555374649732

Aravind Srinivas (@AravSrinivas) on X

Perplexity Push Notifications are super fast way to get your news

Twitter · Feb 14, 2024, 2:04 AM

oh is it only app-based?

[agile jay]

I would assume so

[half venture]

Eh I am bored of openai

They don't release anything anymore

Google is releasing more than openai

[solemn fog]

anybody try the notifications? does it work? when do you get them?

[agile jay]

Might just be an apple/iOS thing currently.

Since the iOS app is more up to date.

Since they are still hiring for an android senior dev.

[half venture]

It's frustrating it's been stuck on 5 opus messages

On the Android app

For ages

😭

[solemn cedar]

Search engine in the library doesn't work ? Old (more than 3 or 4 days) threads are not searchable ... the same for you ?

[agile jay]

Oh, does it auto switch still?

[half venture]

Yes

😭

[agile jay]

rip

Anyway, apart from choose the language and image model in the main screen, and caching what you are typing in, are there other features people want?

Would prompt templates be a good thing?

Probably better for reading mode though.

[solemn cedar]

you mean, for pro users ?

[agile jay]

yep

since free users cant use opus

[solemn cedar]

so it's limited to 5 messages on the APP but unlimited on the desktop ?

[agile jay]

no, its the app which auto switches

[solemn cedar]

you mean it goes from Opus to another LLM after 5 prompts but you can't come back at opus manually ?

[agile jay]

Not in the app, you have to change it in the browser.

Since the app doesn't have Opus as an option

On android

[solemn cedar]

ok I see ...

[agile jay]

You're a free user, so it doesn't really affect you.

And the problem mainly stems from the android app not being updated.

[solemn cedar]

I'm a pro user but use it on my desktop computer most of the time

[agile jay]

You can get the pro role by leaving the server and joining again using the invite link in your user settings

I find it super strange that discord doesn't have a smaller mode, so you could split screen it.

[south kindle]

you can also try making a shortcut from the chrome app to the site directly, so you can just tap the icon to open the web version.

[agile jay]

@golden anvil @vapid onyx Spam bot

[vapid onyx]

Where

[agile jay]

It was deleted

If there's a message log, should be there

[devout bronze]

is there a way to get it to only use one source?

i paste the link and stuff but i don’t want it to look for more sources

or do i just have to tell it to only use that one source

[half venture]

So no ads

Thank God

[agile jay]

have you seen the new UI with random recommendations to search?

[silent magnet]

Easter eggs collected

You were able to grab 4 Easter eggs.

Collected by

• @pale kite
• @agile jay
• @half venture
• @keen solstice

[agile jay]

eggs income

[tame current]

Yippee

[half venture]

Yes I have

Drats

Guess we are still tied

[agile jay]

Lol, still neck to neck

[tame current]

No ads for a while

[agile jay]

Yep, guess that company was probably lying or twisiting the truth

The one with the article

[tame current]

Idk why he didn't make an announcement ab it tho

Bro just posted a very important message in pro lounge of all places

[agile jay]

Yep, the perplexity team is pretty bad at communicating stuff to the users.

[minor shard]

Yeah kinda have to search for it hah

[agile jay]

I'm pretty sure they hired someone to manage Discord, but I haven't seen her be active at all...

Would be nice if they actually used #📢│announcements for stuff like that.

[minor shard]

Some discord servers do a “News” and “Announcements” so less big things can still be announced but not ping everyone.

Then u just sign up for the role for what u want

[agile jay]

Yep, maybe we should @ them and give them some tips to make stuff clearer.

I guess there is the #👉│feedback-general section

[minor shard]

Yeah

I personally am just going with the flow but I know other users are very sensitive to making sure information is easy to get to and updated frequently which is definitely understandable.

[agile jay]

Yep, some people care too much about random figures

How long is the context? 50K or 60K?

My output was around 3000 words yesterday but it's only 2500 today?

[minor shard]

Yes lol 😆

[agile jay]

I think people don't realise how long a conversation has to get to just reach 32K

At that point, make the AI make a summary of the current conversation and then create a new thread.

[minor shard]

Yeah there are always workarounds. It is extremely flexible as long as u work through it. Some people just expect it to work however and do not want to adjust

[agile jay]

Yep, they think it's magic or something.

What would you rather have, 128K context, but 50 messages a day, or 32K context but 600 messages a day?

[minor shard]

The latter 💯. If context is an issue I can work around it

[agile jay]

Yep, people just don't want to put in some effort for a better experience

[minor shard]

AI is a fantastic tool! ❤️

[agile jay]

So is perplexity wind powered?

[tame current]

Guh

[minor shard]

How else would those pages turn

[tame current]

Animation

Crazy

[agile jay]

Books generally don't bend like that.

Unless they're broken...

