TheaterFire

The latest ChatGPT is supposed to be ‘PhD level’ smart. It can’t even label a map

Posted by pulse77@reddit | LocalLLaMA | View on Reddit | 4 comments

Reply to Post

4 Comments

sammcj@reddit

Off-topic, does not relate to local LLMs or local LLM tooling. Please keep posts about proprietary models and providers to other subs.
View on Reddit #64199771

Qual_@reddit

Well, it completed Pokemon Red way faster than any other model.
View on Reddit #64196133

ShengrenR@reddit

It's the same crew who were running around like decapitated chickens spamming Strawberry everywhere for months as though it was some rebuttal to LLMs.. in this case they just saw "As smart as PhD" and decided it must need to literally do everything at that level.. and anything less is some grave failing of the technology. I will say in 'those folks' defense.. it's sortof how the stuff gets advertised to them, rather than "this is a new tool that is very powerful if used correctly" which would require them stretching themselves as well, rather than magical-thinking-box-as-a-service. It may or may not get there.. but for now.. apparently they're upset it can't perfectly reproduce obscure image generation requests - I'm actually really impressed it did as well with the 'presidents' as it did..
View on Reddit #64197688

ShengrenR@reddit

Why is that relevant here? This article is written by somebody who has no clue how LLMs work.. based on a tweet.. by somebody who has no clue how LLMs work.. written for an audience.. well.. you get the picture.
View on Reddit #64194900