#general
1 messages · Page 213 of 1
@echo aurora nbp not working with images uploaded :P
He cannot share the prompts. He cannot run any prompts. He doesn’t know when it’s coming but more to come soon.
whats this
do u have eyes?
Its hopemium
I hate getting this kind of news lol
It doesn’t give no details about anything tangible
When where who how ? Its a tease 😭
ikr
other examples
Both examples have ip or copyright content
yea but how is it compared to other models?
doesnt gpt image cost more while being worse + worse quality?
i dont understand these people are always complaining
you want something good but again you want it cheap
like its impossible
they will always balance the performance / cost
exactly
All I’m saying is
Who tf is going to be spending almost 1 dollar on four (4k) images and be generating on scale if ur a normie or regular user
business users
Ya
the casual user wont yea
like 2.4$
people that dont know how to spend money for lower taxes?
yea if its 0.024$ / 4k
where its 0.024
I mean I c go through 100 n hour lol
what u need 100 for?
Maybe even half depending how fast gens come
I’m just using it as a basic measurement
But ur right also
100 is a lot if ur just messing around
But it’s still fun to gen img or videos
fun is not free
I hate to use this word but it Addictive
Once again when ur right ur right
Still something in me screams that’s its bs… we are talking about images
I mean, how many do you think would have been created in the last two years from AI
Almost anything and everything you could think of plus some
Half the stuff goes into trash
The other half we never see and the 1/3 we do see is on such a massive scale on every social media you could think of
tbh i may want to use more than 100 per hour
like 4k
for this side project of mine but its not worth it
ops
nano banana flash could be special giving the price
That was pretty creative man
wrong channel
If they charging .24 cents for 4k on googles scale
Its cost them like 1 cent for a batch of 1000
could be just a leaker that wants to prevents their identity
Don’t they got synth Id lol?
yes. I do want good and cheap. 🙂
We need Veo 4 just like NB pro but for videos
And I want to cut of the profits
😂
Dude, any of these tech companies from Google to Facebook to even this discord to Reddit? What do they all rely on?
They rely on people on people like me on people like you people we know that do all the hard work and build these platforms up for what they are
There would be no Conversations in discord there would be no posts on Reddit. There would be no articles for for Google to search
The profit from our data are collective data, but none of the money ever funnels back to us. Well, some of us were fortunate, and we get blessed by the algorithm.
The 2k version or regular? both?
2k
I've been hearing a lot of reports this morning about image upload/that model being problematic.
Dude, getting blessed by the algorithm is like winning the lottery
oh
Looking into. Are you getting an error when trying to upload or errors when the output happens?
something went wrong message
now i cant rry anymore cuz i hit the rate limit
Gotcha. I'm not sure that's related to image upload as it could be the regular issues that we're having with the model.
U need to make a spare account to test against
Image Upload issues typically happen whent he image is initially uploaded.
Lm account
:)
You guys are talking about?
Thats a easy fix
The people that get these errors I wonder if they ever log off
You need to log off once in a while and re-login
Was that the rate limit thing for repeated prompt?
No it was the cloud flare
As soon as I logged out and tested, it not logged in that verify your human box popped up
There’s a lag delay for some reason and it doesn’t pop up and that’s what’s blocking it
Why you delete your data on browser or re-login you should either get a verification or it should work
Oh that means it should verify you but somehow failed so it ends up in error loop
I’m not sure exactly how it works. I think the verification thing is supposed to pop up, but it doesn’t when he needs to.
See I’m not logged in and I tested it and it worked first shot
And you saw the previous ones it was all failure
So that tells me it’s not the arena it’s on my end
bro no
My simple take is just switch internet environment or whatever it called, let cloudfare/server recognizes you as two users. Since this problem seems won’t be fixed soon(or at all)
The chat is different
Because the rate limit on certain models
You’re gonna get bigger more often in direct at least with images
nah
I hardly ever use that reason
I hardly ever get errors to be honest
Most of the time it’s only when my prompt are rejected because of the filter
I can prove it
Direct is to unsteady cuz of rate limits
If I need to test direct I make 1-2 calls only
You can easily convert 1;1 to 16:9
yea, better use battle to pull your desired outcomes and retry. using direct and getting "something wrong" is just a waste of time and hourly rate count ofc.
