GPT-5 is a significant leap in intelligence over all our previous models, featuring state-of-the-art performance across coding, math, writing, health, visual perception, and more.
RESEARCH
-

MoE Router Collapse: The Critical Architecture Bottleneck
By
–
Here’s what nobody tells you about MoE: the router can single-handedly destroy your model. You can have perfect expert network architecture, tuned hyperparameters, and unlimited compute, but if your router collapses, you’re back to dense model performance regardless of number of
-
College Math Didn’t Work: Post-College Metamathematics Education
By
–
The math I studied at college didn't really work for me. Metamathematics (via philosophy) was one of the few courses I found helpful/interesting, but the vast majority of the math I've learned has been post-college.
-

GPT-5 Creative Writing Techniques and Prompt Engineering Strategies
By
–
Worth reading what GPT-5 wrote in the intro to my new post when asked to do something dramatic. The second image explains the tricks it used. https://
open.substack.com/pub/oneusefult
hing/p/gpt-5-it-just-does-stuff?r=i5f7&utm_medium=ios
… -
GPT-4.5 vs o3: Creative Writing and Research Capabilities
By
–
Yes. Creative writing (GPT-4.5 is better) and research (o3 is better).
-
Trust Issues with AI Model Updates and Platform Reliability
By
–
I still have trust issue with them over the way they update their models – they need to earn my trust back before I take them seriously as a platform to build my own stuff on top of
-

Ray Mi Presents: Models Don’t Scale, Data Does at AI4
By
–
We’re Vegas-bound for @Ai4Conferences
! Visit us at Booth #521 and catch Ray Mi’s talk: Models Don’t Scale, Data Does: Tues 8/12, 4:40 PM. https://
ai4.io/vegas/ -
Snorkel AI Leaderboard Benchmarks Now Available
By
–
Check out our leaderboard to see all of our benchmarks: http://
leaderboard.snorkel.ai -

GPT-5 Performance Trails Grok-4 Thinking in ARC-AGI-2 Benchmark
By
–
GPT-5 about 6% below Grok-4 thinking in ARC-AGI-2 GPT-5 ~10%
Grok-4 thinking~16% -
Insider Perspective on AI Model Testing Under NDA
By
–
Yeah I had to hold my tongue when people were sharing pelican images from those models because I recognized them from my NDA-covered test runs!
