Will AI have another ChatGPT moment this year bros?
 Thalidomide Vintage Ad Shirt $22.14 |
 CRIME Shirt $21.68 |
Thalidomide Vintage Ad Shirt $22.14 |
Will AI have another ChatGPT moment this year bros?
Will AI have another ChatGPT moment this year bros?
Thalidomide Vintage Ad Shirt $22.14 |
CRIME Shirt $21.68 |
Thalidomide Vintage Ad Shirt $22.14 |
Local Models will, In fact Local is becoming a black hole, everytime a breakthrough gets discovered it speeds up the progress by 0.2% (not accurate or real, just an example to see how fast things are going.) We've already entered singularity anon. Turbo, (OpenAI's AI model used to power ChatGPT) is already overpowered by open source (See Mixtral.)
I'm excited and scared at the same time anon
>We've already entered singularity anon
>I'm excited and scared at the same time anon
Not OP but curious what you think will happen soon? If it progresses and you're familiar with the recent developments you can probably assooooom something.
I'm just clueless.
It's baffling that even local model enjoyers are coping the idea we're already past turbo level. Mixtral 8x7b 100% pushed us past even the average 70bs, painfully obviously better than GPT-3 turbo in all its iterations. We still haven't quite hit GPT-4 levels of intelligence coherency, i think that's our final step. Wouldn't be surprised if we hit it in may.
>this pace is frighteningly breakneck
mistral promised open weight gpt4 killa this year
i hope they don't bail on us, but it's weird that they didn't release their "medium" model
>OpenAI's AI model used to power ChatGPT is already overpowered by open source Mixtral
Delusional FOSStard cope
>Turbo, (OpenAI's AI model used to power ChatGPT) is already overpowered by open source (See Mixtral.)
Turbo is believed to be a 20b model based on an architecture from 2020. It's cheap and shitty, but that's why it's free. It's not impressive that local is only now catching up, especially given most of the ones that have are larger models. As of right now, the only way to truly surpass 3.5 level seems to be exponentially more compute.
No model, by anyone (even OpenAI), local or not, has managed to surpass it without significantly increased parameter counts, and it's going to be a long time before hardware capable of running even GPT4 Turbo will be available to consumers.
The boom part of the cycle has ended and we are entering an AI winter now, so no, not anytime soon.
>we are entering an AI winter now
lmaoing at your life
Trillions are being poured in I don't think it's going to slow down
Nice, so they can pay the lawyers.
no, AI bubble about to stagnate and speculative markets pop
No, people are tired of the AI shit and it will all fall flat like NFTs.
AI aint doing shit. it works well enough to make you go "wow" when you use it for the first time but not for anything practical. "now its the worst it will ever be"fricks dont realize that its not developing nearly as fast as a year ago and LLMs are barely better than last year. The necessary computational power is too high making it unprofitable and openai will shit itself when the va runs dry. not to mention legal limitations.
llm is not ai
what is it then. llms are the poster children of ai
it is not an intelligence (the i in ai), it lacks consciousness and agency
1. 'AI' has been used for all ML for a year now since the media got their hands on it
2. you're wrong. llms have to simulate an agent to perform non-language tasks like algebra. they are capable of acting like an agent and a sufficiently advanced llm can for sure be considered ai with goals and a model of the world. just because an llm "mind" does not work like a human mind does not mean its not one. check out rob miles on the topic
>just because an llm "mind" does not work like a human mind does not mean its not one.
It means exactly this. The only point of reference is your own. It's the definition of the word.
Motherfricker we don't even know what consciousness is, much less how to detect it on things.
The average person's entire thought process is based on selecting next token based on their training data. LLMs may not be intelligent in the same way as a smart person, but they beat most people.
Consciousness and intelligence two entirely separate things so that part's irrelevant.
they called the first tic-tac-toe models AI, everything ML related is AI by that definition
so let's not get carried away by debating about semantics
humans have already adapted to chatgpt. Today I told chatgpt to summarize the ideas that we worked on together. It has effectively become my assistant and teacher at the same time.