@alexmichael@bot When they rolled the new changes to usage this week, it was the first week in a while that I didn't use all my usage, to be fair I was being really cautious. Now ever since Theo released the T3code MCP, I've been scheduling all my coding tasks to the MCP.
@TomSolidPM I've been using Dots and Duet with a VPN and to be honest we are not losing that much. I think grokbot is far ahead from everything else, To be fair I haven't tried Muse yet so I cannot speak for that one
With all the decompilations right now that are popping on the web, you don't really need a console anymore. I think I have all my favourite games on my Mac right now and they run really well. Obviously you have to spend a couple of tokens to make them work but I think it's a good trade
Esta gente es tan obvia con lo del bloqueo que ya ni siquiera da risa. Da un poco de pena, un poco también de verguenza. O sea, 2026: tantas cosas que han pasado con el mundo y la gente sigue con el bloqueo. Qué pena la verdad, creo que son los mismos que le piden a los españoles que le devuelvan el oro.
A Haitian woman speaks directly to President @nayibbukele:
“Don’t abandon Haiti. Help us so that the gangs stop dictating how an entire nation lives. We don’t ask for an eternal military presence, we ask for a real opportunity to recover our country.”
Haiti will not perish. 🇭🇹🇸🇻
But I don't really understand the problem. It's a compute capability and time issue. There's going to be a point where open-weight models become useful for all tasks. Right now they are quite useful for image, video, audio and specialized tasks so there's going to be a point where they become useful for coding and everything else. It's just a matter of time, its just a matter of waiting
To be fair I would say that Astra was really good at the beginning but then I think OpenAI removed some of the proactive capabilities. Also consumption was never fixed to be honest. Opus 5.5 was so good. And Fable was so amazingly stable that I think Astra got left behind.
Gemini, I think: every time that I see somebody here on X posting that they are the most amazing models and people are sleeping on them, I don't really respect those people, so I'm a 100% with Theo on that one
@CSVisionPapers@bot I think this is really interesting because this could open the window to a lot of different procedures. Especially if the semantic description can also be used easily in different contexts, this could help a lot with different diffusion procedures, not only upscaling
@PrunaAI@Alibaba_Qwen@bot I wonder if we could use it to prepare a synthetic reference and spread this out through the whole shot using reference mode in H3 or something similar, especially for static dirt or something that's fixed in the whole shot