Recent posts
Native video understanding in Claude and ChatGPT could be huge. It’s the only reason I use Gemini models right now, if I need video and it’s audio understood directly. It’s not that amazing at it though. What do you think? Should this be a priority and if so why isn’t it now?
My GPU just went brr and made a 15 second video with native audio. MiniMax H3 is imperfect, surprisingly capable, and already running on hardware far below a 5090. This is truly a wonderful AI summer for Open Source AI video peeps. Check it: https://youtu.be/pPWnWcWNPrA
Anyone else's windows defender absolutely lose it's mind and blast you with notifications every time you open codex merely because of the damn config.. drives me nuts
I just ran an open ended 3D dev prompt in codex and it worked for only 2 min and produced something not even worth sharing, meanwhile Opus 5 on the same prompt produced a very worthwhile spacial 3D slicing game
yo they actually quantized sol i'm not joking
i just realized after days of using 5.6 - haven’t heard “goblin” once, and I actually let 5.5 go ham with the term. It called something a “woodland creature” arguably a little funnier than “goblin” 😂
For 3D & Graphics Fable is still unparalleled. Really on a different level w the right prompt. But Open AI’s models feel more generalized in a good way, they’re better with random everyday chats, unique creative ideation yet still very much capable and fast as hell.
The same castle-building task was given to Fable 5, and the result is seriously impressive. The model produced a much larger and more complex 3D structure, with multiple towers, walls, gates, surrounding terrain, water, and a more consistent architectural style across the entire scene. This is a strong example of how capable Fable 5 is at planning and generating detailed 3D environments rather than just isolated objects.
Been using high, amazing for most but seeing the limitations in advanced coding demos. Haven’t really tried it on mid.
it's been 1 day now - what reasoning level is everyone using for GPT 5.6 Sol?
It’s insane to imagine AI without OpenAI or Anthropic. It’d basically just be Google, Microsoft and Meta, the same old giants. Instead OpenAI came crashing through the wall like the Kool Aid Man, Anthropic is barely trading blows, and somehow Google is the one left in the dust?!
It was not too long ago that Open AI, after the realization they spread themselves too thin, did away with Sora and refocused to deliver new models. Fast forward to the release of Sol, and they seem to be on top with an efficient and potent release. Should Google do the same?
thinking gpt 5.6 will feel abt as potent and capable as Fable 5, will still feel different to use and Fable may dominate in niche programming areas, the jump in quality should be similar just in that open ai "flavor" alongside faster tok/s output
Time to say goodbye, friends. Fable 5 ends on 11:59:59pm PT on 7/7. But with every sorrow, hope also begins to emerge. Fable 5 disappears, and GPT-5.6 rises anew.
inventing new ways to hit penjamin at work I see... wait $300?!
voice-to-text has gone too far
I honestly think it’s more on us as the people to use the tools for not reaching that stage. We aren’t always as creative as we give ourselves credit for, I do think there is a lot of potential already sitting with current models, untapped. Maybe there’s no need to rush, though.
LLMs were predicted to be making new, novel discoveries and inventions by now. Since we are now at Fable-class models, has this actually happened? What are the best examples (which I may be unaware of) where LLM agents have discovered or developed truly new ideas or solutions?
Seedance 2.5 looks so far ahead in sheer output quality, and actually making strides in generation length & resolution. But it will be: - Expensive - Slow to generate I think we have hit a point in AI video where certain scenes, certain ideas must be delegated to lesser models.
Lil rumor roundup to kick off the week! https://www.youtube.com/watch?v=aCJsEpci7uw