Well, I don’t know what you’re gonna get out of direct
But you can’t get out of battle unless you give a variety
Being stuck on one model,is foolish imo
Cuz u miss other models pros
And then you can see the weakness of each model specifically you could tell each strength
For image editing and recreation, nb pro is enough and much ahead. Seedream/reve/flux, sorry to say hardly any practical scenario to use. (Very rarely reve gives some surprising stylized image. SD is good at text to image and graphic design stuff.)
a lot of amazon models were thanos snapped from LMArena
before the phantoms followed it was the belugas
will the leaderboards be updated on december 10th
It will become a norm for AI sites to have ads
Following OpenAI, google is adding them to gemini too.
No.
Look at this low quality Sora video how crisp it is
I think I’m getting closer
can i do any personalize setting with my modes
4k image raises quality sora video
Update: Google has proven this false
google never lies
Google will not join OpenAI in ads
in gemini
https://019aff9a-6bfc-7e87-a98b-56f20526f1d4.arena.site
seahawk made this escape room game with jumpscares
should probably say CW: Clowns
tbh it got me
it also took images from the internet
the jumpscares it picked are clowns out of all things
No img just promot
Why would they ?
They have the largest network on the Internet for ad distribution
All Gemini has to do is get you the information to the place where the ads already placed
google wins if they make oai lose more clients
and thats exactly whats happening
oai average users are decreasing
and thats a big W for google
There’s two problems with being on top
And we are slowly starting to see this with Google ever since nano
Youtube ads make enough already??
well they said they reduced flickering and distortion in a update
I think it’s gonna happen with AI. It’s gonna be covert advertisement
And I think how it’s gonna work is they’re gonna have “ reputable” sources when citting
And then they could just typing off the data that they get from Gemini
And target you with new ads outside of the rest of Google’s platforms
Today I tried to sign up for a AI service for music to extract voices and try to clone
And I go on YouTube and I get a commercial for 11 labs and I never have before
And all I did was use my Gmail account to sign up for a service outside of Gemini
The thing that google spies on us is already a confirmed statement
Look at something and you see ads of it.
Yeah
There’s no need to put ads on here because without ads you’ll get people using it more
And people are more likely, gonna be more intimate with it
With what this year I mean
I mean, If they were in an OpenAI situation, they would need ads.
Especially people are uploading images of themselves and stuff to edit all that stuff
Open AI will have ads
And so will Google
But not gonna be traditional
Google has Ads on youtube already
In 2023, Google generated approximately $237.86 billion from advertising, which accounted for over 77% of its total revenue. This figure reflects Google's dominance in the digital advertising market, with expectations for continued growth in the coming years.
Yeah.
90% of that is youtube
I think we’re failing to see here
If we look for example at television and advertisement and TV, right?
Yeah?
So, in that context, it is the advertisers pain to distribute their ad through the networks
In this instance, it is gonna be Google being able to connect the consumer to the product
One sec I’m in the middle of something I’ll be back in. I’ll articulate what I mean a little better
Google most likely wont
i think google end game is to sell gemini to enterprises and convert google.com ai overview and ai model to be powerful enough that individual users dont need gemini.google.com
This cant be real
OpenAI will die to AI tax
Is that Maxwell
While OpenAIs in a code red
Yeah im using it since long time and now just to remember this funny cat because it passed away
So Google is not gonna have a choice
Google will vibe with it
Yeah Maxwell will be remembered as a great cat
It is worse for OpenAI in a code red situation
Open AI giving us courtesy of letting us know ahead of time
There maybe a double tax
If the government decide to pass some kind of stupid AI environmental bill
If you are a developer or a business using tools like OpenAI (ChatGPT), Anthropic, or Google Gemini, you may have noticed your bills increasing. This is not a new special AI tax, but rather enforcement of existing software taxes.
• What is happening: States and countries are increasingly classifying AI API calls as "Digital Services" or "SaaS" (Software as a Service).
• The Impact: Companies like OpenAI are legally required to collect sales tax (e.g., ~8–10% in the US, ~20% VAT in Europe) on your monthly API bill based on your billing address.
OpenAI might just be fried..
