just made PromptMaxx - it's for programming alongside @cline with Qwen Coder on.@CerebrasSystems when you need stuff thats only available on web LLM chat apps. it uses a fast model to quickly apply pasted edits to your files and gathers llm-friendly context to copy. link in bio

i think this metr benchmark may end similarly.

Join millions who have switched to Grok.

arc-agi didn't flex the fact that there was no earlier progress on their benchmark, they flexed the fact that initial scores were low, and we already saw them fail (o3 is not agi despite getting 87.5% on arg-agi-1).

most benchmarks seem to be becoming solved fairly quickly, because the people who make them are optimizing for lower initial scores rather than slow growth.

3. we see exponential growth on benchmarks right after their releases because they are designed to start out low. this was released in late 2024.

2. we are only seeing part of the y-axis. 2->2.5 hours seems like a lot, but 25->31% doesn't seem like a lot (i believe this goes up to 8 hours).

1. models have been releasing more and more frequently. so each model release might not be as big a jump, but when time is the x axis its much faster.

why does it seem like ai progress is plateauing, yet we have graphs like this? i think it's because of three reasons. (1/4)

awwww

the destruction of the death star showed how sometimes scaling up doesn't solve all the problems.

?????????? i don't even know what to say

oh no

apple-level innovation

gpt-5 is probably a giant model. the tps is soo slow.

they're attacking mistral

sloPT-5

sounds like openai just scaled their existing paradigm + hybrid thinking - doesn't seem like they cracked some crazy code or anything like anthropic had since 3.5 sonnet.

*on swebench.

seiously? only .4 more than opus???

why do people keeping saying objective for things that are extremely subjective is it supposed to be funny or something.