it's still going fyi. can do multiday coding now but still crashes for unknown reasons. previous session ran for 6.6 days straight. still not done implementing like half of the pseudocode cc @walden_yan
@swyx
-
Setting up AIUX collaboration request with team mentions
By
–
@vibhuuuus can we set this up for aiux cc @Mappletons @thesephist
-
Kaplan Scaling Laws and Chinchilla Ratio in GPT-3 Sizing
By
–
i think this implication is drawn from the context re: kaplan scaling laws paper. and yes could extrapolate but i think the chinchilla ratio contributed to gpt3 sizing?
-
400B Model Expected to Surpass GPT-4 Performance
By
–
they were “only” trying to match gpt4, and now 400b is likely to beat it
-
Opus Brief Reign Ends as Llama-3-400B Emerges as GPT-5 Contender
By
–
Opus only lasted like 2 weeks at the top RIP Llama-3-400B finna be gpt 5 level i think these guys may have overshot lmao
-
Jason Exploits Variant of Shunyua Yao’s Koala Architecture
By
–
and jason is exploiting a variant of the @ShunyuYao12 Koala architecture
-
GPT-4 Training Cost Estimates Questioned
By
–
most estimates have gpt4 all-in cost at least 9 digits. idk how they calculated the compute
-
Learning from Open Source AI Apps Feature Development
By
–
take @nutlope
’s notesgpt repo (or really any nontrivial oss ai app) and watch them add a feature -
AirChat Gaining Popularity with TIES-Merging Explanation
By
–
ai airchat is popping. i asked for an explanation of TIES-merging (ft @maximelabonne
’s model merging guide) and @nisten obliged: https://
air.chat/nisten/post/82
887bb6-5f2f-4c8e-9941-a6bb625f1282?thread_id=17843165-f817-3f7b-84f0-14476b6cd8f6
… airchat is basically twitter remade in @naval
’s image, but one benefit is that you cant write in pompous corpo language, -
Moving Goalposts: Redefining Success for AI Software Engineers
By
–
are we so good at moving goalposts that we call it "disappointing" if Devin merely replicated a remote junior software engineer? its up to us to figure out how to use it