- Replying to @GregHBurnhamWrong 4/4 times with only the Thinking setting. Wrong only 2/4 times if I also append “Think really hard!”
- Replying to @zzdoublezzThis tweet was one scroll down from yoursloved that silly little quip bro. my favorite part was the deep sadness masked just faintly behind it
- Since seeing this, any time I pass a samoyed in the street I mutter to myself, “hypoallergenic, if you can believe it”Age-restricted adult content. This content might not be appropriate for people under 18 years old. To view this media, you’ll need to log in to X. Learn more
- Replying to @GregHBurnhamThis is the one GSM8K problem frontier models consistently struggle with, so it's a bit adversarially selected in that sense. I also think it's hard for humans! I got it wrong on my first try and... emailed the authors to point out the error. 🤦♂️ Source: platinum-bench.csail.mit.edu
- Replying to @peligrietzerComparative advantage in the home economy
- Pretty happy with how my predictions are holding up. 5/6 was the gold medal threshold this year. OAI's "experimental reasoning LLM" got that exactly, failing only to solve the one hard combinatorics problem, P6. My advice remains: look beyond the medal. Brief thread. 1/
- New from me: the USAMO problem that Gemini 2.5 Pro Deep Think newly solved happens to be one I wrote about a lot earlier this year. It's solution differs from the most common human solution in one interesting way. Excerpts here, link in reply
- Replying to @smithsmm and @lastpositivistAnother great story of Sendak understanding kids’ perspectives
- No urban chaos quite like the block in Queens bounded by 58th Ave, 58th Pl, 58th Rd, and 58th St.
- The 2025 AIME has fallen! On MathArena, GPT-5 (high) has solved the one problem that no prior model had solved, 2/4 times.