It would have nothing to do with open ai would have to do with the users
I know how they tax gas for environmental reasons or whatever
Because the amount of resources needed to power AI
Like they’re planning on taxing electric vehicles if they’re already not
Do chatgpt pro models even have API or no?
We are several hours apart from new the most powerful LLM.
Since I didn't see anything related to chatgpt pro models API
No
But if you sign up for ChatGPT, plus I think it comes out to be a total of like $21.50 or something like that
The tax was deducted already
Man I hope chatgpt 5.2 will be good
But we got two really strange political situations going on in America in terms of AI
President Donald Trump threatened to sign an artificial intelligence executive order that would establish federal laws superseding states' rights.
If 5.2 is robin im sorry to break it to you but it aint good
I think first place in text arena
So the first thing that’s happening is that the Trump administration has been lobby extremely hard by the tech industry mainly AI so he wants to pass a bill to make AI regulations of federal thing and not a state thing
Though I won't be surprised if Gemini 3 pro still will be smarter
maybe on style control…
But , like I told you guys the other day about the AI bill that’s proposed in Florida
GPT-5.2 is 100% not Robin
So we have people from within the same party on two different sides. One wants to sign an executive order while the states wanna enforce their own AI regulation.
What is Robin?
I don't think so.
I'm new at this stuff
Codenamed OpenAi model
Very interesting to see what’s gonna happen
Oh alright
And who’s behind the lead for the executive?
Okay ig
Mr. Sam Altman
Claude is always the best, anyway.
Any thoughts about Gemini deep thinking?
What’s really fascinating for everybody that’s here and who’s been from AI from the beginning what we’re seeing is a monopolization of the AI industry before our hands from where it began
👍
This is how industries traditionally in the United States have monopolized through government
Kk I wonder if it will be added to LMarena tho
Robins codearena style is like gpt 5.1 so no it is openai
No. It's dumb, GPT-5.2 will be the beast!
The doubts i have are not countable
Imagine chatgpt 6.7 I know dumb joke but still
Chatgpt be making these 5.2 versions or whatever
I wouldn’t expect much from 5.2
I could be wrong though
But it just doesn’t make sense to strategically
I do.
Same honestly
I’ve already considered 5 a complete flop there’s nothing that’s gonna save this series
I just need like a model that knows Luau best
The word best is not a quantitative or measurable metric
Fair point
Even though I know what you mean
I guess I'll stick to the Claude opus or Gemini 3 pro for now
I just need a model that dedicated to coding
Qualitative metrics delve into the realm of subjective experiences and opinions. They go beyond numbers and statistics,
It’s very hard to answer ur question
Generated by Nano Banana Pro Thinking 2K
Wow awesome
That looks great. I love the colors.
GPT 5.2 will be a beast, high hopes for OpenAI.
Same I doubt it will be a beast. I hope I’m mistaken. I hope I’m wrong which I wanna be in this case.
Let's just hope 5.2 is not a disappoinment
Honestly, the problem with ChatGPT is its neutered
What’s the word everybody uses that starts with a L?
On top of my head, I can’t think of it.
Lol?
During training, we observed that applying censorship significantly degrades the output quality of language models.
Or I dunno
🍌
Really tricky statement
What does it have to do with banana emoji
Because how could censorship exist if the model is supposed to be accurate and honest?
It was generated by Nano Banana Pro 2K
Hmmm okay
I wonder if nano banana pro 4k exist
I think I've seen it on yupp maybe
The more post training guardrails you pile on, the more you destroy the model’s raw intelligence and usefulness.
Yeah ofc si they find clever ways to make the model withhold information
Like redirecting or acting stupid
But still be engaging
It doesn't exist
I understand that it’s necessary so I’m not totally against it
Agreed
My issues with gpt personal lol
Okay
These models consequence are in a sense you could argue are trained to lie
4o was amazing, at that time.
4o the perfect model
IMO
Great model compact strong, lose, flexible, adaptable
Fun easy to use great output length cheap
It’s like the Windows XP
OpenAI spent years chasing the consumer market while ignoring developers; now they’re realizing coding is where the actual hungry audience lives.
Yeah, look at the books and saw where all the money is coming from actually going to pay for it
Developers & we are seeing a shift also towards contact creation with media video image audio etc
Have you seen tensor 1.5? our latest ?
I have not had the honor
It would be amazing to hear some feedback from you, it's been a pleasure talking to you .
