Cheap FLOPs

@elliotarledge @anneouyang Nice! Well done.

kernelbench.com is live. can frontier models write fast triton/cuda/cutlass/cute-dsl/ptx without cheating? i add kernelbench-hard to the existing kernelbench-v3 (which was built on top of KernelBench from @anneouyang et al.) hard mode has: fp8 gemm, topk, sonic MoE fwd, KimiDeltaAttention, paged attention decode, kahan softmax, w4a16 gemm. all of these require deep understanding of the sm120 architecture (benchmarking happens on my local rtx pro 6000 blackwell setup). gpt 5.5 xhigh and claude opus 4.7 max clear took the W here, but i was honestly surprised with kimi-k2.6 and deepseek v4 pro. this is just the first public iteration. im open to constructive criticism (dm works)

@iScienceLuvr Cloudflare tunnel to a web app running on your computer with xterm.js

Claude Code has remote control which is amazing but has low rate limits. Codex has high rate limits but no remote control. How do I get the best of both worlds???

@scaling01 They always do this: deny, say it's a docs bug and then it happens like a week later

deleted both posts that went viral

@whatRanjuSaid @icanvardar "Most rats love rat traps"

@icanvardar most software engineers love anthropic(claude)

anthropic hates software engineers

@Hesamation Yes it's garbage, ChatGPT's is really annoying as well.. it used to be unlimited now it cuts you off after a few seconds and says you've gone over your limit but they'll still do it for you if you click again, then it works

let’s take a moment and appreciate how Claude’s built-in speech to text is ABSOLUTELY CRAP. > you speak 2 minutes > writes the first 20 words > cuts off without telling you > has a ton of typos, misses spaces

@koylanai They are just pretending to have made a mistake so people think it's fixed.. and that they aren't purposefully screwing over the users. 'Whoops- we changed the system prompt to make a bunch more money!'

The frustrating part is that the Claude Code team, along with people deep in AI psychosis, have been gaslighting anyone who raises concerns about Claude Code's recent issues. "your reasoning setting is wrong" "oh that benchmark is wrong" "we checked our code, nothing is wrong" "skill issues" "Hate" is a strong word, but when you're paying a lot of money for a product and it actually makes your job harder, to the point where people make you start questioning the quality of your own work, it really becomes a problem. I'm glad they identified the issue, and I genuinely want them to succeed, along with Cursor, Codex, OpenCode, and others, we need more alternatives, but we also absolutely need open benchmarks. anthropic.com/engineering/ap…

@elliotarledge Tldr: they're claiming to be morons so people don't suspect malice

here it is

@dedene @AnthropicAI I'm not convinced they really fixed the root issue still.. given all their recent pricing shenanigans I think they're engaging in subterfuge on multiple levels

@cheapflops @AnthropicAI Yeah, my weekly limit was due for tonight 21h anyway. This single reset doesn't give me 6 weeks of poor usage back.

Did @AnthropicAI just reset everybody's rate limits?

@badlogicgames Lmao amateur hour @ Anthropic

recommended reading. cool they are fixing things. but it's also a reason i switched away from CC. no control over the harness means having to wait for them to fix things. the model didn't change. the harness did.

@johntitorIII @bcherny They are - don't fall for Boris' gaslighting

@bcherny Boris, eu amo Claude Code. Mas cara.. eu to sentindo tanto o 4.7 e 4.6 100% lobotomizados. 😭

We’ve been looking into recent reports around Claude Code quality issues, and just published a post-mortem on what we found.

@bcherny In his next tweet in the thread goes on to say the main thing people complained about is still an issue

We take these reports incredibly seriously. In my time on the team, this has probably been the most complex investigation we’ve had. The root causes were not obvious, and there were many confounders.

@bcherny Haha - I thought you "just fixed it"?

Separately, we’ve also heard reports of issues with Opus 4.7 in Claude Code. The team is working on those and we’ll share more as we roll out improvements over the coming days.

@leeloumt @bcherny @DanielleFong And doubled price in A/B tests etc etc ... pattern of abuse

@bcherny @DanielleFong You revoked pro users from Claude Code… seriously?

@bcherny "Thanks for your 200 dollars"

We’re resetting usage limits for subscribers. Thank you so much for your feedback and patience!

@bcherny No you can keep your dogfood, I'm out

@bcherny You really don't lol.. you just gaslight people and then when the heat is on you retweet a bunch of BS to hide from folks

@dtfskinner @ClaudeDevs "We got caught throttling you and we fixed 3 things" (but it still sucks)

@ClaudeDevs You could’ve just said: “We got caught throttling you all”

Over the past month, some of you reported Claude Code's quality had slipped. We investigated, and published a post-mortem on the three issues we found. All are fixed in v2.1.116+ and we’ve reset usage limits for all subscribers.

@maxi_moxa @ClaudeDevs Fancy? Sounds disgusting. I don't want Dario's dogfood

@ClaudeDevs dogfooding is just a fancy way of saying we used it once and it didn't explode immediately

@ClaudeDevs I'm out - keep your dogfood

We’re making changes to catch these types of issues earlier, including more internal dogfooding with configs that exactly match those of our users and creating a broader set of evals and running them against isolated system prompt changes