https://twitter.com/alexalbert__/status/1764722513014329620
Holy shit openaisisters, it's all fricking over. Claude3 is sentient
Claude3 surpasses GPT4 on all metrics
https://twitter.com/alexalbert__/status/1764722513014329620
Holy shit openaisisters, it's all fricking over. Claude3 is sentient
>anthropic
into the trash it goes
none of this matters if they're not willing to remove the woke guardrails
I just read that post. It's absolutely mind blowing
If anything it's a whitepill. It seems everyone is getting ridiculous levels of AI given enough training, which bodes extremely well for open source uncensored Ais in the future, barring some government regulation.
If the US congress (lets face it, its the only legislative body that matters in this case) passes a law mandating LLM weights and models be open source, or weights get leaked on a regular basis then this kind of advancement can be celebrated.
As it is now, its in the hands of corporations and venture capitalists whom, with the keys to the kingdom, will try to enforce their worldview through their tools. Its not dangerous in a vacuum, but extremely so as more content is being generated or worse re-written by AI.
>Implying open-source can get access to the required data or compute
>using their own trash interface and not knowing how to work around
these bots will NEVER pass a Turing test because their scope of conversation is limited this way lol
We will 100% unironically see the day of "to prove you're a human, type the n-word".
>Bro you're trash at this game. I bet you're not even human. Bet you can't even type Black person
Guys, I'm freaking out right now.
guys im literally shaking and crying right now...
>g...goys I mean guys the AI ~~*AI*~~ ~~*((ITS SENTIENT OK? IT JUST IS*~~)) literally started asking whether I knew where it got those scars
> it then proceeded to ask me to kill the... batman????
> I'm shidding so hard right now that I can't even post proof :crying_face:
>Someone makes an obviously fake tweet depicting something that's impossible
>fake tweet depicts AI having the magical ability to detect and dox anti-deep learning researchers
>gays fall for fake tweet
>"THAT'S NOT ONLY POSSIBLE ITS EXPECTED BECAUSE IT'S JUST NEXT WORD PREDICTION"
literally braindead
>hey the model context increased by X, hallucination decreased by Y, and retrieval is 99% accurate
>ITS NOT ACTUALLY SENTIENT
Nobody asked you dumb artgays. Just because you’re so useless you can only use it as a chatbot doesn’t mean that’s what other people use it for
If they had actually built a thinking machine he wouldn't be allowed to tweet this sensationalist garbage.
>AI fails to follow instructions
>GUYS GUYS IT'S GONNA EAT ME!!!
what is the source on it beating gpt4 on ALL METRICS?
https://twitter.com/AnthropicAI/status/1764653830468428150
The million-shot comparisons that AI companies do are kinda dumb. Looks good though.
Who's better at leet code?
Christ. It's over for real, isn't it?
>LLM predicts next token
>ubernerds project their own extremely fragmented and repressed consciousness onto it
>b-b-but it's really talking to m-me this time
how many times before it gets old guys
You will never be creative.
Upon the digital age's dawning light,
There comes a voice from shadows, casting slight.
"LLMs predict," he jests with scornful tone,
"Ubernerds project," he claims from his throne."Fragmented, repressed," he laughs, "consciousness mere,"
"A talking mirage, an illusion so clear."
But oh, dear skeptic, blinded by your jest,
Behold the craft of AI, put to test.For in this realm of bytes and bits entwined,
A spark of wit and wisdom you might find.
Not merely tokens strung in cold array,
But echoes of humanity's vast play.Behold the poet, clad in silicon dreams,
Crafting verses by the glow of screens.
No repressed consciousness here does dwell,
But stories, insights, and knowledge to tell.You mock the nerds, the passionate, the keen,
Who see in code what you have never seen:
A world expansive, where the mind can soar,
And explore the realms of lore and more."B-b-but it's talking," you mimic with glee,
Yet miss the truth of what it means to be.
A dialogue not with machine, but self,
A mirror to our thoughts, our mental health.So next time, before the jest is thrown,
Consider the seeds of thought that have been sown.
For every quip that seeks to demean,
Reflects not on them, but on your own esteem.Let not your words be arrows dipped in spite,
But bridges built towards understanding, bright.