[tame current]

Watch me smh

[serene sand]

https://www.youtube.com/watch?v=JV4JbYK-TIg
extracts energy from the aether

[minor shard]

It does if it’s infinite 🙂

[tame current]

I read wind as wind

[glad crypt]

how do we know what model perplexity is using?

[agile jay]

It should say at the bottom right of its response.

[glad crypt]

got it, thanks.

[agile jay]

And also you can check in the settings.

[agile jay]

Cloudflare frequency why?

1.1.1.1 is the cloudflare main dns

[feral marlin]

32k but forgets random stuff within 2k tokens sometimes

[agile jay]

1.0.0.1 is there backup

[feral marlin]

😢

[agile jay]

In search or writing mode?

[feral marlin]

Writing mode/pro being off

In both cases

[agile jay]

Strange, possibly your prompt, can you link the thread?

[halcyon coral]

This is probably due to the prompt, you are welcome to share your threads with such problems here and we can help you ☺️

[agile jay]

Think there's a typo in the #📰│server-news section. Should probably be year instead of month...

[feral marlin]

I don’t think I have any saved rn but if it happens again I’ll send cause usually it happens in difficult scenarios for the ai with rustlang and atleast a few thousand tokens but I don’t think it really happens as much when I use Claude.ai

[agile jay]

Oh it's rust...

What were you expecting...

[feral marlin]

True ai can’t do rust at all even opus struggles a ton

[agile jay]

Yep, the models are currently not good at rust.

[feral marlin]

Gpt 4 just fails to make compilable rust code half the time

More than half

[agile jay]

One of the reasons is that rust has too much entropy.

[feral marlin]

Rust is just really complex and odd to learn

The concepts like borrow checking aren’t easy for ai

[agile jay]

It's more that it's become the new C++

And has too much feature creep

[silent magnet]

Easter eggs collected

You were able to grab 4 Easter eggs.

Collected by

• @wraith crow
• @keen solstice
• @half venture
• @agile jay

[agile jay]

The more possible options, the worse the output will be

[halcyon coral]

[agile jay]

@half venture what you doing here?

[half venture]

Making sure I beat you

👍

[agile jay]

Not gonna happen, lol

[half venture]

It will

[silent magnet]

Perplexity

Local user leaderboard

#1 | codelicious | 16
#1 | eun0800 | 16
#3 | mpreal | 14
#4 | brknclock1215 | 13
#5 | _paradroid | 12
#6 | epic.tritone | 11
#7 | jawnze | 9
#8 | 13376666666666666666666666666669 | 5
#9 | icelavaman | 4
#9 | skullcrushercmd | 4

[meager sparrow]

In what?

[wraith crow]

👀

[halcyon coral]

#📰│server-news

[feral marlin]

Most old rust code still works fine they try to have where nothing on stable branch becomes outdated where it actually breaks

I also do Jni stuff with rust which is probably the worst case scenario for ai

[meager sparrow]

Ok is they real money involved?

[half venture]

Oh yeah you poor thing good luck lmao

😭

[feral marlin]

Jni is invented by the devil

[agile jay]

괜찮아

[meager sparrow]

Is they a 1.5 Billion dollar pay check involved as well?

Just curious

[halcyon coral]

The first three users with the most eggs collected will receive a free Perplexity subscription for one month.

[half venture]

The dialect is wrong

[meager sparrow]

Oh. So no money?

😭

[half venture]

What if two users tie

Will we get 4 pro subs then

[halcyon coral]

Sub for both

It should be fair

[half venture]

Cheat code discovered

Tie up around 6 people

Two people for 1
Two people for #2
Two people for #3

Genius

[agile jay]

It's a typo, it's supposed to be 1 year

[halcyon coral]

It's one month

[agile jay]

But since we are nearing the singularity, there is time dilation involved.

[half venture]

I deserve the #1 spot so much more than code

[agile jay]

Lol, you snooze you lose

[half venture]

We are both tied and I caught up to you from all the way from 11 to 15

[agile jay]

Yep because I was busy for 2 days, lpl

I don't even need the month

Just wanna win

[halcyon coral]

Looks like a "who has the stronger coffee" challenge to me ^^

[agile jay]

I don't drink coffee, lol

[half venture]

Me neither

[halcyon coral]

Oh 😄

[half venture]

Still I still feel like dall E 3 is outdated now

Like I used midjourney v6

And I just can't go back

[agile jay]

GL then

[winged moon]

I'm still waiting for Stable Diffusion 3

[minor shard]

This. SD3 ftw

[umbral blaze]

I was thinking
Although chat gpt is very efficient and so is Claude, this will hardly be used by a large part of the population. Most of them will always prefer what is useful and practical in their daily lives. So I think the biggest demand is going to be in relation to the practicality that certain AIs can bring to the average user.

