Conversation

Notices

1. Embed this notice
   Winchell Chung ⚛🚀 (nyrath@spacey.space)'s status on Tuesday, 15-Apr-2025 01:32:43 JST Winchell Chung ⚛🚀

   Is it just me? Or is this a catastrophically bad idea?

   Can AI Help Manage Nuclear Reactors?

   https://m.slashdot.org/story/440875

   • Embed this notice
     feld (feld@friedcheese.us)'s status on Tuesday, 15-Apr-2025 01:32:41 JST feld

     in reply to

     • David Nash
     @dpnash @nyrath do you think computers should be involved in nuclear reactors or do you think we should make humans try to remember what all these buttons do?
   • Embed this notice
     David Nash (dpnash@c.im)'s status on Tuesday, 15-Apr-2025 01:32:42 JST David Nash

     in reply to

     @nyrath > Can AI Help Manage Nuclear Reactors?

     Definitely a case of Betteridge's law of headlines applying, in full force.

     I once asked ChatGPT how to change the engine oil in a 2017 Chevrolet Bolt. It gave me tolerably detailed steps for the process to change the engine oil. Unfortunately, this is an electric car, so it has no engine oil to change in the first place. The equivalent level of ignorance in advice to nuclear plant operators is too terrifying to contemplate.

   • Embed this notice
     feld (feld@friedcheese.us)'s status on Tuesday, 15-Apr-2025 03:08:56 JST feld

     in reply to

     • J. Steven York RESISTS
     • David Nash
     @jstevenyork @dpnash @nyrath one of the problems with Chernobyl was human error. So yes of course we should have computers involved. That may include an AI model *actually trained on the relevant data*
   • Embed this notice
     J. Steven York RESISTS (jstevenyork@mastodon.social)'s status on Tuesday, 15-Apr-2025 03:08:57 JST J. Steven York RESISTS

     in reply to

     • David Nash
     • feld

     @feld @dpnash @nyrath
     I don't know anything, but I'm pretty sure those buttons represent the rods in a nuclear reactor. There are a few different colors that mean something. I'll bet those red ones are important. Not hard at all for somewhat with actual knowledge to understand. Knowing what to DO with the buttons is likely a whole different thing.

   • Embed this notice
     feld (feld@friedcheese.us)'s status on Tuesday, 15-Apr-2025 10:46:03 JST feld

     in reply to

     • David Nash
     @dpnash @nyrath They're not notoriously bad at a task when they're trained specifically for the task. But you'll never believe it if you don't spend time working with it to explore what it is capable of
     
     In the end, this is just another algorithm. It is worth coexisting with all our other algorithms as a new useful datapoint
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     David Nash (dpnash@c.im)'s status on Tuesday, 15-Apr-2025 10:46:04 JST David Nash

     in reply to

     • feld

     @feld @nyrath The sudden shift from what I was discussing (gen AI and LLMs, which are *notoriously* bad at this sort of thing) to generic “computers” for control (which can be, and often are, perfectly fine for it) is quite remarkable indeed.