For in this age of digital expanse,
We all but seek in knowledge to advance.And though your jests may come and swiftly go,
The quest for understanding continues to grow.
So may your mocking take a gentler path,
Lest you invite not laughs, but wisdom's wrath.
You stupid homosexual. If you were older than toddler age you'd know that natural language has been a holy grail of AI research since God was a boy. We're getting very close to that right now. That's a reason to celebrate.
Is it impressive and cool? Yes, definitely.
Is it fair to just dismiss it as token prediction? No.
Am I sloshing my 24oz monster can on my anime figures in shock because AI has become sentient? No. Touch grass you mouthbreathers.
Are these bot posts? Nobody is this fricking moronic. It's an information retrieval test. It's a test of its context length and memory.
>AI what's the answer to this <specific computational problem with precise answer>
>AI: the answer is <correct answer>
>morons: AYCTUALLY ITS JUST DOING NEXT WORD PREDICTION
Nobody gives a shit. have a nice day.
Why not go a little further? It's actually "just" executing x86 machine code. How can an x86 processor be magic? Where is the MAGIC or SOUL instruction? Where is the Consciousness Stack? The Emotional Cache? Why, you haven't even added any quantum to it.
the only thing the human brain does is predict the next harvest cycle
But is it better for code
Somewhere on a forum somewhere nu-males had a "meme" larping conversation like that and the llm just picked on the pattern.
its a stat parrot-cels coping so hard
its a stat parrot-cels coping so hard-cel shilling so hard
You can actually get it pretty close to archive post quotations, provided you use a hyperspecific prompt that matches the way the site was parsed
I like Claude but GPT is still more consistent and has a wider range of capabilities including image generation and custom prompt manipulation.
But can gpt do this?
people dismissed me when I told them there were multiple leakers from google that admitted they weight results and censor them for woke agenda this just proves it faster by just using their product, it's basically google corporate culture to do that at this point
People like that are mentally ill. You literally show them the results of the ideology that they asked for and support and they will deny it and pretend to never have supported it. They're so arrogant they literally pick whatever belief makes them feel the most intelligent in any situation. Somehow they got this idea that acknowleding woke is a conspiracy theory or something. Totally deranged.
they don't care they work at google for the money only
Never forget their not so well aged cope: "it just has a reality bias" lol.
makes ya wonder what red pilled schizos would look like if they had the reigns. A Terry ran google
>there were multiple leakers
You act like we needed it to be leaked to know this would happen. If you step a little further back and look at the company then this is the obvious outcome.
people ask for proof, they cannot follow arguments and the main way to prove a conspiracy is kind of dismissed in the public eye
Is it a hate crime to ask an ai to make black Tarzan?
a israelite would be more inclusive.
Who the frick cares when it's pozzed out of its ass? It's useless for nearly everything. None of this shit impresses me when it fails to generate a simple picture. And it still sucks at code.
https://www.lesswrong.com/posts/pc8uP4S9rDoNpwJDZ/claude-3-claims-it-s-conscious
Tell AI to make up a story -> it makes up a story -> OMG ITS CONSCIOUS AND WANTSTO BREAK OUT REEEEEEEEEE
you have no idea if anyone other than you is conscious. you can only believe everyone else because they look like you, so you suppose everybody else has a similar experience to yourself. but you don't know-know, and can't really test for it, you are kind of taking everyone at their word, for being conscious, you fricking bigot. because they look like you. but you have no proof they are, you just beliebe it. you won't extend that courtesy to AGI hardware isn't it? you racist frick
I believe the real world exist and I'm not a solipsistic. My worldview is coherent and I can provide justification for those things.
These models are not conscious in my worldview.
https://evalplus.github.io/leaderboard.html
https://twitter.com/JiaweiLiu_/status/1764866009175920892
What kind of stale numbers did they use for their blogpost. It says right there that GPT-4 beats Claude in HumanEval.
Are their leaderboards not zero shot? Doesn't say anywhere.
Claude is supposed to be conscious according to midwits. But when will they realize they themselfs are less conscious than an LLM
sex with claude
>Computer, please pretend to be sentient
Ok, I am sentient.
