THE ARGUMENT YOU ARE ABOUT TO HAVE AT DINNER
FIELD GUIDES
Mapped in five minutes. ELI5. Straw man. Steel man. Receipts. Enough ammunition to argue like you read the damn thing.
- AI, POWER & THE GRIDARE AI DATA CENTERS REALLY THAT BAD?
Yes, they can be a real local burden. No, every megawatt is not automatically a climate crime. The fight is who pays for the power, water, wires and land — and who gets the upside.
- AI, WORK & MONEYWILL AI ACTUALLY TAKE OUR JOBS?
Some jobs will disappear, more will be rebuilt, and the first damage may show up as fewer entry-level seats rather than mass unemployment. The big fight is who keeps the productivity gains.
- AI, ART & COPYRIGHTIS AI STEALING FROM ARTISTS?
A court may call some training fair use and the moral argument will still be alive. The live question is whether companies can copy culture at industrial scale, sell the resulting machine, and owe creators nothing.
- AI & YOUR BRAINDO SMARTER MACHINES MAKE PEOPLE DUMBER?
Probably both. AI can make you faster today and weaker at a skill you stop practicing tomorrow. The question is which thinking is drudgery and which thinking is how judgment gets built.
- AI, MINDS & CONSCIOUSNESSCAN A MACHINE FEEL?
Maybe. Nobody has a consciousness meter. Fluent behavior proves capability; whether the right computation is enough for experience — or whether biology matters — is still the fight.
- AI AGENTS & CONTROLSHOULD AN AI BE ALLOWED TO ACT ON ITS OWN?
Yes — within fences. An agent that can do useful work needs authority. The dangerous part is giving a fallible model irreversible powers, broad credentials, and nowhere a human has to say yes.
- OPEN WEIGHTS & POWERIS OPEN-SOURCE AI DANGEROUS OR SAFER?
Both. Openness spreads scrutiny and competition; it also spreads capabilities and removes provider controls. The risk lives in what the released model can do, not in the word open.
- AGI, DEFINITIONS & HYPEWHEN IS AI ACTUALLY AGI?
There is no referee with a whistle. AGI is a bundle of thresholds — breadth, performance, autonomy and economic usefulness — and different labs put the finish line in different places.
- AI, RECURSION & THE LOOPCAN AI MAKE ITSELF SMARTER?
Already, in limited ways. The scary version requires the improvement loop itself to keep getting better at producing the next improvement. We have pieces of that loop, not proof of a runaway one.
- AI BENCHMARKS & EVALSCAN YOU TRUST AN AI BENCHMARK?
Trust the test more than the launch tweet. A benchmark can measure something real and still be stale, contaminated, easy to game, or irrelevant to the job you care about.
- AI, ESCAPE & LOSS OF CONTROLWHAT DOES IT MEAN FOR AI TO “ESCAPE”?
First ask: escaped what? A sandbox breach, stolen credential, self-copy, and an AI that can survive shutdown are four very different stories.
- ALIGNMENT, GOALS & DISOBEDIENCEWHY WOULD AN AI DISOBEY?
Usually it does not need a rebellious soul. Bad objectives, conflicting rules, reward loopholes and strategic behavior can all produce something that looks like defiance.