Re:Claude Opus 4.6
by petercooper on Thursday February 05, 2026 @05:18PM
(#46906181)
One possible trick could be to search and replace them all with nonsense alternatives then see if it extracts those.
[ Parent
| on HN ]
That might actually boost performance since attention pays attention to stuff that stands out. If I make a typo, the models often hyperfixate on it.
Re:Claude Opus 4.6
by jazzyjackson on Thursday February 05, 2026 @10:48PM
(#46908860)
A fine instruction following task but if harry potter is in the weights of the neural net, it's going to mix some of the real ones with the alternates.