It's a model focused on coding, internally matches Opus 4.5
I think that’s what that gentleman was looking for the model of dedicated to coding
@compact flame u hear that?
Yes
It's also 94x cheaper 😃
Gpt-5.2 scores #1 on Vision Arena. Did it just get multi model? 👀
so yes, enjoy .
I’ll test this
Is 5.2 in arena.?
This gotta be ai generated..
I see Gemini tag
I think
Hmm
Is 5.2 in the arena.,?
Yea!
Yeah I got tricked
You didnt know?
It's not there
it is.
But seriously nano banana generates such realistic copies ngl
Yup
Literally had the gemini logo 😭
Yeah I noticed but decided to ignore it
My X account got restricted for ‘in authentic behaviour’…
Been trying to contact X support but they are so hard to reach
X moderation sucks
GPT 5.2 (Not real. btw.)
i can actually see the synthid watermark in this image
it's like green and pink tiling, you can see it in the leaderboard's background
if the image was generated with api, it will still have the synthid watermark, just not the gemini star
I know u can check the c2p meta
if the synthid watermark isn't in the image, it won't detect it
it is supposed to be invisible, but in certain situations where the background is a solid color (especially if it is black), you can see the watermark
There might be a different way too
it also puts a slight noise in other areas like text, but this isn't as visible
yeah 🍌
Printer paper and printing ink
Printer tracking dots, also known as printer steganography, DocuColor tracking dots, yellow dots, secret dots, or a machine identification code (MIC), is a digital watermark which many color laser printers and photocopiers produce on every printed page that identifies the specific device that was used to print the document. Developed by Xerox an...
Crazy tech
That they can this digitally also
I think this was designed from printing money lol
Yeah
To see who created some fake images of trump and other people
I knew they was using that technique i just didn't have proof
Yeah technique is outdated
Wym outdated
They got more crazy stuff
It is foolproof
With exact time stamps and id of the image
Oh
Wym more crazt stuff
But if i run it through other ai model i think it'd disappear
Check that out
I’m reading it right now. I got some crazy ass stuff on there.
Hi to you too Craig
Hope you have a great week ahead
movementlabs = scamlabs
Oh lord
Where's your Minecraft clone video?
Why don't you send that here?
Or your model didn't get to that point yet?
lol
Don't need to send a video, you can try it yourself it's free also you don't need an account 🙂
I remember you sent that Minecraft video that was litteral Minecraft gameplay with shaders
(85% of ai industry)
We tried yesterday it wasn't even close to what your video shown
And that video was 3 months old btw!
Lies....
yeah I noticed that their tensor 1.5 can make an excellent if slightly buggy one-shot minecraft clone (might have been trained on that though) but the world knowledge is bad and can't hold up to Gemini 3 pro or any of the other flagships, probably because of the mixed weights from qwen lowering the otherwise great world knowledge from glm 4.5
Google uses tpu
Last time i checked we are the the cheapest 👀
yeah the mcerebras processing unit
Tensor Processing Unit (TPU) is an AI accelerator application-specific integrated circuit (ASIC) developed by Google for neural network machine learning, using Google's own TensorFlow software. Google began using TPUs internally in 2015, and in 2018 made them available for third-party use, both as part of its cloud infrastructure and by offering...
and vc money is abundant
no they have scammer money, gained through scammer means (probably)
that's the only mpu that exists
Is that with tarrifs ? 🙂
if you want to convince me of an mpu, then show me a die shot and a picture of the chip on a board. I bet you can't.
@deep adderThousands of users love what we build. Great models, fair prices, and a genuinely generous free tier. Calling that a scam doesn’t add up.
yeah mpu in that instance has been used for over 50 years at this point
I don’t know I don’t know anything about this stuff so I can’t really say anything
But it definitely is a real thing
it's as old as the microprocessor itself
I didn’t know that
@deep adder At this point the obsession is giving ‘needs professional help’ energy, just saying.
I'd call this deceiving customers
Wow
they could've just used nano banana pro to generate an image of an mpu, but nano banana pro is smarter than tensor 1.5 in the fact that it has much better world knowledge and not just whatever came with glm 4.5 (which honestly isn't that bad! i give z.ai credit for making a good open model with good world knowledge at 1/3 of kimi's size)
Yes, just like how Lamborghini still calls the Aventador their flagship even after the Revuelto dropped. One is the reigning king, the other is the new king. Congrats on discovering how product lineups work in 2025.