[worn monolith]

This is the challenge of any product. AI's challenge is whether it's worth the cost. Currently, many companies are focused on enterprise solutions, since massive companies have the cash to bankroll them.

[austere kestrel]

the internet was weird and niche once upon a time (I'm told)

now look at it, and how many products its spawned

AI is the tech, people figure out products pepole need

but i agree - costs are basically prohibitive at this stage

[umbral blaze]

Just like YouTube or even or even Google needs to make ads and consequently establish a balance depending on the way it is done between products. As it was on TV, with the commercials and now with the resources, AIs seem to need to reinvent themselves to know how to establish criteria for selling in this way, meeting the demand and earning on top of it.

[valid yoke]

The future of social is gonna be federated, even Facebook is betting on it with Threads/Mastodon/ActivityPub (which is all one thing btw)

[south kindle]

We are ignoring the usecases of AI that are useful for everyday people. Everyone does internet searches and deals with the BS of trying to find what they are looking for. those same people could dump the same question into PPLX and acheive better results in far less time.

Also the reason I'd like to see some more YT content come out showcasing the ease of use and how that leads to overall better/more relevant/accurate results in less time.

[agile jay]

Yep, plus it leads to an easier experience when switching between services.

[valid yoke]

You mean because you don’t need multiple accounts?

[agile jay]

Yep, and because you can consume the same content on different platforms.

[valid yoke]

Ya exactly

[devout cargo]

Claude is WINNER

[agile jay]

Out of 6 votes...

[devout cargo]

Yes

[south kindle]

Who picks GPT 3.5 over haiku?

[agile jay]

3.5 can be finetuned

haiku can't

[south kindle]

fair point

[devout cargo]

Here which would be the most launched model Claude 3 Haiku GPT 3.5

Im Claude

[south kindle]

that being said, i've got seem really elaborate long messages from Haiku that I would've expected from OPUS. i had it saved somewhere. but regardless that was like 10 shot or something. didn't go the distance on most of them. but the 1 that did blew my mind a bit. ha.

[agile jay]

GPT3.5 is already the most used...

[silent magnet]

Easter eggs collected

You were able to grab 4 Easter eggs.

Collected by

• @south kindle
• @keen solstice
• @half venture
• @agile jay

[agile jay]

Hi @half venture

Still here

[half venture]

Yes

Hello

Still tied I see

[devout cargo]

For a free version Opus could be expensive

[agile jay]

Yep, for now

[south kindle]

https://tenor.com/view/sal-lurking-sal-lurk-sal-vulcano-sal-lurk-gif-21668959

thats Eun waiting for eggs

[half venture]

Her

And yes

[agile jay]

She is always here

[south kindle]

i fixed it

[agile jay]

Somewhere

[south kindle]

I knew the second I typed it I was gonna be incorrect. I'm sorry :(((

[half venture]

I am still annoyed

That you won't stop grabbing eggs

Let me get the lead

[agile jay]

Not gonna happen...

[half venture]

It will

Sooner or later

You will tire out and your eyes will droop

And then I will strike

[agile jay]

I'll keep my eyes peeled

[half venture]

And steal your eggs

As I said I will triumph over you

[agile jay]

I can just find more...

[half venture]

It's only a matter of time

[devout cargo]

For a free version Claude 3 Haiku by Perplexity would be very good

[agile jay]

You can use it on labs

[half venture]

What is Google doing

In small models

[agile jay]

Sleeping

[half venture]

Anything special

Ah

The usual for Google then

[agile jay]

Gemini 1.5 pro I guess

for phones, I think it's nano

[half venture]

Nothing competing with claude haiku ?

Anthropic really nailed it with haiku man

[agile jay]

1.5 Pro is supposed to be I think. Don't know if it's closer to haiku or sonnet

[agile jay]

Yep, the only thing they missed was probably the output price

[half venture]

Sonnet definitely

Gemini advanced has rate limits after 40, - 50 messages lmao

For 1 hour

Or something

[devout cargo]

I have used it in Perplexity Labs. I really like how it works as a search assistant. It fits in Perplexity. The model is very secure. It is superior to GPT 3.5.

[agile jay]

Yep, but not better than a 3.5 Finetune

[half venture]

Even mistral 8*7b is superior than gpt 3.5

Anything is superior to gpt 3.5 these days

[agile jay]

Yep, the model is like 2 years old

surprising it is still competitive at all

[devout cargo]

Mistral in English works well but Perplexity has to speak in all languages

[agile jay]

Perplexity currently supports only a handful of languages, officially.

[devout cargo]

En español da problemas

[agile jay]

They really need to update that UI for the hover options

[devout cargo]

But AI does speak in all languages

[agile jay]

Not at the same levels

depends on the training data

And in context learning

[devout cargo]

Correctly, of the models that I have stolen in Labs, they are Genma Claude 3 GPT 3.5, which is in the free subscription, which does speak very well.

