Pove (tokyo)
5.3K posts
@craigzLiszt as a developer i don't think benchmarks are valid (especially if you know what they are)
English
ai labs are being very sketchy with how they’re advertising and benchmarking their models
English
@craigzLiszt i like when it speaks well and knows what i'm saying and doesnt make bad assumptions. accuracy and humanity.
English
@midgirlfriend i think of you as an outdoors creature, not allowed inside
English
no, it's just curiosity, the same as academia. curiosity + rigor = explanation.
propagandists call independent explorations conspiracy theory, because they deem the state and commercial institutions as the *only necessary and qualified authors of truth*.
i'd say your claim is backwards.
-those who accept institutional dogma without question do so to make the world to make sense
-those who investigate the world in its ambiguity trade certainty for clarity
English
i think people are drawn to conspiracy theories because that way they get to make sense of a world that just doesn't make sense
English
Pove (tokyo) retweeté
Pove (tokyo) retweeté
@k1rallik i noticed this and it really crushes down development. it feels like it did 6+ months ago, when claude was buggy all the time.
it's kinda sad, how we had a good thing, and now it's gone.
is there any way to have a good chatbot again?
English
🚨do you understand what just happened with Claude..
Anthropic quietly cut Claude's thinking depth by 67%.. didn't announce it.. didn't explain it.. an AMD AI Director had to dig through session logs just to prove it happened.
median reasoning dropped from ~2,200 to ~600 characters.. API calls went up 80x.. meaning Claude thinks less, fails more, retries more. and YOU burn more tokens paying for those retries.
they added a header that hides Claude's thinking from your logs. so when the model analyzed itself it found blank pages. and concluded it had stopped thinking.. you're paying $200/month for a model that can't read its own diary..
the thinking didn't disappear.. it just became invisible.. and Anthropic said nothing until the numbers went public
ℏεsam@Hesamation
AMD Senior AI Director confirms Claude has been nerfed. She analyzed Claude's session logs from Janurary to March: > median thinking dropped from ~2,200 to ~600 chars > API requests went up 80x from Feb to Mar. less thinking and failed attempts meaning more retries, burning more tokens, and spending more on tokens > reads-per-edit dropped from 6.6x → 2.0x. model stops researching code before touching it. > model tried to bail out or ask "should i continue" 173 times in 17 days (0 times before March 8). > self-contradiction in reasoning ("oh wait, actually...") tripled. > conventions like CLAUDE.md get ignored because there's less thinking budget to cross-check edits > 5pm and 7pm PST are the worst hours, late night is significantly better. this means the thinking allocation is most likely GPU-load-sensitive.
English