OpenAI's GPT-5, will show progress on solving "hard problems" like reasoning. OpenAI's COO Brad Lightcap says these new models will be able to "take on more complex tasks in a more sophisticated way" compared to current narrow AI systems.
Meta's VP of AI research Joelle Pineau says they are working to get their models to "not just talk, but actually to reason, to plan...to have memory." Meta's chief AI scientist Yann LeCun explains that adding reasoning would allow AI models to "search over possible answers", "plan the sequence of actions" and build "mental models" of the effects of their actions.
https://www.ft.com/content/78834fd4-c4d1-4bab-bc40-a64ad9d65e0d
Shopping Cart Returner Shirt $21.68 |
>AI models can reason
inb4 it has hardcoded 'shut it down, goy-ai knows' function
BUY AN AD!
>BUY AN AD!
how much does hiro pay you to shill? double the salary of a janny?
>how much
the quadruple of expected UBI
Good
I can't wait for AI to steal my job so that i can go back to being a NEET living with my parents. At least with AI it won't entirely be my fault for being out of work so no one can blame me
Living in a soulless shithole mega city just to work at some moronic company (and being grateful for the opportunity!) is the ultimate psyops. There is no way i am ever coming back to a large city in my life. It's just too fricking depressing and oppressive. How bugmen can stand living here is a mystery.
When i was a neet i was still earning enough with my niche progressing skills to buy groceries and basic things, and i was only working like 3 hours a day at max. My parents were glad i was living with them, I was there to witness my younger siblings growing up.
Never again am i going to be misled into working for a soulless corp
>just too fricking depressing and oppressive.
what? explain further anon
not him, but you are just a cog in the machine, some find it oppressive
>what? explain further anon
You don't really know anyone in the big city. Everyone is there to make money and they have no attachment to the city itself or it's people. The kind of people who move to cities for employment are also homosexuals with chip on the shoulder who think someone enjoying their time or doing well is somehow going to be bad for their success so they get jealous quickly but are still massive cowardly subhumans so they don't say it to your face.
The isolation you feel after you come back to your cold empty apartment after a long commute and work is horrid, and dehumanising. Why must you be away from your friends and family? For what, just so you can earn more money? Is it really worth it when you could survive just as easily on your family farm income or by doing odd jobs around your Village? At least you would be able to see friendly faces at the end of the day
why don't you try to find dates? Is it really hard to find? What city are we talking here?
you dont/can't "make" money, you can only transfer from pre exisitng ledgers.
Only ADI's (banks) can issue loans that makes money by FRB (10x infralition through loan issuance).
Or the mint by printing money.
I'm so with you, brother. Currently neeting after getting a taste of the 'end-game' society was so proud to introduce me to. Never had money in my life - that changed for the time, yet I felt so incredibly hollowed out and generally depressed. Now I work on my own projects, can't afford many luxuries, but I am my soul is back. There is no comparison between being sad + wageybrained and sad while neeting.
As a neet, I get up and do pretty much the same shit every day regardless of my location so I don't really understand this sentiment other than you just want to be away from people in general.
I don't believe it until we can see
Reasoning means pretty much ai in real sense. I don't see it happening.
I think it's pretty believable. They clearly found the correct paradigm with large NNs, the rest is just figuring out the best architecture and training approach.
Once that is solved it's good night for people who have jobs with low interactivity.
Wrong, until AI can do things perfectly every single time at best they will be assistants for professionals. Even if the AI is correct 99% of the time, you still need someone for oversight. Ultimately this just means 1) you'll need to use AI to be competitive and 2) you will be an AI tard wrangler for tasks you don't like to do anyways.
No it only needs to be as good as humans. The tard wrangling is not something that is beyond AI, neither is consulting stakeholders on technical matters.
moron, humans don't hallucinate bullshit in the middle of their code. AI literally can't be trusted, every single line of code needs to be checked because the AI will make up non-existent libraries. You can trust a trained, standard human to do a job correctly. Not that there won't be people who try to fully automate with AI but they're going to get fricked. The second their code base is more than 100 lines.
If you think hallucination cannot be solved I have bad news for you lol. The reviewing can also be done by AI btw.