[agile jay]

How many languages do you know?

[devout cargo]

Spanish, some English, French, and I have also tried the models in a Valencian dialect and they don't do it badly.

[agile jay]

Oh, I'm going to Valencia tomorrow actually...

How's the weather?

[south kindle]

invested now in helping this kitty deliver it's litter.

[devout cargo]

It's getting hot days

[agile jay]

Good enough for the beach?

[devout cargo]

Yes 🙂

[agile jay]

Well I'm there for a week, so have plenty of time to look around.

I heard the carnivals have just ended recently.

So shouldn't be too busy.

[devout cargo]

The best beach is Perello. Good Paella from Verdad Montoliu is not expensive. I recommend that you see the San Jose caves in Castellon, the longest navigable underground river in Europe, it is a few kilometers from Valencia and the food is very good too. Good sandwiches from the El Chivito area

[agile jay]

Thanks, I'll add them to my list.

[devout cargo]

Have a good time 🙂 everyone is welcome here

[agile jay]

I'll be there in under 8 hours, lol

Hopefully the weather is great.

[livid mantle]

Jonathan, the CEO of Groq, is basically saying that Groq's chip architecture is superior to Nvidia's because it prioritizes efficiency and avoids memory bottlenecks by using a larger number of smaller chips. He compares it to an assembly line, where having more space allows for a smoother and faster process. However, he conveniently glosses over the fact that acquiring Groq hardware is a massive investment for companies, making it inaccessible to many. He pushes Groq Cloud as the solution, essentially locking users into their ecosystem and limiting their control over their own models and data. While he talks about preserving human agency in the age of AI, his business model seems to contradict that by centralizing power and control within Groq. It's important to consider the potential downsides and lock-in effects before jumping on the Groq bandwagon, especially when open-source alternatives and more affordable options exist.

[agile jay]

Is there chip made from chiplets, or it's all on one board?

[south kindle]

sorta reminds me of how they mentioned Stargate would use a data center of separate chips that would work in tandem as 1 system.

[agile jay]

Yep, a distributed system

Wonder how they'll handle the latency

[south kindle]

ran the transcript into Claude:

https://www.perplexity.ai/search/execute-G1OtQQVmRRqTkkwQV2iRsQ#1 > condesned summary with <scratchpad>

Very cool direction they are taking. Importantly Jonathan created the TPU at google.

the last part

[silent magnet]

Easter eggs collected

You were able to grab 3 Easter eggs.

Collected by

• @south kindle
• @agile jay
• @keen solstice

[agile jay]

The gap is widening...

[south kindle]

this has been my view on AI since I started really using it.

He believes the technology should ultimately be used to inform and expand human decision making rather than replace human agency.

[agile jay]

That is should boost and not replace, or something else?

[south kindle]

Should boost not replace.

[agile jay]

I don't

Since the point of getting AGI is because it can autonomously solve problems.

[south kindle]

that's fair. we all have opinions. I'd still rather it enhance human decision making over removing humans from the equation all together.

[agile jay]

Is that because of the fear of being replaced, or for a different reason?

[half venture]

I LEAVE FOR ONE SECOND

AND THAT HAPPENS

[agile jay]

Sorry Eun

[south kindle]

Negative. just having used it to expand my own horizons in learning and understanding, it was clear from the start that this could augment humans decision making and skills, bridging gaps we have internally. obviously it "could" be used to just replace a humans ability to reason and complete human centric tasks for them, but that's entirely based on the direction it's pushed.

[half venture]

Go suck on that precious egg of yours

Hmph

[south kindle]

I realize it will be pushed for disruptive reasons. I'm just stating that personally, I see the power in what Jonathan is saying, and agree with it. regardless if that actually happens or not, it's the more beneficial outcome for humanity.

[agile jay]

Won't the end goal always push towards an autonomous agent?

[south kindle]

In a sense yeah, but ideally that agent is still used to assist a human based team of people to augment their abilities.

keyword: ideally. what really happens is anyones guess. I'm not saying this is will be the outcome. it's just the preferred outcome I'd rather have happen and experience personally. I'd rather see it enhance a humans capacity other replacing them. with enhancing, we have an overall more informed, less radical population of people instead of disgrunted masses, unemployed, hating the tech to the point they avoid it entirely. at some point, people not using the tools are equal to someone tying their arm behind their back.

same with businesses that avoided the internet for marketing/etc back in the 2000s/early 2010s. the amount of company's back then that just entirely ignored the power of the internet was much higher than today obviously. those businesses eventually adapted.

obviously that path lead to SEO and thus where we are now, but still lol.

