Re:Ask HN: Why is GPT4 better than the other major LLMs?
by takinola on Saturday November 25, 2023 @04:43PM
(#38416974)
I have a test harness that runs all the LLMs side by side to compare the quality of output. I have tried Claude 2.1 (the new Anthropic model) and it is better than PaLM but not as good as GPT4.
I am skeptical that Google is holding back a better model. It would go against every instinct that all the tech companies have, especially if you believe that AI is the future, why would you hamstring yourself by putting out an inferior product when you have a better one gathering dust in the back? Google has oodles of cash, so not sure why cost would be an issue. I don't get the safety argument so I can't speak to it but unless you have AGI it seems premature.
Re:Ask HN: Why is GPT4 better than the other major LLMs?
by ilaksh on Saturday November 25, 2023 @07:04PM
(#38418000)
It's widely reported that Google have models that are not publicly released and that part of the reasoning at some point has been cost effectiveness.
Re:Ask HN: Why is GPT4 better than the other major LLMs?
by gtirloni on Saturday November 25, 2023 @04:49PM
(#38417036)
> Also Google and Anthropic are basically deliberately holding back their strongest models because they are too expensive and/or they are worried about safety or something.
Any place I can read about this?
And note that there are new versions of models from Anthropic that have just released or could release within a few months.