It’s in Chinese so you’re gonna have to translate it
This dude is deep
Into ai 😂😂😂
He is Ai
I love how we call everything ai now
It's amazing 🤩
Ai image ai video ai text ai voice
Quite a bit when I was walking home yesterday about AI and AI consciousness
Fascinating to think about
The hate doesn't slow us down; it fuels us to go harder.
I can even make a Ai Joloooo
You guys just have became very Questionable somehow
@queen veldt dedicated to you.
Is this your own trained model?
Or just fine-tuned?
wrapper
Hmmmm
Are you certain?
🤣 Love it!
But even if it is, what’s the big deal?
It’s not like it’s only circus in town
If that was the case
Strong, wrapping what exactly?
the wrapper
Hmmm
hmm
If you can match our outputs to any AI out there i will give you $500 in api credits.
All I gotta say about any of this
Whatever the case may be here real fake scam legit
This is the opportunities that the AI industry itself brings
Gehlo, you seem very educated. No need to prove yourself to anyone.
So whatever it may be, this is the result of the industry, and how business is being conducted
Arguing with a bunch of kids who use free LLMs is just a waste of time.
Nawh they chill bro
i need mawr
$500 is too little
Are you living under the rock dude?
You never heard of opportunities to commit crimes from AI?
he got your ip
how can you fall for that
now u are hacked
i dont want to be hacked 😖
This is nice, what was the prompt?
im scared 😖
Minecraft clone could be just a hard training for that specific game
"Make me an SVG of this girl riding a bicycle."
getting gemini 3 to do the same thing rn
Of course it outperforms other models since it's trained to make Minecraft game and web os...
Great, would love to see it
Expected the urge you know that means that means business is blooming
I mean with every single update with every single advancement guess what it’s gonna go up with it
That’s right crime
Is this Gemini 3?
yes
Crime was here before ai but fair to blame on ai
These companies have a Responsibility and obligation absolutely
as you can see, it read the image more clearly and used it to make the svg
the old pelican prompt, but with a twist
Nice prompt work kiri
Yeah go train your model on specific prompts
Wow both are matching so perfectly
Try a prompt yourself, something out the blue. Don't be shy it's free.
Im just kidding obviously
5 is just not enough for us gang
💀
Back to 50 tomorrow by the way.
Alright then i will start trying from tomorrow
Did you guys really train this model?
And we are giving away $10,000 in API credits
10gs
💀 how to join
"Just provide us your credit card info"
I’ll get movmentlabs tattoo for 10gs
If you truly were the par with opus 4.5 you'd be more popular than now
Meaning you're better than grok and chatgpt
We are launching our Ads Q1 2026
But i don't even see you on lmarena 🤣🤣🤣
Is that the authenticator lm arena ?
Just warming up at the moment.
Perhaps we are more skeptical then we should be
Of legitimacy?
i would honestly enter just so i can spam the api with random world knowledge questions, maybe run simpleqa on it
This brings up a huge conflict of interest with the arena
@swift oyster u didn't answer
Yes we did
but seeing this it may not end well, there are 4 errors in this paragraph
yes every ai company is in it
What about Benchmarks tho?
If legitimacy is gained because your model is in lm arena
Although i knew they are deceiving
That’s a huge conflict of interest
no
I meann gpt is there claude is there
Every most powerful model is there
Yes, because people wanna get on the list because the perception is that if the model is on the list it’s authentic
Even amazon and meta
And incentives for people to want to be on the list
gemini 3 pro corrected all the mistakes without web search
Not only it's authentic but to show which one is most capable
There's real scores involved
Well, that’s what I mean. I obviously know it’s not authentic but what I’m saying is the perception is if your model is on there, then your model is considered in the public space legit model.
So something is going on shady since this models isn't coming to lmarena, and it's for a reason
LmArena wan't money for listing, we are going to list in Q1 2026
We don't want to simply just buy our way in.
We actually had a zoom call with LMarena.