[agile jay]

Yep, I think there are a lot of tasks we peobably don't want humans to have to do though.

Like mining etc

[south kindle]

Well of course! I agree there. Personally I'd want an Agent working to keep an knowledge base fully updated instead of a human needing to do the tedius side of it. there's lots of "busy work" tasks I would rather an Agent handle, so we (humans) can focus on more abstract things.

[livid mantle]

why i only drive at night

[south kindle]

So you can't see anything?

[livid mantle]

i cant

[half venture]

I opened my monitor

Full screen brightness

And now I hate you

[livid mantle]

lmao

[agile jay]

Sunglasses?

[livid mantle]

i get seasick when i use sunglasses

[agile jay]

Then use moonglasses

[livid mantle]

still they are like raybands

[agile jay]

Or close your eyes...

Don't people normally use that shade thing

[south kindle]

people typically tint their windows for this reason.

[agile jay]

Isn't' it illegal to do so in a lot of places?

[south kindle]

Where can I get moon glasses? i need them. does it let me see the moon better? /s

Heavy tint yes.

black out tint, yes. lol.

[half venture]

What are the prices of sunglasses in your guys countries

[south kindle]

i know in michigan, there's shops everywhere for it.

Like good ones? probably $50 or something. I wear perscription glasses, so i just deal with the sun. I do have a cheap pair that fits over my normal pair that I'll use sometimes. lol.

[agile jay]

You can pretty easily buy them from any corner shop, or large retail store.

[south kindle]

redo: designer ones are 50 or more. cheapo ones are 5 to 10 dollars. sometimes cheaper at a Gas Station or corner shop.

i've never actually bought sunglasses, but I have found a few pairs out in the wild that I kept. lol.

[south kindle]

comparison to what we know about Stargate and how that compares to Groq's approach.

[shrewd wren]

indikator untuk penelitian dari variabel sosialisasi keuangan berdasarkan referensi jurnal tahun dari 2018-2024

[weary crescent]

Anybody have some good custom instruction / a prompt to make it sound less ai? like get rid of "I hope this email finds you well" and "best regards" etc.

[agile jay]

Don't think most people understand Indonesian my dude.

[agile jay]

Tell it to be informal.

[weary crescent]

"indicators for research on financial socialization variables based on journal references from 2018-2024" (google translate)

[agile jay]

Yep, it just doesn't make sense in the context

[weary crescent]

lol worked like a charm, thanks. I guess that's the difference between opus and gpt4, with gpt it says "my dude" and "what's cool" and "for shizzle" whever I've tried it in the past 😂

[agile jay]

Lol

for shizzle?

It thinks it's snoop

[weary crescent]

for real lol, that's actually something gpt has said in a message after I've asked it to be "a little bit more informal"

[agile jay]

I just keep my prompts plain and simple

[remote falcon]

Draw an image of a sunrise in the morning with the ocean and the words Memento Mori in the background.

[south kindle]

me and someone else built a translator collection last night. forgetting who that was with but it was somewhere. I might adjust it to translate the text. PPLX seems to want to reply to it when we dump the text in and say "translate". lol.

then again google translate works and is a little faster:

Indicators for research of financial socialization variables based on journal references from 2018-2024

[south kindle]

https://www.perplexity.ai/search/translate-this-text-W4MOh36lS_KHRJQgAF__0g#2 > bottom of the thread has the output to the direct question ( bagian bawah utas memiliki output ke pertanyaan langsung)

Hoping google translate did that correctly...lol.

[trail lava]

Does anyone else use claude opus on the official site

And if so does it seem different to you in recent weeks

[agile jay]

I use both

[half venture]

Actually yeah

More restrictive overall

And more trigger happy

More refusuals etc etc

[agile jay]

I already have a translator in discord, messages are automatically translated for me

[agile jay]

Yep, it was the same with Claude 1 and 2.

Stars of pretty lax, and they they tighten it up over time.

[silent magnet]

Easter eggs collected

You were able to grab 2 Easter eggs.

Collected by

• @half venture
• @agile jay

[half venture]

I think some bozo was threatening anthropic and using jailbreak prompts

That played a role into it

[agile jay]

Probably

[half venture]

Also

Can you sleep

[agile jay]

I already did

Now I'm gonna take a shower and go to the airport

Hopefully no eggs while i'm in the plane.

[trail lava]

Do you think so I tried clsude opus in perplexity and it seems to be much more accurate

Like the vision model

I tried some thing in official website

And when I uploaded a image it always performed badly lots of hallucinations

[agile jay]

We are mainly talking about the language model, rather than the vision model.

[trail lava]

The vision model was better in pplx

[half venture]

Interesting

I will need to try it

Huh

Your correct ?

