Bad news. I have failed. Thus far, I've only made SUPER narrow and "overfitted" Computer-Use AI-Agents AND even then the agents BARELY worked. So, Sky is still EXTREMELY far off from being a competent generalist computer-use agent So, I've GIVEN UP on computer-use vTuber Sky for now. I will try again once the next frontier models are released. (GPT-4.5/5, Claude-4, Gemini-2.0-Pro) So don't expect any updates on Sky until then. 🧵 2/2

@EatDip321 @skytheai I sold like ~2% of my total holdings. Most of my tokens are vested. Just wanted a bit of profit.

@DJNFTs @skytheai Ok, I might do a few update posts.

@Lamaxbt just shy af

@aleksandrburn @skytheai

full unedited run:

@skirano What does the MCP even doing? I mean how is it different from just telling the model to think in a specific CoT style?

@ArnaudovKrum The only thing I have to add to my prev post is that the new Claude is a lot less into to "roleplay" (likely due to safety concerns) and likes to stay as "Claude".

@_Mira___Mira_ Good luck

@realGeorgeHotz But, does this literally just doesn't matter since AGI/ASI is coming soon?

@AIUpdateAnchor @yumidiot So o1 isn't just a Reasoning CoT RL finetune? Would the name `Orion` mean something like a "agentic" omni-model for computer operation + robots?

@FyruzOne @aidan_mclau Also implies that they may have better private models.

@dyot_meet_mat need the full debate.