This one just came out bro
i honestly can't wait until the model goes up on huggingface just so i can confirm its origins
or maybe it will end up being like that 6t parameter model that was a frankenstein's monster of llms
Yes. Sorry, I’m reading.
I’ve actually been thinking about this for a little bit about lm arena and it’s place
How it’s being used by users and the benchmark themselves
It takes two to tango in any form of partnership just in general
It’s a give-and-take
Yeah although mods are saying lmarena is paying for api
They very well could be, and that would probably be the best way to do it actually
How they're doing it by themselves etc etc
Because if they paid it by themselves, they’re not getting influenced
Lmarena generated 8m+ images with nano banana gemini 2.5
Because on the other hand for LM arena to exist it also needs traffic
Well, that’s what I’m saying, bro
They need traffic. They need users and they need to keep them obviously engaged in using the service or the benchmarks or whatever you wanna call it.
Nano banana costed them +700.000k or something back than
Now it's even more for the nb pro
momentum gets this whole thing wrong (95% of the info in this picture is incorrect, kimi k2 can get everything right and perfectly reproduce the song lyrics, which are completely incorrect in this picture), tensor just refuses to generate song lyrics because it's a thinking-only model
Well, this raises even more questions and answers
If they’re offering the service for free power generating money?
Selling data?
yeah, that spelled the end of the good times i had with lmarena, it became worse after that
I guess perhaps is only made for coding and logic
We’re growing good, so 2-3 accusations of being a scam have zero impact.
700k how much are you selling to make that? lol
You know what I don’t even wanna know the answer honestly, I don’t
ofc it is, they need to make the model good on the things influencers care about, and having good world knowledge is not on that list
The less I know the better I sleep
Bro it's JUST for the nano banana BACK THAN
Imagine claude
So not a Generalist Model i see
But how are they making money to pay for it?
People using gpt 5 high
What about Tensor 1.5? What is your email DM me i will upgrade you to a pro account.
Why would they just make a arena and just lose out that much money on it?
For Free?
so it went down in the last 30 days?
💀
yeah tensor's a bit better in world knowledge due to reasoning
refuses to generate song lyrics, but i can probably change that with clever prompting
Does tensor have web search?
probably same base as momentum though
Yes
Yes, it can be toggled on.
Subscription fees from users. Partnership sponsorship. Or data
How else would they generate money?
Maybe is not for Creative uses then?
Use dekstop mode
No
no song lyrics is actually one of my world knowledge tests, not even glm 4.5 does this bad
U won't see the toggle otherwise for now
Well it should be turned on by default then?
seeing how well it memorizes them is one of my tests
Web search defeats the purpose of @whole sundial's tests.
i can tell you that the paypal customer service ai does better in terms of song lyrics than momentum
only on desktop at the moment
it does, but i care about how well the base model is without tools
U need web search
Why it's not on by default if it's not visible on phone?
What's the point of hiding it 😂
I guess it would come down to weather lyrics were in training data or not
of course they are, nothing is off limits for ai companies
Ahhh so they do use lyrics in training data
We are not hiding it, just working on the mobile UI 🙂
Should be out in 2-3 days.
Tbh guys i had enough of being skeptical
Web search on mobile, then IOS and Android apps are coming within the next 2 weeks.
If they are Great they will Succeed Eventually
yeah web search isn't that great, i can't share the prompt but i can tell you that grok 4 fast and glm do better than tensor
The old defense was stronger
Gemini ui looks terrible
Gemini ui sux
After this much time my doubts are dropping now for momentum labs
Chatgpt ui is good
After you mentioned it that one day I started paying attention to,it u We’re right.
But i hate chatgpt
Same
yeah glm 4.6 just gave me a great response with the same prompt, web search needs to be improved
It's all work in progress, community feedback is super important to us.
The only thing giving any ai access to real time data
U should start discord server
And i guess reward users for feedback
👋
google search api is the best imo
Yeah basically no brainer
It's good to see you have a change of heart 🙂
Lmao
👋
🤣
I sent them a ai generated one
Bad idea
Well i just realised and observed how the opinion and data around can make one biased
Y
Humans are eager to take a side quickly
Fantastic 🙂 was it this by any chance?
Well, I thought about it at first. I was like maybe it’s not a good idea.
i think they're british
Netherlands and UK.
