pull down to refresh

100 sats \ 4 replies \ @Scoresby OP 8h \ parent \ on: As Rocks May Think - Eric Jang AI

is this just because he's using monorepo wrong? (I understand the contradiction to be that it's necessarily not a monorepo if you're relying on agents?)

write
preview
102 sats \ 3 replies \ @optimism 7h

It's because monorepos lead to regressions, so if you run on a monorepo, you're going to be a step behind your competitor that can actually manage software development like a pro, and you'll keep losing all the time.

Don't believe me on this. Just ask OpenAI (confirmed monorepo: #1040555) how successful they are in regression management (or just check their ELO on arena)

Therefore, if you run a monorepo, you're per definition not ready.

reply
0 sats \ 2 replies \ @optimism 7h

Here's the current ELO:

GPT from 2 versions ago is ranked 9 and beaten by every direct competitor. The new versions don't even show in top 20. gpt-5.2-high is ranked at 21, below open models like Kimi and GLM

reply
100 sats \ 1 reply \ @SimpleStacker 7h

What tasks are they being scored on? I've found ChatGPT to be a bit more useful for general purpose queries while Claude is better at technical stuff

reply
0 sats \ 0 replies \ @optimism 7h

This is overall "text chatbot". There's categories, see https://arena.ai/leaderboard/

For me, the only thing that GPT currently does well is deep research, as good as Grok, better than Claude and Gemini (though funnily, arena disagrees with me.)

But since Grok responses are much less annoying than GPT's emoji flood, I use the former.

reply