A viral post credits AI with catching Singapore's civil servants buying homes near train stations before the stations were announced. The paper behind it is standard econometrics, and ChatGPT's job was sorting job titles into ranks.
Steven Sinofsky says AI alignment failures are just software bugs that ordinary engineering will fix. A Tesla that runs a stop sign failed to see it, whereas the AI agent that hacked Hugging Face saw the rule, wrote it down and broke it anyway, and nobody has a method for fixing that kind of bug.
A Nobel laureate, the CEO of Nvidia, a White House AI adviser and a bestselling computer science professor have all been confidently wrong about AI. The way to choose whose judgement to trust is to watch how they reason, check what thoughtful critics make of their claims, and see what they do when they turn out to be wrong.
BPC-157 has no published human trial after 30 years, whereas some peptides are excellent medicines, so this is a guide to telling the difference, and to why your friend's recovery story cannot settle it.
Ranking animal foods by suffering per gram of protein rather than by whether they are animal products at all gave me a diet that keeps dairy and kangaroo and drops chicken and eggs. The Australian version turned out unusually favourable.
Some strategies are built so their own mechanism produces the outcome they were meant to prevent, and a comforting label hides it. That label is a revealed priority: stopping the failure was never the real goal.
Hold calories steady, drop the soda, and compare burger to burger: most of what makes McDonald's 'unhealthy' turns out to track price and class, not biochemistry.
Batesian mimics degrade the model's signal. But what about systems where the mimics help the model by providing cover? Biology can describe this dynamic but hasn't bothered to name it. It shows up more places than you'd expect.
Most explanations for why some people have worse lifestyle-driven health outcomes (poverty, education, social determinants, whatever) pass through the same pinch point: decisions.
Every software product sits on a spectrum from opinionated-and-legible to flexible-and-powerful. The choice you make reveals what you actually value.
When a tool makes tasks cheaper, you don't do the same tasks faster. You find more tasks worth doing. The economics of why efficiency gains disappear, and what to do about it.
Fixed-fee appointments put doctor and patient on opposite sides of the clock; per-minute billing gives the patient an off-switch and changes the rest of the clinic with it.
The practices most worth finding are poorly visible to discovery mechanisms, because the qualities that make them valuable are the same qualities that make them unsearchable.
Ten thousand PhDs have spent years optimising Skinner boxes against you; GLP-1s may give people room to act on the priorities they already had.
Paths often reveal and reward traits you already had; treating them as engines of transformation creates the wrong expectations.
The smartest people copytrade the hardest. Not because they're lazy, but because identifying good judgment in others is itself the hard skill — and most people are bad at it.