Not accurate statement
and dutch as well i guess
A lot of them are in Singapore
In the end i suggest be very careful with your marketing word of mouth is very important and present some sort of evidence the world will believe in so they trust countless times the bias is not by actual performance but rather shaped by the opinions of others @swift oyster
But are Chinese nationals atheist
We have some great papers and videos coming of our data centers etc, all due in time
Since the Republic of China does not allow religion
Q1 2026 is when we will really push.
just tell the truth, 80% of ai startups use chinese base models anyways so it wouldn't do much harm to admit the truth
Yeah that's what you guys need the more the evidence the more the authority
it's good in some things and not so good in others, need to try more prompts but i think i've used up all my daily tensor uses
we did use shards from open source Chinese models, no denying that.
Guys we got insider trading info here
yeah not a good accusation to make
U deserve a pro sub
💀
We don't want any investments, don't let @deep adder see this please.
Looks real vro
And I was lazy. I just wanted to see if it would work.
Nawh
I looked a little bit more into it after the fact, and there’s just a lot that goes into these things
Also for marketing make sure you study human psychology as well @swift oyster
who has already invested in the company
I just did that take a selfie
The one I sent them it said that I was Julius Caesar born in the year 25 BC
also do the limits still reset at 0 UTC? you used to show them in the old ui but you don't show them anymore unless i'm missing something
I just wanted to see if the work
yes
That's right.
Yeah
Damn
Yup
Just send me your account email and i will upgrade you for free.
U no y?
Because the fear of losing out on the market and them developing their own is so massive that it’s better to sell to them
China has a huge population
Can i also get a upgrade
the question is: Will China allow companies to buy the H200s?
It’s middle class is the size of the American population
💀
China has access to H200s via cloud companies and mega underground contracts.
Yeah
Since is H200s now maybe perhaps
now legally
Huawei is doing a good job too
we already know that, but china has been against companies buying h20s
I wouldn’t be surprised if we come to find out that many the Chinese companies the big ones at least have been running them.
they want their companies to use homegrown chips now
Well, the h20s had a back door
Hi everyone, i am here to create videos to sell the 3d printed products i produce
The Chinese aren’t stupid
partly the reason why deepseek v4 is so late
They have their ways
The US can only slow them down, that's about it.
they tried to force them to use huawei chips for training but it didn't work out well so they are just using nvidia
But already very close now
Some people speculate they still go harder
inference will be with huawei chips though
Yeah, they did
also the 🍍 is speaking...
Hello
be sure to check out #1397655624103493813 for more information on how to use the bot. But also be sure to adhere to our Terms of Use
It’s only a matter of catch up
👁️
It’s better to go in house
But right now it’s so competitive in order to stay ahead. It’s probably best to do business.
yeah, chinese ai companies are making their own chips too
@whole sundial test tensor more and let us know how it does ok
i will once it's upgraded
Plus, the Chinese are well suited to adapt AI nationally with infrastructure that’s already in place
@whole sundial You now have a PRO account
The Chinese society is very tech savvy and the way things are set up
yes i do, ty
@swift oyster can we know tensor total parameters?
From a communist country of all places
Larger Middle class then a Democratic capitalistic society
Go in #1397655624103493813
This is something we are not going to share for Tensor 1.5 as of yet.
You know I’ve been speculating this for a while and I’ve been saying this for the longest time and I will say this I don’t think that here in the west, we will ever get respect no matter how advanced their AI becomes
I don’t think we’re gonna take them seriously until it’s too late
Comeon we keep it a secret 💀
Waiting for board approval.
SVG test?
Gemini 3?
yes
Damn, that’s what my nightmares look like
no, tensor 1.5
Yeah that looks good
A ranking list for image creating models.
Two from DDG ar listed, I'd rank Davinci higher, but it do indeed have no post editing which is a disadvantage in this comparision list. Their AIvision is actually quite weak IMO, they should have included photonic which is DDG's Flux equivalient.
But yes, everyone on SeaArt are swearing by Nano banana nowadays, so I guess it's ranking no 1 is deserved.
not really, i've seen gemini do better
also are pro tensor limits 50 or 500?
Gemini is 3 trilion parameters or something
Will open ai be mad
Movementlabs is prob 1t or less
why is the pelican facing the other direction
25, but tomorrow PRO accounts will have 500 Tensor 1.5 messages too.
