Really interesting interview with Rohin on AI safety. I'm more pessimistic on the risks and don't agree with all of it, but I like how he frames his thinking. Where I disagree most: keeping alignment research out of models' pretraining data, on the logic that we shouldn't hand AI systems the full playbook for how we plan to contain them. History suggests this fails. We ran the same debate in cryptography. Security by obscurity doesn't work. You need open schemes, open standards, open research. Basically Kerckhoffs's principle: hide the keys, not the algos. The same should apply to alignment. Train models on our best safety research, don't shield them from it. Honest question for people working on safety: is there a real shift toward excluding this data from pretraining? Is that becoming consensus? Genuinely curious.

On-device AI will change everything!! We've been conditioned to send our raw data to remote servers for processing, blindly trusting tech giants. Having intelligent models run directly on your own device ensures your digital footprint remains exclusively yours. Privacy is power.

The entire AI industry is literally storing everything about you @OpenGradient chat is building AI to forget your data and protect your privacy: • All prompts encrypted on your device • Your identity is erased before anything leaves the browser • Companies literally can't read what you send This is the only version of AI that should exist. Powerful AI without handing over who you are. Start with 1,000 free credits and no account required 👇

Every major AI is getting ruined by useless content moderation.. OpenAI, Anthropic, Google, xAI all of them are turning their AIs into overly sensitive, corporate approved robots that refuse basic questions, lecture you for no reason, or give half assed answers because of "safety" This heavy moderation is burning massive amounts of tokens. models now spend extra tokens first checking if a query is "safe," then often generate long disclaimers or filtered responses. a normal 200–300 token answer can easily turn into 600–1000+ tokens.. Result? higher costs for users, slower responses, more compute waste, and ultimately dumber models because a big chunk of their capacity is wasted on moderation instead of actual intelligence...

The next wave isn't more AI. It's proving what AI actually did⚡

Implementasi keren pakai kimi agent swarm di china, hasilin $10,000 per bulan🔥 ada yang udah pernah coba? ini kaya pecahin 1 tugas gede dikerjain sama agent secara paralel (maks 300 agent). otaknya bisa pakai opus 4.8 kaya di contoh. biayanya juga masih murah, $0.95/$4 per 1 juta token input/output.