past statements: „Google has the mandate” → “Google is not a serious company” → “Gemini is cooked” → “Google is in free fall” → “not sad, but deserved.”
Then the “Anthropic is mogging OpenAI” and OpenAI is “so fucking far behind” propaganda for weeks, mind you… all of it before Mythos/Fable was available and many engineers (including me) preferring codex gpt 5.5 over slopus 4.8.
He is an exceptional AI reporter at times, but also very, very emotional
@FakePsyho Ahhhh sweet old AlphaZero mogs AlphaGo flashbacks are kicking in again. In similar energy the model of this year is mogging the result of last year…
@omni1896837@Lentils80 I fear a world where the speed is so fast and intelligence so stagnant, that Im CONSTANTLY doing work, since any delegation gets handled easily… basically all my threads woulds would wait for me all the time and the context switching gets me into an early grave. 💀
From what I've seen, GPT-5.6-sol will be comparable or stronger than Claude Fable 5 in agentic capabilities specifically, while being much cheaper
What it will NOT be is smarter or more capable generally than Fable 5, and the jump from GPT-5.5 to 5.6 will not be as big as the jump from Opus to Fable
@Lentils80 If that’s accurate, it’s a little disappointing. At the GPT-5.5 launch, OpenAI chief scientist Jakub Pachocki predicted “pretty significant” short-term improvements, “extremely significant” medium-term improvements, and said the pace of capability gains should keep increasing…
I gave Fable my project from last year and it just knocked it out of the park.
Creating a FF7-style game world with gpt-image-2, detecting a navmesh, pre-fetching connecting routes on the overall map, and at the edge of the map it's building new scenes on the fly. 🤯
@_sholtodouglas Memory needs a lot more attention. We need to move on from interactions that live & die in scoped chats. I wonder if we really can only get there by solving continual learning…
@_sholtodouglas Sholto, like this post if Fable to cheap to meter is on roadmap. We can tell it’s amazing, but on MAX it burned like 150€ for one tasks for me…