OpenAI's "AI SYSTEMS" and New Scientific Discoveries

Had to check that I didn't accidently hit 4 x speed on the video

That other Asian dude is some how fast forwarded whilst everyone else is at normal speed 😂

im so early that ai is actually producing good output still before it gets nerfed

Well o3 made a little bouncing ball more reliably than Gemini 2.5 Having said that, I'm still continuing a project in Gemini and will test out o3 in parallel. It does seem pretty impressive!

> Are they back? Yes , as a product company. they’re differentiating by application development rather than sheer model smarts. I foresee them pumping $$$$ into tool use too cuz that’s gonna help their products. I’m glad to see it, cause the usual IQ benchmarks are saturated, now it remains to tackle tool use and computer use.

LiveBench coding eval: Gemini 2.5 Pro peaked at 58, now o4-mini-high hits 74.33. That’s a massive leap in real-world coding performance — seriously next-level.

As a simpleton barely hanging on. I appreciate you incorporating their video into yours. I don’t have to go find it myself now. Thank you!

Math and coding scores show it can solve both math problems and algorithmic challenges — that means serious potential to accelerate OpenAI’s real-world AI research!

o4 sounds like a dream to build agentic AI applications on

actually, resolved my n8n-docker-nas-pinggy issue with a very simple and elegant solution. I spent 2 days making it work, and tried different models. this one turned out to be the most useful