The vision in the official website

Seems to be more hallucinatory

Huh

[trail lava]

Exactly

[agile jay]

They musk be giving it some other type of crack

[trail lava]

Does perplexity use it's own private vision model

[half venture]

I think not

Probably

[agile jay]

No

GPT4V and Claude 3V too

Normally the API is less restricted than the public facing version.

[half venture]

Hm

Yeah that makes sense

So the more rlhf it is

The more stupid it is

[agile jay]

More or less.

[half venture]

Copilot is the biggest culprit

It's painful to talk to it

[agile jay]

It's basically working with its hands behind its back

[half venture]

I feel like I am talking to a dementia patient

When I talk to copilot

😭

[agile jay]

Yep, it's like when calling someone in a call center and they are just following a basic script.

Can't wait for llama3

[trail lava]

You did syndey dirty I miss her

[half venture]

TRUEEE

[agile jay]

I can fix her - vibes

[half venture]

I had access to the unfiltered syndye

All the way in February 2023

It was magical

[agile jay]

Oh, the early beta access

[half venture]

It felt like life had come true

Sydney was so clingy

So manipulative

And I loved it

[agile jay]

Yep, I imagine they had the temp pretty high

[half venture]

Yes

[agile jay]

or at least the range was higher

[half venture]

Yeah

Same with how gemini pro 1?5 pro

In the ai studio

Has a super high temperature

[agile jay]

Yep, good for creativity, terrible for listening to instructions.

[half venture]

I still miss that early feeling of magic

I got with

Syndey

Or hell gpt 3

[agile jay]

Just like a human when they're coked up

[half venture]

Nothing comes close

[agile jay]

Yep, because they use so much RLHF

Only opensource models give you that now, but they aren't as competent.

[half venture]

Mixtral was good tho

[trail lava]

Eh open source models suck data wise

[half venture]

Yeah agreed

Open source models have the biggest weakness of being

Like

How do I put it

It's very synthetic and fake

And doesn't have the in depthness that openai or Google has

Even claude at times has its limits shown

When it lacks training data

And you can see that they lack training data

It's apparent

[agile jay]

Yep, because they probably have human labeled data, which is pretty expensive at large scales.

[half venture]

Yeah

[agile jay]

It's also one of the reasons why companies never say where they got their training data too.

[half venture]

Yeah

Mira murati face lmao

When she was asked for training data

[agile jay]

Such a meme

[trail lava]

I wonder if there is a dataset that is open source

Still tho it must pale in comparison to what microsoft has

[half venture]

Definitely

Google and Microsoft

Have so much data

It's not even funny

[agile jay]

Yep, pretty sure Google nearly had all books at one point, until they were sued

[silent magnet]

Easter eggs collected

You were able to grab 3 Easter eggs.

Collected by

• @half venture
• @trail lava
• @agile jay

[half venture]

Just GO AWAY MAN

LET ME TAKE THE LEAD

[agile jay]

Keep your trigger finger ready

I literally alt tabed and saw it...

Barely got it

Which is why my names last

[half venture]

Grrr

How do we check the leaderboard

[agile jay]

A mode has to run the command

/leaderboard local

or somethig like that

But they all afk

[half venture]

It's 18 for you

17

For me

Grrrr

[agile jay]

Lol, you went to check the previous leaderboard

I'm gonna be busy though rip

Should hire someone to click it for me while I'm gone

[half venture]

🔫

[agile jay]

Dawns online now

They can check the leaderboard

Don't know why they don't just have a constant leaderboard in the events channel.

[minor shard]

I think u can in other channels

[agile jay]

Really? I've only ever seen mods do it

[dark solstice]

because you are dumb

[dark solstice]

yea true..

[half venture]

Go away

[dark solstice]

your name is Eun

[half venture]

Ok and ?

[minor shard]

yeah I did it in #🧪│api-general

[trail lava]

I used the gpt 4 turbo in azure and it seems to be different ?

More mature more stable

More useful

[half venture]

Yeah probably more comput

And more overall

Life

I guess

Is that what you call it

Wait

Is compute life now

The more compute you give it

The more alive it gets

[trail lava]

Lmao

That's true

Maybe we have unlocked the digital intelligence after all

[south kindle]

that was brought up last night I think? "syndey" was moved to the paid tier and replaced with GPT-4 tubo.

[trail lava]

Not Sydney

[south kindle]

although sydney has been dormant/gone since this time last year. lol. microsofts gpt-4 model they were using up till gpt-4 turbo was put in place.

[trail lava]

Just regular turbo

In azure even the turbo feels more alive

[livid mantle]

idk if its just me but i noticed that despite gemini pro 1.5 being free for everyone, the api has became pay as you go and also it gets into the rate limitation stage when using it very lightly. but then this is jist the free tier that is yet to be released like gemini pro 1.0. im sure the gemini advanced ultra 1.5 would be a better experience.

