Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
@danellisona Buy land in Iceland. Drill a hole to get free electricity. Buy some hardware to run decent llm models. Use it to provide local inference for Icelanders... Shocking how dependent we are on external intelligence.
@DylanFeltus Yes we probably are, but it's also hard to tell how different this period in history is. Wish X had a time-travel feature to experience Twitter from a specific date... That being said, this acceleration feels different...
Opus 5.5 is almost too good... In few more generations, where is this heading? When managers realize they don't need employees, and employees realize they don't need managers anymore.
@DavidOndrej1 Strange... I had the opposite reaction when I tested it yesterday. Kept producing horrible UI and buggy mess.
Gonna keep testing, maybe try it in Antimatter...
@DavidOndrej1 GPT-5-Codex has blown my mind multiple times. One-shotting large complicated refactors. Very impressive!
However, I rarely use it as a daily driver due to speed and flexibility.
Claude seems to breeze through simpler tasks faster.
@omarsar0 Feel like I'm jumping between Codex IDE extension and Cursor with Claude 4.5 Sonnet & Haiku. Haiku has a real surprise due to the speed.
I agree with some of the other comments here: Large refactors or UI design = Codex (give it plenty context). Most other work = Claude.
3 addresses aren’t centralized or less secure. They’re P2SH (script-based), often used for multisig or as SegWit wrappers. The format itself is neutral: it’s only “centralized” if you hand control to an exchange or custodian, not because of the address. Cryptographically, 3 uses the same Bitcoin security as 1, and with multisig can be more secure than legacy 1 addresses.
Great way to put it. What a lot of people misunderstand about term is the "vibe" part of it. Like it's all rainbow and sunshine. But if you're doing it right, it's huge mental effort. Long term planning, short term execution and a lot of learning. Most people leave out the last part. That's where you get apps with API keys in the front end, or vastly inefficient tech stack.