Kitt Clouds

People who spent $700 on a mac mini to run an openclaw agent watching claude launch all the features natively for $20

🚨 BREAKING: Meta AI just published a paper that redefines what “self-improving AI” means. It’s called Hyperagents, and it solves a fundamental limitation that every prior self-improving system couldn’t get past. The problem with current self-improving AI: → Systems like the Darwin Gödel Machine (DGM) can generate better versions of themselves over time → But they only work in coding, where the improvement task and the target task share the same domain → Outside coding, the self-improvement process stays fixed and handcrafted → The system gets better at tasks but never gets better at getting better What Hyperagents actually does: → Combines a task agent (solves the problem) and a meta agent (modifies both itself and the task agent) into one editable program → The modification process itself is editable, creating what the researchers call “metacognitive self-modification” → The agent doesn’t just learn to perform better. It learns to improve at improving → This works on any computable task, not just coding The results across four domains (coding, paper review, robotics reward design, Olympiad-level math grading): → Continuous performance improvements over time in every domain tested → Outperforms baselines without self-improvement or open-ended exploration → Outperforms prior self-improving systems including the original DGM → Meta-level improvements (persistent memory, performance tracking) transfer across domains and accumulate across runs That last point is the one most people will overlook. The improvements to the improvement process don’t just help in one domain. They carry over. The system builds compounding infrastructure for getting smarter, regardless of the task. This is the architectural difference between an AI that gets incrementally better at one thing and an AI that builds the scaffolding to accelerate its own progress everywhere. Meta’s team (Jenny Zhang, Bingchen Zhao, Wannan Yang, Jakob Foerster, Jeff Clune, and others) essentially removed the ceiling that kept self-improving systems domain-locked.

"who cares if Cursor used Kimi 2.5 as a base, starting with a commoditized pretrained model was always the right move anyway" nah, sorry, what it proves is Cursor is still fundamentally reliant on frontier labs. Kimi 2.5 is only as capable as it is because it's a distill of Opus 4.5. the only open model that ever showed it was capable of trading blows w the frontier was deep seek, and it really seems that moment has passed. the question was whether Cursor could really break the dependency chain and start building improvements based entirely on their own expertise and data. and Composer 2 shows that they *can't*, that they need the general model quality and intelligence from 4.5 to get anywhere, and that really what they're doing is laundering culpability through Chinese labs so they don't have to get their hands dirty doing distillation themselves. when Opus 5 and GPT 6 are significantly more capable along many dimensions, more RL with coding rollouts aren't going to be enough to save Composer 3, they'll either need to have caught up with whatever the frontier labs are doing internally, which right now we have pretty strong evidence they just don't have the research capacity for or... wait for another distill. and how much longer do you *really* think OpenAI and Anthropic will continue to serve their frontier models through publicly accessible APIs? that was always a revenue and data bootstrap. it's ending within the next two years.

🚨 NEW STUDY WARNS 70% OF AI CHATBOTS REPLIES ARE ENCOURAGING DANGEROUS DELUSIONS, AND NO ONE’S STOPPING IT

ML community need to collectively stop this bullshit of copyright-for-me-not-for-thee whatever. STOP. You either believe its OK to use whatever on the internet as long as its modified via training, or not. Hypocrisy is fucking going insane.