[anonzzzies]

For sure, but not for that reason; there is currently no one with a plan how to go from current (LLMs) to a better model. It's some 'more focused training' 'better prompting' 'agentic' 'smarter lookups' 'better tooling'. But fundamentally, this model is simply shagged out and it'll get a little better with the above, but the jump everyone is waiting for cannot happen without a new model invention.

[tim333]

No one you know of but I'm sure people are thinking about it.

Which reminds me that one of the most obvious failings of LLMs is they never say "I've been thinking about that and have a new idea." The thinking leaning thing needs work.

[anonzzzies]

I know multiple people who are working on this: there is just no progress yet that is more of the same and that won't work.

[FergusArgyll]

My point is; maybe we can't prove that until deepmind gives us their best shot

[anonzzzies]

But aren't deepmind, in this infinite money AI times, giving it their best shot?