>>24832 (me)
>For the science people outside, health, linguistics and electronics looking in wondering how those fields are getting results while your field is spinning its wheels, it's because those fields require systems thinking to really get anywhere so you're constantly taking notes and rethinking and reviewing your prior notes seeing whether you can find something systemic that fix a lot of things at once and make thinking about the whole problem simplerOn a practical level, it's a tool
When you're working with a chat bot the information in the middle of the Context, ie the stream of text/tonkens so far, has a tendency to get log jammed by the things at the beginning and end of the context; this is due to the structure of the attention based networks most models use
Theoretically, recurrent networks maybe; it may just be a fundamental limitation
The information is still encoded, it just doesn't make it into the output, so on your next prompt you gently nudge it back along with your next note, and if it's an alignment issue, since the inline, in context learning is to oversimplify it a little just back propogation, if you're careful and detailed with your notes the information jammed in the middle should come out
There you go, a machine summarising your notes on every note
Great for science, but you still have to do your own thinking; and then go back and double or even triple check everything
Like I have to go to China anyway to get a specific kind of ink anyway, so I might as well get the parts – since 中国 is the only place that makes them on an industrial scale anyway, or at all in a lot of cases – while I'm there to prove some things in practice
Post too long. Click here to view the full text.