Also just as a reminder: 12-13 years ago the hottest shit in AI was being able to recognize a dog in a photograph. Unless something arrests human progress pretty soon this thing will take your job. Low interactivity knowledge work is on its way out.
Hallucination is a fundamental part of transformers. But keep holding your breath for cold fusion, cheap energy! Any day now!
You literally only have to add something like a parser and a compiler to check if the code compiles and regenerate it if it doesnt.
WOW ANON I CANNOT FATHOM HOW THIS MIGHT BE A BAD STRATEGY
>moron, humans don't hallucinate bullshit in the middle of their code.
lol
lmao even
I bet everytime you click 'build' your code compiles succesfully.
I bet you don't need to look into documentation either when writing it.
>please keep the vc money... please... 2 more weeks and our llm will start to reason... it will be giga profitable... costs will go down... don't let us freeze...
buy an ad, homosexual.
Microsoft/OpenAI wants to brute force it by building a 100B AI servers. "Stargate" that unlocks the potential for AI.
This. Once you have some kind of vague feedback mechanism + a reasonably competent AI, you can brute force the solution. You might not be able to do it for competition level tasks, but crudmonkey tasks could be easily automatable.
Let me guess it's going to have an inner monologue solution that we see for AI Agents but still be fundamentally an autocomplete algorithm that corrupts into memes and repetition and still can't yet determine fact from hallucination.
This isn't really a surprise, there were very simple improvements like ToT and whatever the frick Q* is. As for memory maybe they're implementing Mamba? Shit, I don't know.
>destroys AI
oh no no no agibros!
>these models are actually just a fancy __________ and they can't really _____________
https://en.wikipedia.org/wiki/AI_effect
Any sufficiently advanced technology is indistinguishable from magic.
>Anon rediscovers what a function is
no they ae not
AI unpacks the entropy meaning or words/phrases ,ssentence-paragraph - context.
then WAI finds the least entropic path it can via gradient of decent, and then clothes that found path into words again.
>hacker israelites midwit discovers the meaning of markov chains
yeah a markov chain with one trillion nodes, so easy bro, right? we can just calculate the transition probabilities by hand, bro
why is it they always have to be the vocal morons
How will it reason 6 million in 6 years?
Can we stop calling it AI and start calling it nvidia brain
>Specifically train on more factual Q&A data
>Now AI can bullshit even better, but even harder to catch
Last time it was some bullshit Devin scam, now it's this kind of bullshit again. AI VC chinks need to be stripped naked and lynched down the streets of San Francisco
This might get scary when people get used to making it think for you. Then get a brainchip to be able to make the process even faster, perhaps even power it partly by your own brain.
I wonder if it's possible to jailbreak women and have sex with them then?
Kek, so naive, anon. You can jailbreak human brains right now, it has always been a thing, its called hypnosis.
If you want to get trippy about it however, AI will inevitably be extremely good at this. Pay particular attention to when AI starts studying individuals in specific and in lenght. Might not be during our life time, but eventually every human will have an "AI physician" assigned since birth that will monitor and collect data of their every bodily function and mental states throughout their lives. When the data is as precise as that, "jailbreaking" through any input will become natural to the AI.
In simpler terms, they will know exactly what buttons push you. I am not talking about convincing, It would be a lot more mechanical. A sound that instantly makes you fall asleep, an image that gives you adrenaline, a sensation that plants ideas. It's eventually happening.
Wow great now we get official CoT but rebranded as QuietSTaR which OAI was already using under the hood can you believe it?!
Wake me up when they use an architecture other than transformers and cut the vram bloat.
They've had that shit since the 90's.
>OpenAI & Meta AI models can reason!
>Can reason
Okay then, please tell me. Assuming that the level of technology is equivalent to that of steel and industrial Europe in the early 20th century, is it logistically efficient to attempt to arrest, imprison, execute, and dispose of anywhere from six to eleven million people through any form of mechanized industrial process?
morons can reason too, their reasoning is just moronic.
Unless they've made actually breakthroughs in the architecture of neural nets, I don't see this making a huge difference in novel tasks.
They're full of shit as usual. Don't trust the marketing israelite.