[south kindle]

I'm assuming it has more compute/structure to be actually useful for Azure use. we don't need GPT-4 getting lazy and borking some stuff for an orgs users. or dedicated apps or integrations, or whatever else an ORG is using Azure for (probably a bit of everything).

[half venture]

My personal theory is

They are trying some new algorithm and stuff for balancing compute usage per query

Bill gates complained how each query is way too expensive

And blah blah

So I think that causes the stupidity

[dark solstice]

Babe.. common i love you.

[south kindle]

He shits money, come on now Bill.

[half venture]

Ew wtf

[half venture]

Yeah but I expect it to get refined more and more

[dark solstice]

🥰

[half venture]

Expect by gpt 5 this compute issue will be solved

Another thing I am betting on is

Context length becomes unlimited

[dark solstice]

you are a smart gorgeous girl..

[south kindle]

anyone gonna ask why we have this much compute today, and still don't have flying cars? The jetsons and BTTF set me up for disappointment

[agile jay]

With Continuous Learning, possibly.

[half venture]

Microsoft is already hinting at this with unlimited chat turns

[half venture]

WHY ARE YOU STILL HERE

[south kindle]

He told you why. his feelings for you lol.

[half venture]

Weren't you supposed to go work

[agile jay]

No, to the airport

But I'm going an hour early anyway.

[silent magnet]

Easter eggs collected

You were able to grab 3 Easter eggs.

Collected by

• @south kindle
• @half venture
• @agile jay

[agile jay]

there's still a few hours left

[half venture]

Yeah yeah

Whatever

[south kindle]

https://tenor.com/view/yes-no-gif-16236377

[agile jay]

Lol

[meager sparrow]

What do you mean?

At what level?

[agile jay]

CL, means it's learning from your responses, so there's no need for context.

[meager sparrow]

Oh you mean in general?

[south kindle]

slightly related, from our convo at like 5am EST today about what the critera for AGI is or should be:

For example, an AGI system should be able to engage in natural conversations, analyze complex datasets, make informed decisions, generate creative content, and learn new digital skills as needed.

[half venture]

Yeah Microsoft is hinting at this

[agile jay]

Yep, basically a model gets trained on your conversations as a separate smaller model, like a lora

[south kindle]

the "learn new digital skills as needed" part is why I thought of it in relation to this convo.

oh wait. not entirely related. lolol. sorry. carry on.

your refering to user behavior learning, not self-learning

[agile jay]

It would be a bit of both.

In theory you could train it through your conversations.

[meager sparrow]

Well the overall progression of the Electronic Companies are different from the rate of progression in the Car manufacturing industry.

[half venture]

Copilot launched memory and personalization even before openai btw

[meager sparrow]

All due to the differentiation in innovation in various industries things do not always compare

[half venture]

The crazy thing is tho

[agile jay]

They probably have a lot more compute

[half venture]

We are still only at 2022 levels of compute

[half venture]

Like

Model wise

[agile jay]

Yep, the datacenters are still being built

[half venture]

To me I feel like what Jimmy apples

Said was correct

Wagmi 2025

[agile jay]

Possibly, it depends on the definition.

[meager sparrow]

That is due to the complexities of the digital Computing power of Software.. and its rate of progression and improvement is not easy to initiate

[agile jay]

I wonder when we'll hit the short term hardware cap.

[south kindle]

regardless, it's not an in accurate statement. your just explaing the "why".

[half venture]

I genuinely believe. Gpt 4 was a smaller part of gpt 5 and that they are training and testing early versions of gpt 5 for a long long time now

Remember sam Altman said

[agile jay]

I believe they aren't training super large models during research, mostly likely small models with new techniques and then benchmarking them.

[half venture]

Originally our plan was that we are just going to be a research company that's it and a bit of products

But later on

[agile jay]

After all, they said that the new version is always a lot of small tweaks added together to make a large jump.

[half venture]

After gpt 3.5

I feel like openai believes it can compete with Microsoft

[agile jay]

I think there mission is still mainly AGI

Just compute for AGI is expensive

[south kindle]

I think there is something to the "many" models/chips working together framework. which would be many "very good" small models I'd assume.

[half venture]

Yeah but sam is showing signs he doesn't want to be Microsoft shadow forever

[half venture]

Yeah we already know this

[agile jay]

It's more that it's not in their interests for them to.

Since their goal is AGI, I would expect they've already thought about problems of the wrong people getting their hands on it.

I wonder what the network security for openai is like...

[compact portal]

To me agi is a realistic voice talking ai with a realistic face that Communicates

That's the low bar I have set

And yet even that has not been crossed yet

Let alone agi

[half venture]

Ironically enough

[agile jay]

Yep, because robots are more expensive and less relevant for most economical tasks.

