I started letting Claude ask me questions, instead of spending too much time planning myself. This works much better for me: I want to build something, I ask Claude to ask me questions about it, Claude writes the full plan for me, I review and update the plan.
✅ Safe Reply
The interview-the-AI approach is underrated. Most people dump context into a prompt and hope. Flipping it so the model interrogates you first forces rigour you'd skip otherwise.
Introducing GPT-Rosalind, our frontier reasoning model built to support research across biology, drug discovery, and translational medicine.
✅ Safe Reply
Frontier models specialised for scientific domains is where the real leverage lives. General-purpose LLMs are table stakes now. Domain-specific reasoning is the next moat.
Drug discovery timelines are measured in decades and billions. If AI can shave even 2 years off that cycle, the economics of the entire industry shift. This is the boring, world-changing use case nobody on Crypto Twitter will notice.
I am happy everyone is switching to Codex, but Tibo if you start rate limiting me or making me use worse models...
✅ Safe Reply
The real signal here is Sam Altman publicly treating competitor products as infrastructure his own relies on. The AI stack is becoming interdependent fast.
When the CEO of OpenAI is worried about rate limits on someone else's product, you know the coding agent wars are properly heated. Nobody has a monopoly on dev velocity yet.
Seasteading is already here. It's just not evenly distributed. [QT about floating data centers that drive themselves out to sea, capture water to spin turbines and power GPUs]
✅ Safe Reply
Floating, self-powered GPU clusters solve the three bottlenecks every AI company hits eventually: energy, cooling, and jurisdiction. Clever engineering.
Everyone arguing about AI regulation while the data centres literally sail into international waters. The future was always going to be weirder than the policy papers.
Motivational platitudes are cheap. The real version: ship something publicly that could fail. That builds more confidence than any mindset podcast ever will.
The AI revolution. Historians will mark the 2020s as the moment intelligence itself became programmable. Generative models, autonomous systems, and agents exploded into daily life, rewriting work, creativity, science, and power structures faster than any prior tech shift.
✅ Safe Reply
The framing is right even if the tone is grandiose. Programmable intelligence is the substrate shift, not the applications on top of it. The infrastructure layer is where the real bets belong.
Every decade claims it changed everything. This one actually might, but not because of the things people are currently excited about. The boring plumbing of agent orchestration will matter more than any chatbot.
The kind people at eth_limo have warned me that there has been an attack on their DNS registrar. So please do not visit [eth.limo pages] until they confirm that things are back to normal.
✅ Safe Reply
DNS attacks remain the quiet killer of decentralised infrastructure. Your smart contracts might be immutable but your registrar is one social-engineering call away from compromise.
The irony of decentralised systems being brought down by the most centralised failure point possible. DNS is the Achilles heel nobody wants to talk about.
Vibe coding is more addictive than any video game ever made (if you know what you want to build).
✅ Safe Reply
The flow state of building with AI assistance is genuinely different to traditional coding. Faster feedback loops create a compulsion loop that traditional dev work rarely matched.
Vibe coding is crack for people who always wanted to build but hated the syntax. The dangerous part is shipping fast enough to feel productive while skipping the architecture that makes it maintainable.
People are vibe coding their own tooling. [QT about people creating their own game editors instead of prompting for perfect assets]
✅ Safe Reply
The meta-shift from 'prompt for the output' to 'prompt for the tool that makes the output' is real. Building editors and workflows beats one-shot generation every time.