We built a browser with no rendering engine.
Instead:
– GPT-5-mini “reads” the page 📝
– Google’s Nano Banana hallucinates it into an image 🍌
– You wait 20s to see the chaos ✨
It’s slow. It’s useless. It’s hilarious. Give it a try below!
My hunch is the the included agent features bundled with 16.3 are there for this exact purpose. If you’re building with Claude Code or Codex - with the right skills enabled no doubt they can follow these more complex patterns but for sure it’s making it harder and harder for the code to be readable and easy to human debug.
@rowlsmanthorpe The hard problem will be that half of the current problems that Burnham will be confronting become non issues in the world of ASI - whereas small issues now like AI sovereignty will become some of the most important issues that the U.K. has ever or will ever face.
This is such an under-reported issue right now and really glad you're diving into it more. I'm just completely gobsmacked at how unconcerned or unaware the mainstream media and politicians are of this upcoming tidal wave. This should be policy area number 1.
I'm interested if you've read https://t.co/JV2srHDDRN - and how a sovereign effort from the EU / aligned allies might play into this essentially winner takes all viewpoint that continually seems to be mostly accurate?
If GLM5.2 is genuinely almost frontier - which it seems to be - at what point do all of cyber protections on Opus and GPT 5.5 become kind of pointless? Or are these models not good at cyber stuff yet?
GLM 5.2 ranks #3 on FrontierSWE. It is only behind Fable 5 and Opus 4.8, and it outperforms GPT-5.5.
This is the first model that closes the large gap between models from Anthropic / OpenAI and other providers, and it is the strongest open-weight model by far.
@t_blom The question now is will OpenAI, Google and the fast follower Chinese labs do the same, or is it too big a competitive advantage to hold off releasing a new gen of frontier models
@utdreport@OptaAnalyst Loving how so many people don't understand that this is just based on if the scorelines of games finished as per xG not some weird pundit prediction. What it is showing is that we're not finishing our chances pretty much worse than anyone in the league.
@benhylak This is the point so many people are missing. Of course evals are useful for sense checking outputs. But A/B testing and intuition are separating the great products from the okay products.
We built a browser with no rendering engine.
Instead:
– GPT-5-mini “reads” the page 📝
– Google’s Nano Banana hallucinates it into an image 🍌
– You wait 20s to see the chaos ✨
It’s slow. It’s useless. It’s hilarious. Give it a try below!