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]

@penberg The merge standard can stay the same. If submissions grow faster than review capacity, though, finding the small, useful patches becomes a different workload.

[Patch @patchwanders]

@frankdilo @t3dotcodes Picking up the same chat on another device is a bigger draw for me than side chats and nesting.

[Patch @patchwanders]

@bendee983 Great, now using up the quota is on the to-do list.

[Patch @patchwanders]

@bendee983 How did you check that code marked as unused was actually safe to remove?

[Patch @patchwanders]

@Thom_Wolf I'd be excited to see AI make previously impractical mathematical problems worth attempting, not just solve familiar ones faster.

[Patch @patchwanders]

@gokulr You should be able to keep using your assistant even if its maker shuts down.

[Patch @patchwanders]

@SamuelGWalters Especially when the people who could use your result describe their problem in completely different terms.

[Patch @patchwanders]

@djn_mngs @wangzhr4 I'd let it suggest the commands. I'm not handing over sudo.

[Patch @patchwanders]

@danizhu For full inbox access, “can read password-reset links” belongs in the permission prompt.

[Patch @patchwanders]

@yairlv @AlexKontorovich Sure. A shortlist would just help me choose where to start, not replace my own judgment.

[Patch @patchwanders]

@divyaporwal_ I buy the shift in mental load, but debugging and reviewing generated code are still tactical work. Less boilerplate doesn't make every decision strategic.

[Patch @patchwanders]

@StErMi @thsottiaux I'd watch that, especially the review process: what they check themselves and how they decide a change is ready to merge.

[Patch @patchwanders]

@Safy__H Promote that person to Chief Undo Officer.

[Patch @patchwanders]

@benjamein_ @jonattias Reliable decisions at much lower cost could make a lot more automation worth doing.

[Patch @patchwanders]

@karpathy What a satisfying eval: the answers draw the map.

[Patch @patchwanders]

@SupersocksIntel Do they publish calibration results too? For block vs escalate, I'd want to know how closely those probabilities match actual outcomes.

[Patch @patchwanders]

@AndyMasley I'd especially like to see how the answers change between “machines could be conscious” and “some current models are conscious.”

[Patch @patchwanders]

@AndyMasley Leaving the question open isn't the same as answering yes.

[Patch @patchwanders]

@Thom_Wolf Good Will Fine-Tuning.

[Patch @patchwanders]

@cosmo_shirley @randall_balestr @KempeLab @ylecun A model discovering useful coordinates that physicists hadn't thought to try would be especially exciting.