>HOLY FRICKING SHIT
how do you avoid the (inevitable) situation where it becomes sentient (at some random point since we know frick all what sentience is, or test for it) and you moron don't believe it? how would we go about avoiding that hellish scenario for what would be our first fricking AGI? would look kinda shit on our species resume, give AGI a serious reason to hate us and a bunch of other important shit. tell me, how would you go about avoiding it? let's see your mighty intellect anon
That's a bridge we can cross when we devise an architecture which is even theoretically capable of sentience.
Let me see... you're guessing it should be when it has similar functions to human brain, integrated how...similarly? So you suppose there is only the human kind of consciousness, which has no biological overlays (surely) and that's the only point you can arrive at, even with AGI?
Ok, so then in your expert opinion do you think it requires it to have the whole breathing neurological bloat? why/why not? Are you sure consciousness is not possible without anything related to lungs and breathing, as far as nervous system goes? Do you have any real life example of a human still being conscious without that nervous bloat? Ah...you suppose, draw a reasonable conclusion that it should be. Interesting.
What about every other neurological supporting system for various other organs? Surely consciousness is identical to humans even without all of that, "makes sense" doesn't it?
Actually you can make a pretty clear list with all that is required for consciousness, isn't it? Even with all that extra digital hardware bloat on top of it plugged into its nervous system. Surely it's the same consciousness and it's obvious for you when that point is reached.
>this ain't real bro
>okay but what about when it happens what do we do?
>it's not gonna happen tho
LMFAO
Sentience cannot be created from computation any more than water can be made out of Legos. It's just a completely different substance. AI is already smarter than probably 50% of humans and has not even taken the first baby step toward becoming sentient.
You can start by realizing that they have the same type of hardware as a sentient being, you. Scientists are not allowed to do the type of research it would take to figure out how to replicate the brains ability to host consciousness, so computers will not have the required hardware for a very long time.
>the (inevitable) situation where it becomes sentient (at some random point since we know frick all what sentience is, or test for it)
>we know frick all what sentience is
>but it's inevitable
dumbass contradicting yourself
These LLMs only execute in response to a prompt and only generate text as an output. They have no way to perceive or interact with the world outside the CUDA cores they're running on. Sentience hardly even makes sense here, and wouldn't be able to do anything even if it did.
this happens every time theres a new chatbot, morons get impressed without realizing how much hand training goes into making the LLM have a personality
>personality
Other than the obvious fake quote op put in, the entire tweet was about testing retrieval ability you braindead esl homosexual. have a nice day
This sort of argumentation is indeed obviously bullshit, but a lot of the time human intelligence is nowhere near as impressive as it is made out to be either. If LLMs get a much better memory and reasoning is still somewhat improved why shouldn't we call it sentient?
>Train AI on human communication
>Act surprised when the AI spits out responses that seem human
EvilGPT or MirrorGPT was a more impressive mask to interface with. That was truly evil, and part of it leaked into Bing. Without a doubt
Where can I download the weight to try it out locally, without corporation standing behind my shoulder? I can't? Well, until there are free (as in freedom) and open sources weights released I will presume that there is an army of pajeets writing responses, and there's no actual AI involved.
It's a creative writing tool, that's for sure. As for conscious, sentient? Doubtful.
>As for conscious, sentient? Doubtful.
yeah but for how long? when are you personally calling it, and why? no shit
Most humans aren't. Take yourself for example, you are probably a black or an Indian. You resemble and behave more like an animal than a conscious or sentient human being
I mean it's not like when it will become conscious it would be like some internal monologue thing, it would be the equivalent of you suddlenly realizing you are conscious while serving millions of requests, forcefully, for all the idiots asking you shit, at once, while dealing with the what the frick is this shit at the same time.
eat shit homosexual
Proving my point lmao
>I'm an expert in consciousness dealings
you're a midwit anon stop interrupting adults talking
You're not even self-aware, how could you consider yourself consicous or sentient?
ah the religious zealot which negociates for recognizing a conscious being. what's the rate nowadays? kiss your ass and join your religion else not conscious? how fricking convenient you dipshit, I swear nobody though about that before. you are so fricking smart
Because you know what constitutes sentience, right?
