Newest at the top
| 2026-05-24 11:46:14 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 11:35:00 +0000 | <int-e> | . o O ( What's 2+3? -- I don't know. My best guess is five. ) |
| 2026-05-24 11:34:31 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 264 seconds) |
| 2026-05-24 11:33:31 +0000 | <tomsmeding> | :p |
| 2026-05-24 11:33:23 +0000 | <int-e> | Oh I'm sure it's easy to get a number. The hard part is making it reflect reality ;) |
| 2026-05-24 11:32:28 +0000 | Alex_delenda_est | (~al_test@5.139.233.99) |
| 2026-05-24 11:32:06 +0000 | <tomsmeding> | if it was really as simple as what Leary said, then you don't need training for that, and it could be a switch in the UI, which would get a bunch of media coverage |
| 2026-05-24 11:31:47 +0000 | <int-e> | But I do imagine that getting a good handle on certainty is actually hard. |
| 2026-05-24 11:31:15 +0000 | <int-e> | Or, perhaps, not desirable (scores poorly in post-training with humans) |
| 2026-05-24 11:30:42 +0000 | <tomsmeding> | I've had that idea too, about 5 secods from hearing about the "can't say 'don't know'" problem. I think all machine learning researchers have. So the fact that people don't do it ought to mean that it's indeed hardder than it sounds :p |
| 2026-05-24 11:29:32 +0000 | weary-traveler | (~user@user/user363627) (Remote host closed the connection) |
| 2026-05-24 11:29:13 +0000 | <Leary> | Hence it can say "I don't know" even if the training data never does. I don't know why this isn't already widely done; it's probably harder than it sounds. |
| 2026-05-24 11:29:11 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 11:28:38 +0000 | <Leary> | Each token the machine generates is chosen according to a weight/probability, from which a confidence score for the entire output can be derived and warnings issued when below corresponding thresholds. |
| 2026-05-24 11:21:09 +0000 | <[exa]> | cool example |
| 2026-05-24 11:21:06 +0000 | <[exa]> | okay nvm |
| 2026-05-24 11:21:04 +0000 | <[exa]> | oh wait no, it's a different `a` than in the class |
| 2026-05-24 11:20:34 +0000 | <[exa]> | is there actually any way to annotate the type family so that the instance would know that `a` is in * ? |
| 2026-05-24 11:20:02 +0000 | <[exa]> | :) |
| 2026-05-24 11:18:22 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 244 seconds) |
| 2026-05-24 11:13:25 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 11:08:55 +0000 | <tomsmeding> | it's not the most likely answer when you read webpages and books :) |
| 2026-05-24 11:08:27 +0000 | <[exa]> | tomsmeding: I love how these things are unable to say "I don't know" |
| 2026-05-24 11:07:21 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 272 seconds) |
| 2026-05-24 11:01:58 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 10:50:27 +0000 | merijn | (~merijn@62.45.136.136) (Ping timeout: 245 seconds) |
| 2026-05-24 10:43:57 +0000 | merijn | (~merijn@62.45.136.136) merijn |
| 2026-05-24 10:33:19 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 264 seconds) |
| 2026-05-24 10:29:54 +0000 | DetourNetworkUK | (~DetourNet@user/DetourNetworkUK) (Ping timeout: 245 seconds) |
| 2026-05-24 10:28:23 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 10:26:44 +0000 | drdo | (~drdo@2a01:4f8:1c1c:4976::1) drdo |
| 2026-05-24 10:26:09 +0000 | drdo | (~drdo@2a01:4f8:1c1c:4976::1) (Remote host closed the connection) |
| 2026-05-24 10:18:04 +0000 | target_i | (~target_i@user/target-i/x-6023099) target_i |
| 2026-05-24 10:17:28 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 267 seconds) |
| 2026-05-24 10:14:09 +0000 | AlexZenon | (~alzenon@5.139.233.99) |
| 2026-05-24 10:12:56 +0000 | Sgeo | (~Sgeo@user/sgeo) (Read error: Connection reset by peer) |
| 2026-05-24 10:12:25 +0000 | merijn | (~merijn@host-cl.cgnat-g.v4.dfn.nl) merijn |
| 2026-05-24 10:09:55 +0000 | Alex_delenda_est | (~al_test@5.139.233.99) (Ping timeout: 264 seconds) |
| 2026-05-24 10:09:00 +0000 | AlexZenon | (~alzenon@5.139.233.99) (Ping timeout: 244 seconds) |
| 2026-05-24 10:05:18 +0000 | merijn | (~merijn@62.45.136.136) (Ping timeout: 252 seconds) |
| 2026-05-24 10:05:05 +0000 | <tomsmeding> | fun: Claude Opus cannot explain why the little reproducer fails, even when given access to GHC to test stuff. It concludes that it's a known limitation of GHC. |
| 2026-05-24 09:58:42 +0000 | merijn | (~merijn@62.45.136.136) merijn |
| 2026-05-24 09:56:50 +0000 | vanishingideal | (~vanishing@user/vanishingideal) vanishingideal |
| 2026-05-24 09:54:54 +0000 | DetourNe- | DetourNetworkUK |
| 2026-05-24 09:52:52 +0000 | DetourNetworkUK | (~DetourNet@user/DetourNetworkUK) (Read error: Connection reset by peer) |
| 2026-05-24 09:52:50 +0000 | <tomsmeding> | fun, this is _very_ hard to spot if you're not aware of how the kind system works precisely |
| 2026-05-24 09:52:45 +0000 | <Noinia> | Thanks for the help! :) |
| 2026-05-24 09:52:38 +0000 | DetourNe- | (~DetourNet@user/DetourNetworkUK) DetourNetworkUK |
| 2026-05-24 09:52:21 +0000 | <tomsmeding> | Noinia: that seems like the correct solution, yes |
| 2026-05-24 09:52:11 +0000 | <Noinia> | (i.e. restricting the Svg type signature to kind Type) |