News for hackers. Stuff that matters.
Re:Markov chains are funnier than LLMs by eru on Sunday August 18, 2024 @10:11PM (#41287196)
Some of the open-weights models come close to what you might want?
Re:Markov chains are funnier than LLMs by dartos on Sunday August 18, 2024 @10:43PM (#41287342)
Sort of. They fill the “locked” requirement, but open models aren’t SOTA like gpt or Claud.
Re:Markov chains are funnier than LLMs by coder543 on Sunday August 18, 2024 @11:08PM (#41287447)
Llama3.1-405B benchmarks at a level that I would certainly consider comparable to other SOTA models, as long as multimodality isn’t needed.

Regardless, OpenAI provides access to quite a few of their older models through the API, since the API lets you pass in a specific model version. I’m sure the older models won’t be available forever, but that is a much more stable target for researchers than just opening the ChatGPT website and typing in things.

Re:Markov chains are funnier than LLMs by dartos on Monday August 19, 2024 @08:47AM (#41290581)
ChatGPT specifically is more than just a chat interface on top of got-4.

Their system prompt includes the current date and time among other information, making it very very hard to run reproducible experiments against it.

But it’s the tool most people are using.

Re:Markov chains are funnier than LLMs by Reubend on Sunday August 18, 2024 @11:11PM (#41287458)
LLAMA 3.1 405b-Instruct is state of the art at the time of writing. It averages slightly worse than ChatGPT 4o, but to a mostly negligible extent for most research.
Re:Markov chains are funnier than LLMs by toxik on Monday August 19, 2024 @06:21AM (#41289655)
Aside but -- it is also aligned to not generate profane text, right?
Re:Markov chains are funnier than LLMs by throwthrowuknow on Monday August 19, 2024 @07:05AM (#41289901)
Abliterate it