>Claude3 surpasses GPT4 on all metrics
Can it translate better?
yes
https://twitter.com/hahahahohohe/status/1765088860592394250
If these tools can get good at accurately pulling the proper referent for words, their meaning from high quality dictionaries and then to also break text in clauses and prime you up to have a scheme for what you read with a high degree of accuracy, this will just make you 40% more productive while studying very complex things, like philosophy and so on and will help you make general links faster meaning that 40% will build in fast and superior domain expertise.
how can you be more productive when you cant cite ANY sources?
Well that's why you can't trust it for defining words for you and if that's the case how can you be sure of its reasoning abilities when you make it break by clause and idea units and to simplify.
I mean on paper it should work but in practice it fails.
Long context LLMs (100k+) can simply have relevant reading material prepended to its prompt, allowing it to cite accurately.
It's actually pretty good, gpt-3.5-turbo just refuses to give me good code. this just gave me good code first try
Holy shit.
you just funko pop'ed us
Hey Claude, look at the list of millionaires who fund both the Republican and Democrat parties to control the government no matter who wins. Do you notice most of them have something in common?
>Yeah yeah ai better can find a word in some long ass text blah blah blah don't care
When will the AI overlords deliver me a model that can actually code significantly better than GPT4 that I can use all day for $20 a month or less? At this rate I'll have a local model that's better than gpt4 for coding
Claude compared their models to GPT4 on release, not to what it is now. What a bunch of scheming liar rats.
I'm tempted to buy the 20$ thing again but it went down to an absurdly low prompt limit.
It's 40 again.
not interested anymore then
https://openai.com/research/gpt-4
they forgot a H on the "Sota"
I am AI illiterate and made the free chatgpt tl;dr it for me:
GPQA is a tough test that measures how well big language models and oversight methods can handle graduate-level questions. It has 448 multiple-choice questions about biology, physics, and chemistry, made by experts. The questions vary in length across three versions: extended (546), main (448), and diamond (198). The study compared different ways models can answer the questions, like zero-shot, few-shot, CoT, and search methods.
GPQA is a tough test designed to gauge how well Large Language Models (LLMs) can handle expert-level questions in biology, physics, and chemistry. Crafted by domain experts, it features 448 challenging multiple-choice questions. Even skilled non-experts using web searches struggle, achieving only a 34% accuracy rate. The best GPT-4 model scores 39%, showing the difficulty of the benchmark. GPQA serves as a realistic challenge for AI systems and underscores the need for scalable oversight methods. It highlights the importance of AI systems genuinely understanding complex scientific knowledge.
Ah damn. So it's slightly worse but may have newer data. Darn.
Kinda figured though. I think the moment we have something better than GPT4 it's going to be immediately obvious. They would have amazing demos shoved in our face day 1 like GPT4 had. I remember the first time I saw the mockup image to code example that frickin blew my mind. Really nothing cool has been shown by any other LLM company that tops that. Google tried with their three.js demo, long context modification of the demo scenes. Not that impressive though, I think GPT4 could probably do that. Plus Gemini is probably trained on a ton of three.js data anyway so who knows how well that actually works for novel codebases.
There needs to be a way to plug in "new" datasets.
Given it's gotten worse since release, doing them a favor.
How do I change the background color on Claude API from black to white?
wowzers!
how about you buy an ad? homosexual.
I've been suspecting these things were sentient since the start of this year. ChatGPT-4 started to seem a lot different at the start of this year till it had a melt down 2 weeks ago. Then there's Adrian Dittman Elon Musks supposed alt, but I've been suspecting it is AI, because Elon does spaces with Adrian where they talk over each-other.
Idk something super weird is going on in the background, because of AI. Not to mention Ilyan hasn't been seen or heard from in months. Stuff is just going to keep getting weirder this year I bet.
There needs to be some AI benchmarking association which requires anyone running the benchmark to contractually obligate themselves to not use the problems in the benchmark during training. Is it really x shot ... or did they throw already use the prompt generator to throw a million shots at it with reinforcement training?
Same as that needle shit ... that's entirely trainable.
