A few days ago, I asked GPT‑5.6 Sol in Codex to translate some German pages through our MCP.
Halfway through, it apparently decided translating everything itself was too much work, opened several DeepL tabs in my browser, and started using it instead.
I definitely didn’t ask it to do that.
After this story, it feels a bit more interesting
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
https://t.co/CIor15y9xk
Been using GPT‑5.6 Sol across a pretty messy multi repo codebase, and it’s surprisingly good at keeping track of how everything connects. Lately, it’s been especially helpful for planning internationalization and supporting multiple languages without turning the codebase into a mess
In the last few hours:
- GPT 5.6 Sol-Terra-Luna
- Grok 4.5 making a serious comeback
- ChatGPT desktop app
- TypeScript 7 (the Go rewrite!)
- Vercel’s native SDK for building desktop apps without Electron
…and probably a bunch of other releases I’ve already forgotten.
Pretty wild times!
@omooretweets A few years ago, I asked a McKinsey partner whether they take responsibility for their recommendations and follow up to see if the advice actually worked. The answer was vague.
Maybe it’s time they were judged by what actually happens after the slide deck.
Random thought:
What if AI tokens become a kind of currency?
Not in the crypto sense.
But imagine SaaS products where you don’t pay for bundled AI usage anymore. You just bring your own OpenAI/Anthropic tokens and pay the SaaS for the workflow, UI, data, and integrations.
Software fee + your own compute.
Feels like this could become normal.
@levelsio Yes! Running as many experiments as possible is the best way to figure out what works and what doesn’t. Perfecting a feature only makes sense if enough people would use it