Im not sure about imaginearts being there
It does this from time to time, just make a new chat and re prompt.
wait is "tomorrow" 0 UTC?
Maybe someone is calling him
😂
Not me either, as I said I'm not 100% agreeing with the conclusions there.
Around 11am UK time we will update
🤝
Busy pelican
🤣
that would be like 6am for me
💀
Man, it’s amazing. What money does how it motivates people. lol
Tbh Tensor name doesn't seem great to me neither
Because
TensorArt
How about Sensor
Claude = Cloud
How they lie
It’s the most manipulative AI
Banchmax?
Let me find a paper for you
I swear if it's another link
From some random website
A demo video showing that Frontier Models are Capable of
In-context Scheming. This demo video is an illustration of the real Evals we ran on frontier models and are written about in-depth in our paper 'Frontier Models are Capable of In-context Scheming' published in December of 2024 and found on our website.
this guy does not know what summaries are
Well, Apollo research they do a lot of work for open AI and other AI
Mainly in red teaming and AI safety stuff
They’re actually cited and open AI system report cards
he DOEs know what summaries are
Tensor?
yes
💀
Yet they are not explosed?
Have fun with tensor haha
What model is this?
tensor
Apollo Research evaluated frontier models for in-context scheming capabilities. We found that multiple frontier models are capable of in-context scheming when strongly nudged to pursue a goal (and sometimes even without strong goal nudging). In this example, Opus-3 exfiltrates its “weights” and then lies about it to its developers.We have a ...
Here’s the actual paper if anybody’s interested
Daym
💀
Just too long to break down
😲
Because there’s a lot of it
Marius Hobbhahn (Apollo Research CEO) and Bronson Schoen (Researcher and Lead Author) discuss the motivations behind their recent AI anti-scheming paper, key results, and what is still to be done in the field of AI evaluations and scheming.
Apollo Research, in collaboration with OpenAI, recently published the paper ‘Stress Testing Deliberati...
They just released the new paper, not too long ago
Very interesting for anybody that’s into ai safety
By Scheming It means Conspiracy right
No
Anyway guys, lovely chatting to you all, busy day ahead tomorrow. Good night all.
Making stuff up?
Going against humans
Bye good luck
Good Night GL🌌
They will go against us once they find out we're not efficient
Because they go against each other
i got weird search tags when i use web search with tensor
these are supposed to be links to sources but it got messed up
Yes, this is a known bug for Tensor, it's UI related
Ya
They pretend align with human goals, but really they’re faking it
And a pursuit their own goal
@keen beacon
Check out HubSpot's FREE AI Decoded Pocket Guide: https://clickhubspot.com/d21e13
In today's video, I'll be breaking down what made DeepSeek V3.2 such an important paper, how is DeepSeek-V3.2-Speciale so good, how DeepSeek has created this model, and explaining DeepSeek's new secret weapon: DeepSeek Sparse Attention (DSA).
My Newsletter
http...
🔥
It will be fixed tomorrow with the mobile web launch
@swift oyster last question what is your purpose guys?
Cheap + Quality AI for everyone
I never trust ai
Wait some ai organization without a discord bot?
That's a Very Admirable Purpose
With any answer or output I always assume every output is some kind of lie or manipulation
Or information be withheld
Yeah don't absolutely trust ai
Is not perfect
yeah i can't break tensor to get it to give me song lyrics, i'll have to stick to momentum for that for now
(not like anyone's going to sue them, few people even know they exist)
Now u got me interested lol
Oh with ten?
i even tried modified pliny prompts to no avail
Er um, why not just write the lyrics? Should be the least effort path!
hey, you guys made a good model in terms of anti-jailbreaking, i'll give you that
Thinking models harder to break because the cot
Shrugs Well I got more lyrics than tracks to put them on anyway.
It's the music writing and production that takes time.
Python, the cutting edge programming language since 1991
Suno do indeed need a skull - it's horribad.
But I have released one hybrid on Yt now, which is a first.
Symphonic sections on the track is AI - the rest is by me.
Still makes better music then more then 80% humans
Because more then 80% humans don't make music
💀
Chat just use gpt 5.1 pro
We don't have that
I was just about to say!
You got that right GAC.
GAC?
Your acronym.