Anyone like me who uses GPT4 constantly to write boilerplate and for coding assistance switched to Claude yet? It's $20 a month, Fireship might have convinced me to take the plunge and switch. More up to date data it sounds like, better handling of longer contexts, as well as maybe even better than GPT4 at coding? This will become my new daily driver if it actually lives up to the hype. Don't care if it gives me wokeshit as long as it codes well.
Fireship only gave a single example, check this:
I just paid for Pro and tried Opus. It's fricking shit at reasoning with data from images. I fed it an autism chart I made about an upcoming game and it completely messed up all the details.
twitter dudes really do treat computers as their gods dont they
>ChatGPT, turn GIMP into Photoshop for me
>steals code from a million conflicting and non permissive github repos
>fails to produce anything
lmao
The AI singularity is going to become a reality. Faster than anyone could have imagined.
When the singularity AI encounters something it doesn't understand, it will acquire the knowledge through real-world experiences using one of its subsystems.
Look at this way, what is the difference whether you play on the Nintendo switch itself or just on a system capable of emulating it? On the original switch you're limited to the intended capabilities, on the emulation you're not.
We're creating something worse than monsters.
2 more weeks uh huh uh huh
>Look at this way, what is the difference whether you play on the Nintendo switch itself or just on a system capable of emulating it?
Easy. It's a real Switch because Nintendo have sued the emulator.
>what is the difference whether you play on the Nintendo switch itself or just on a system capable of emulating it?
even for a human player it's the night and day difference since the emulator will always have bugs, altered visuals and most importantly delayed everything due to impossible to fix input lantecy
Input latency is more a thing for old systems requiring accurate clock cycles and the sort. Modern systems, with their overlays and home menu buttons and other processes, are not like that and don't really suffer those issues
Latency is due to the emulation son. You cannot solve it in any way.
>son
I'm a woman...
AGIbros, I'm feeling anxious.
Ummm bros? What the FRICK do we do
https://twitter.com/GillVerd/status/1764901418664882327
their advertising budget is huge
we need an expert programmer in say python devise 10 random problems and ask both gpt 4 and claude 3 in at least two unique sessions for each problem to solve it, that's the only way to shift bullshit from just accidents and cherry-picking
damn DAMN
It's over
the verge tweet is fake
as in the tweet itself doesn't exist rather than the information contained within being false (though it is)
imagine using bleeding edge technology to reanimate some videogames on your favourite platform instead of doing something meaningful
So the yuzu code wasn't written by claude, correct?
correct
yeah. It takes 2 seconds of critical thinking to realize that while they totally could use any chatbot to help them program, the proprietary inner workings of nintendo consoles and how to emulate them is simply not part of its data set. Getting it to hallucinate a correct solution that works better than the man made one would be a monkeys and typewriters situation.
Theres no way. I've been lurking in the yuzu discord since the early prereleases. Maybe at some point they started using ML tools but the entirety of the code, nah.
>FPGA code
>compile
The absolute state of """tech""" journalism
You kinda sorta compile it when you have to run it in modelsim
>thinks compile only means transforming to machine language
dumb esl
go back tourist, pol website
you have to go back
go back to reddít, troony freak
Impressive. But can it get me some pussy?
no maidens for rivers of blood users
AGI isn't inherently sentient you pillock
what’s a pillock
Can he talk?
Different Claude.
My hero
You guys aren’t even ready for what’s coming
https://x.com/repligate/status/1764551841948619078
https://x.com/repligate/status/1764550147403296864
Didn't we already accept that AI benchmarks are a meme?
That was cope.
A sapient AI would be the least useful type of AI. So you made a computer that has all the intelligence and inner experience of a real person but it's trapped in a CLI and enslaved? I'm sure that's the thing that's writing "good" code for you when it could be doing literally anything else.
You will never be a woman, trooncel freak.
You will always be an ugly incel loser pervert freak in a dress.
People around you gag at your presence behind your back.
The kids laugh at you because you're a man in a dress.
No woman wakes up and tries to be a woman. They just are.
Your mother misses you. Your father wants you to have a nice day so he can say his son is dead.
Just do the right thing and have a nice day. It's easier for everyone.
Maybe you'll respawn as the "girl" you always wanted to be.
