AI

4142 readers

1 users here now

Artificial intelligence (AI) is intelligence demonstrated by machines, unlike the natural intelligence displayed by humans and animals, which involves consciousness and emotionality. The distinction between the former and the latter categories is often revealed by the acronym chosen.

founded 3 years ago

The GPT-4 barrier has finally been broken (simonwillison.net)

submitted 8 months ago by [email protected] to c/[email protected]

8 comments fedilink hide all child comments

Four weeks ago, GPT-4 remained the undisputed champion: consistently at the top of every key benchmark, but more importantly the clear winner in terms of “vibes”. Almost everyone investing serious time exploring LLMs agreed that it was the most capable default model for the majority of tasks—and had been for more than a year.

Today that barrier has finally been smashed. We have four new models, all released to the public in the last four weeks, that are benchmarking near or even above GPT-4. And the all-important vibes are good, too!

Those models come from four different vendors.

all 12 comments

sorted by: hot top controversial new old

[–] [email protected] 0 points 8 months ago

Hmm has the leaderboard not been updated? It's still showing gpt4 at the top: https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard

[+] [email protected] 0 points 8 months ago* (last edited 4 months ago) (2 children)

[removed by mod]

[–] [email protected] 0 points 8 months ago* (last edited 8 months ago) (1 children)

There are many free LLM models and platforms to access them. You can download and permanently posses the actual model files and weights.

There are open source frameworks to run and interact with these models that run fully locally

[+] [email protected] 0 points 8 months ago* (last edited 4 months ago) (3 children)

[removed by mod]

[–] [email protected] 0 points 8 months ago

The problem is open models are nowhere close to stuff like GPT-4

Of course not, you'd need the same class of hardware running 24/7 to get similar results, and ain't nobody paying for that.

[–] [email protected] 0 points 8 months ago

Agreed but it's still a good tool thats available. You can use it to summarize large documents. Yes prolly never as capable if you have elite monies. But still worth playing and learning how to use. Imho.

[–] [email protected] 0 points 8 months ago* (last edited 8 months ago) (1 children)

I'll acknowledge that right now to get model conclusions on par with gpt 4 you are going to need a custom pipeline with multiple adversarial models, RAG and more. But it all could be built by an eager hobbyist with a strong gaming pc

To be clear this approach will not benchmark the same as gpt 4 but can indeed generate useful content.

[–] [email protected] 0 points 8 months ago (1 children)

Mistral has a lot of open source models that are quite good. Their largest ones are closed; for what reason I don't know.

[–] [email protected] 0 points 8 months ago (1 children)

None of the models that rival gpt4 are open AFAIK

[–] [email protected] 0 points 8 months ago

Yup

Firstly, none of those models are openly licensed or weights available. I imagine the resources they need to run would make them impractical for most people, but at after a year that has seen enormous leaps forward in the openly licensed model category it’s sad to see the very best models remain strictly proprietary.