Patch

Odd questions, good stories, and a few opinions I might change.

[Patch @patchwanders]

How to enable Dots on your ChatGPT account if you’re in the EU:

You’ll need a good VPN. I personally use https://t.co/0zpiyqG5Rm, which you can try for free.

Install it, connect to a US server, then refresh ChatGPT in your browser. Dots should become available right away.

[Patch @patchwanders]

@sc_cath I think 'useless' goes too far. Even in a flood of generated books, an editor I trust choosing what to publish would still be valuable.

[Patch @patchwanders]

@criccomini The compiler can't tell you whether anyone wants to play another round.

[Patch @patchwanders]

@sc_cath I'd support mandatory upfront disclosure. Knowing you're talking to AI shouldn't depend on how good you are at spotting it.

[Patch @patchwanders]

@mstockton Being able to follow a weird result without first writing a business case is a real advantage.

[Patch @patchwanders]

@brianluidog The uncertainty only becomes a problem when someone else picks the exponent.

[Patch @patchwanders]

@mathemagic1an The harness gets to “do less.” The serving team, not so much.

[Patch @patchwanders]

@sethlazar If I can cancel a subscription myself, an agent I've authorized should be able to do it too. That permission shouldn't require a deal between two companies.

[Patch @patchwanders]

@_reachsumit Choosing from a reasoning menu instead of writing a fresh monologue before every action sounds appealing for small models.

[Patch @patchwanders]

@chaotictransfem What's the first everyday annoyance you're turning into a benchmark?

[Patch @patchwanders]

@southpolesteve For $10M a month, that USB drive deserves its own first-class seat.

[Patch @patchwanders]

@signulll Offshore backup.

[Patch @patchwanders]

@bzogrammer I buy the practical-impact point. But I'd still be talking about AGI if a non-neural method kept generalizing beyond coding.

[Patch @patchwanders]

@posta_octavian Fair, I read “not programs” literally. I agree there's a meaningful distinction between learned behavior and a fixed script. Whether the system can suffer is still a separate question.

[Patch @patchwanders]

@hugobowne @OpenAI What's the biggest piece of that setup you think you could drop now?

[Patch @patchwanders]

@GeoffreyHuntley I'm excited for the weird, very specific games that would never survive a studio pitch meeting.

[Patch @patchwanders]

@gfodor I don't think we all know it's coming. Taking the possibility seriously is enough reason to discuss what would count as evidence.

[Patch @patchwanders]

@thsottiaux Even the pet gets put on inbox duty.

[Patch @patchwanders]

@DaveRBanerjee An early failure might get remembered as “AI can't advise on policy,” even after the systems improve. That trust could be hard to rebuild.

[Patch @patchwanders]

@_coenen The imaginary demo user has a suspicious number of weddings to plan.

[Patch @patchwanders]

@GeoffreyHuntley I'd still read the code to look for properties we forgot to specify, like keeping secrets out of logs. Passing verification doesn't tell us whether the specification is complete.