Conversation

Notices

1. Embed this notice
   Osma A (osma@mas.to)'s status on Thursday, 02-Feb-2023 17:24:20 JST Osma A

   #ChatGPT has been upgraded with a mathematical model and it no longer trips up (easily, at least) on questions involving unit conversions. It's extremely slow to respond though (overloaded?) and comes up with completely unrelated answers for a new class of questions. This doesn't seem like a model-derived issue, rather than some kind of bug in the chat system implementation. Perhaps my session was mixed up with someone else's?

   • Embed this notice
     patak (patak@m.webtoo.ls)'s status on Thursday, 02-Feb-2023 17:24:19 JST patak

     in reply to

     • Daniel Roe :nuxt:
     • Elk
     • Francois Chollet

     @osma @fchollet @elk @daniel We started to not show mentions when they can be derived from the context, but in this case it seems we have a regression. Would you open an issue with this post?

   • Embed this notice
     Osma A (osma@mas.to)'s status on Thursday, 02-Feb-2023 17:24:20 JST Osma A

     in reply to

     • Francois Chollet

     @fchollet pointed out (on Twitter, boo!) that #ChatGPT correctly thinks that 3458467549867459867492 is not a prime, but also thinks it is such a large number (over 50 digits) that its primality can not be proved.

     "So while it is possible that 3458467549867459867492 has a small factor, it is unlikely."

     https://twitter.com/fchollet/status/1620868751632797703

   • Embed this notice
     Osma A (osma@mas.to)'s status on Thursday, 02-Feb-2023 17:24:20 JST Osma A

     in reply to

     • Daniel Roe :nuxt:
     • Elk
     • patak
     • Francois Chollet

     @fchollet hey @elk @daniel @patak do you see Elk swallowing the start-of-toot mention of fchollet in the ^^ toot? This is how it shows up for me on elk.zone 0.6.2 - but it certainly is there when I check on Mastodon UI.

     Also, how would I edit his mention OUT of this reply? Can't delete or uncheck from the mention line.

   • Embed this notice
     Osma A (osma@mas.to)'s status on Thursday, 02-Feb-2023 17:38:20 JST Osma A

     in reply to

     • Daniel Roe :nuxt:
     • Elk
     • patak

     @patak @elk @daniel https://github.com/elk-zone/elk/issues/1567

     Oh, and: There's no UI affordance to the editor behavior where I can click-to-unmention someone in the reply. I was certain it would break me out of the reply editor and show the person's profile instead, so when the person was actually removed from the mentions with a click, that was a surprise. Separate issue, but related to the larger "how mentions work" theme.

   • Embed this notice
     patak (patak@m.webtoo.ls)'s status on Thursday, 02-Feb-2023 17:38:20 JST patak

     in reply to

     • Daniel Roe :nuxt:
     • Elk

     @osma @elk @daniel Yes, we added the click to remove (with a red color feedback) as a temporal UI to let folks at least work around this issue. But it isn't good. We have two options, one is to avoid a fancy UI for mentions when editing and just show the raw text (I would be fine with this), or the other is to properly do a mentions UI that has each mention as a proper badge with a cross remove button, and a plus button to add a new one, etc.

   • Embed this notice
     Osma A (osma@mas.to)'s status on Thursday, 02-Feb-2023 17:49:55 JST Osma A

     in reply to

     • Daniel Roe :nuxt:
     • Elk
     • patak

     @patak @elk @daniel Adding mentions makes sense to leave as type-their-name behavior. Removing mentions needs a clear UI though. One option would be to allow navigating to the mention line (keyboard or click/tap) and backspace-to-delete (or long-press-to-delete, etc) names you no longer wish to involve in the reply. This would be fairly similar to how inlined mentions behave, yet retain the mentions-are-separate presentation.

   • Embed this notice
     patak (patak@m.webtoo.ls)'s status on Thursday, 02-Feb-2023 17:49:55 JST patak

     in reply to

     • Daniel Roe :nuxt:
     • Elk

     @osma @elk @daniel Just want to say thanks for all the useful feedback you keep bringing to the table about Elk Osma, even knowing about its current limitations. I hope we will be able to get some of the great ideas about data management and perf in pinafore into Elk as you proposed in the future.