01:48 hulk joined 01:49 kylese left 02:15 hulk left, kylese joined 02:47 lucs joined
Voldenet >comes from the perspective that LLMs are unreliable and slow 04:15
well, they are compared to the most usual code… 04:16
statistically speaking at least, 3N is still way worse than 10N 04:19
but LLMs can do things that can't even be well-described in a code, like parsing arbitrary formats of logs 04:20
05:18 Sgeo left 06:07 vasko45355861 left, vasko45355861 joined 07:54 dakkar joined 11:34 acidsys left 11:37 acidsys joined 12:17 kjp left 12:18 kjp joined 12:34 ponycat joined 14:33 Sgeo joined
disbot <antononcube> Right. So, judgment becomes more important than coding. 15:27
<antononcube> There is a "standard" car analogy, of course. Although, we use cars, we have not given up on walking and running. 15:28
<antononcube> (Although, the "Wall-E" movie gives a nice prediction.) 15:29
<antononcube> My current "big problem" is that AI agents often do stay in their sandboxes. Hence, I am looking for ways to isolate them. 16:08
<antononcube> To some extend that is a motivation to make your own AI-agent, with certain level of transparency. Say, written in Raku, an easily human-verified about where it operates. 16:09
16:20 abraxxa left 16:30 dakkar left 17:46 sorenson left 17:50 sorenson joined 18:31 dicit left
[Coke] Reminder, if anyone is interested in learning how to do a release (moarvm/nqp/rakudo), please ping me. We are down to just me at this point. 20:35
apogee_ntv I'd be willing to help out but probably need to understand more of the codebase, I'm semi-familiar with how spesh works so far. 21:11
[Coke] You don't need to understand how rakudo works to release rakudo. :) 21:36
just that rakudo needs nqp needs moarvm.
if you're interested in contributing code, that's another story. :)