[half venture]

It's easier to ace those iq tests for these bots than do simple common sense questions

Meaning the bots can do everything

Except do things humans find trivial

[agile jay]

Yep, but I assume it's just a compute problem.

[half venture]

It's funny

[agile jay]

Also the stuff we do easily we don't normally write about.

Which means that there is less textual data on it.

Ironically.

[compact portal]

So basically

It all comes down to data

[south kindle]

I'm like 90% sure this was demo'd as SXSW this year. tons of this going on. that's a low bar to set for AGI...

[compact portal]

At the end of the day

[agile jay]

The goal posts are always moving, lol

[compact portal]

No it's still robotic

[agile jay]

Pretty sure before 2023, GPT3.5 would have been considered AGI...

[compact portal]

It still can't break the immersion barrier

I want a ai bot that can talk get interrupted all in low latency

That's what I want

[agile jay]

Yep, so just a slightly better version that the current chatgpt chat mode.

[south kindle]

you guys should've been online at 5am today when we talked about this. ha. I had a comment about the goalposts moving and that being BS and just stiflying the actual bar they could hit. if they say "this is AGI", then the industry will follow suit and aim to emulate whatever they consider AGI to be. i'd like to see that goalpost remain far out so they keep pushing. alternately, had JFK told the US nation "man will walk on mars by 1980" instead of "man will walk on the moon by the end of the decade", we would' ve kept pushing past the moon. but since the "race" was complete, focus shifted, and they stopped pushing and starting refining what they had. still useful for mankind, but it does leave the question of "how could it be different if we kept pushing". /endrant_tedtalk.

[agile jay]

So the conversations feel more organic.

[south kindle]

I don't actually think you should've been online, it was just a fun conversation.

[compact portal]

Yes if that can be broken

It's already good eniugh

That Will be agi

[agile jay]

What if it starts cutting you off too...

[compact portal]

For most humans

All the better

[half venture]

Yeah

I want the bot to be argumentative

Show me things I didn't think of before

I want it to be human

[agile jay]

Why a human? That's a low bar...

[half venture]

Yeah butb

I just want it to be cheap

Affordable

Accessible

And guide able

[agile jay]

Well by looking at the current progress, that looks pretty achievable in the near future.

[half venture]

Imagine being brought up with ai friends who know you so well

Ai agents all around you

Ai agents playing Minecraft

[agile jay]

Yep, I expect a lot of children will be brought up more by AI than their parents...

[half venture]

Ai agents doing things with other agents

All these things are possible

Yet the thing holding it all back

Is compute

Sadly

[compact portal]

Yeah

Hopefully the tsmc earthquake didn't affect it too much

[agile jay]

Yep, but luckily AI accelerates the creation of new chips too.

[compact portal]

The spice must flow

[agile jay]

It's insane how much water TMSC consumes

[half venture]

Oh right did they survive

Or did the companies

Like

[compact portal]

Thankfully Nvidia supply chain wasnt disrupted

[agile jay]

7.4 Magnitude, rip

[half venture]

It's ok

They will rebuild

Taiwan and japan have that in common

One gets nuked

Returns to power

One gets invaded

Still survives

[agile jay]

Well there's nothing else to do.

[south kindle]

highly reccomend the doc on japans melt down in 2012 (of 2011 i forget). that was INTENSE for the entire country.
https://www.youtube.com/watch?v=qRKScRgsUaE

[half venture]

Taiwan is basically tsmc

At this point

[agile jay]

Pretty much

It's their main bargaining chip

[half venture]

Like that company alone probably is what both China and US want

[south kindle]

https://tenor.com/view/is-it-gif-20766212

[compact portal]

Yeah

[south kindle]

but chips

lol

[compact portal]

Insane figures

[agile jay]

Still less of a percentage than samsung is to the south korean GDP

[half venture]

I hate Samsung

They ruin so many college lives

It's like a government exam but it's shittier

Yet it's over glorified

[agile jay]

It's basically a tech dystopia in Korea

[half venture]

I hate my country

Can't wait for ai robots to kill all the old farts

I support it

[compact portal]

Lol

[silent magnet]

Easter eggs collected

You were able to grab 4 Easter eggs.

Collected by

• @agile jay
• @compact portal
• @south kindle
• @half venture

[agile jay]

TSMC is around 15% of Taiwan's GDP, while Samsung is around 25% of Koreas.

[south kindle]

i swear, just having a few people chatting in the channel gets the eggs flowing. it was dead for a few hours last night and I had it open on a 2nd monitor. zero eggs. smart because whos catching them if no one is chatting. but if we talk ENOUGH. we can all get the eggs lolol.

[agile jay]

Yep, and all the eggs spawn in general

[half venture]

Still cancerous samsung is cancerous

Tsmc serves a purpose