Chris Rabl

you should be gremlinmaxxing

thank you @modal for existing! serverless H200 instances with snapshots are so nice for quick experiments

@haki_xer cards and border radius is a dead giveaway

Be honest, does this look vibe coded and if yes why?

@VictorTaelin you have to hold the model differently, they don't work the same

people who swear 4.7 > 4.6 (if anyone): what are you doing

opus 4.7 is a VERY GOOD model

@joshcrnls bc their target audience is not "designers"

can someone explain to me why everyone thinks an ai lab made up entirely of engineers is going to build a better design product than a company who has lived and breathed the needs of designers for a decade

@JakeKing @realsigridjin building @chaosgremlinlab here in AB, llm eval tooling and mech interp

Who's building devtools in 🇨🇦right now? want my algo to be filled with cool people building cool shit up here.

anyway. the anthropic team (@bcherny @trq212 and everyone else) are clearly working hard on this. just wanted to share concrete data instead of vibes. it's easy to criticize those in the arena, but please remember they are humans working their butts off for your benefit <3

in contrast to others i don't think the fix is more thinking tokens. tool selection and "thrash" detection are two places to start: make the model verify against actual source material before it starts claiming things, and detect when the model starts swirling and second-guessing

been data mining my claude code transcripts to figure out why conversations "feel worse" lately. pulled out 264 "tilt" incidents across 127 sessions, ~9k tool calls and this is what i found...

@kcosr @sdmat123 yes, this takes into account reading files through both the standard Read and Grep tools as well as cat, head, tail, diff, grep, git diff, git show, sed, etc. via Bash

@crabl @sdmat123 Does this account for files read using other tools like Bash or Exec (or whatever Anthropic calls them)? These can output file contents from shell commands like cat, grep, sed and diff.

Thariq, Firstly, huge respect to you and your team for the innovative work on Claude Code over the past year. But "mostly" doesn't cut it here. Yes, one of the metrics in the report is flawed due to your change hiding the thinking traces. That doesn't invalidate the other objective behavioral metrics showing severe degradation. This is a high profile user who put a huge amount of effort into communicating exactly what they were seeing with more rigor and precision than Anthropic provides regarding changes. Thousands of other users are shouting about their similar experiences. Anthropic has won a lot of love from developers. There are only two ways to retain that love when performance drops: 1) Fix the problem 2) Be transparent about necessary compromises/tradeoffs, including explaining what changed in detail If this level of transparency will alienate users, that requires some soul searching.

@ns123abc grok and gemini combo in opencode

What are we supposed to do now that anthropic completely nerfed claude?

@sdmat123 the single strongest predictor of my own frustration is the model editing files it hasn't read, i have quantitative proof of this