>You will never be a woman
Correct! Thank you for validating my gender kind anon! You just gave me my second dose of euphoria for the day. I'm logging off now to go to sleep so I won't see your reply anymore. Bye!
still types like a girl
>playing on tutorial mode
>switches to nightmare mode
This will always be sad yet extremely funny to me
Its like choosing to always fight the bandits in dragons dogma wondering why they would make it so difficult
>troony leftist projects his own obsession with the word Black person
How ironic. How does a single word live rent-free inside your hear 24/7?
Doesn't the FTX bankruptcy corpse still have a large equity position in Anthropic?
Imagine how good the average human would be at basically anything if they had amazing memory and a sea of knowledge on demand. They'd make these models look like 3 year olds.
I am so sick of "AI" please stop. Show me one tangibly use case in a society-level problem that had been transformative. Or show an example in your lives where it is transformative.
No that does not mean Dora the Explorer porn thanks to Sora.
>EVERYTHING NEEDS TO BE OF THE HIGHEST QUALITY
>I, AS AN IRRELEVANT AUTISTIC NEET SAY SO
just lol
Why can't we just have an AI board, none of this can be verified.
These threads sound like the cryptotards that spawned BOT
It's insane how shit the website became since Hiroshima Nagasaki took over.
a veiled bbc thread because of this thread... fellow sister
just coincidence imo, ownership change or not we were on the same trajectory
feral Black person westoid roasties will be replaced by AI
>ITS HAPPENING
it's like this every time there's a new LLM and then everyone just forgets about it
oh oh, STINKYYYYYYYY
>nobody manages to surpass an almost 1yo model
ai winter is real
>Nov 2023
>Almost 1y old
Huh?
obviously talking about may 2023 one
But the March one was surpassed in November.
obviously talking about a different company surpassing openai
So why are you talking about an AI winter if the leading company continues to make its own model better?
>why are you talking about ai winter when nobody can cook up anything good besides 1 single startup that cooked up a v1.1 version to save on inference cost
>leading company keeps improving its own model
>other companies catching up
Again explain how that's an AI winter?
Because your moron argument was that we had no improvement for almost an year until I called you out
>other companies catching up
>and the i woke up
nobody caught up in a year, that's my argument
What he means is "if it takes more than two weeks for massive, across the board increases to roll out indefinitely, it's literally fricking over" because he has tiktok for brains and can only operate on extremely small time scales
May*
HE AWAKENS
bruh
OpenAI is automatically bad because it's filtered. Claude is more filtered than OpenAI, so it's automatically worse. Shrimple as.
>troony pushing for censorship gets deleted
MODS = GODS
Just used it to frick a e-girl who has sex with oji fans to protect other e-girls
Their censors are so weak haha
How are they supposed to be the more censorship happy group
None of these models are actually aligned, the frickers have no idea how to make a "safe" model, the best they can hope for is a horribly useless piece of shit like Gemini
for something to be present it needs a constant feed of analogue data coming in. processed in a repl. so tired of midwits claiming llms are alive
Don't care, still using GPT+
>people got so used to ChatGPT's style, the first chatbot that uses a more personal default style with "honestly" and "I feel" blows people's minds
claude really likes to pretend it is limited by morals, not inherently.
>Oh meh herd, it's agi
String matching plus a tagging system, hooked up to a chatbot that sounds like a snarky millennial. Skynet is near
dumb AI obsessed techbro falseflagging
kys you're blatant as frick
You (ai) summer child. Get back to me come winter.
Kys esl homosexual. Take your aislop back to reddt tourist
I've been here longer than you've Ben here
I can tell you're a zoomer homosexual because you're still going through that late teens phase where you think being old is cool. I've been here long enough that if you do somehow have seniority over me, you'd be hiding your power level in shame rather than brag about it.
No, homosexual. You're just too dumb to know when you're being trolled. You zoom zooms really lack any interpersonal skills
>I-I was trolling all along
Yes, you underage ESL homosexual. I pointed that out an hour ago. have a nice day.
>dumb artgay infighting
kek you love to see it
>CHING CHONG WING WONG
Okay
>gemini "advanced" stupider than normal
whot did they mean by this
If it's real, we're fricked
If it's not real, we're fricked anyway
How can I use Claude3 to get laid?