News for hackers. Stuff that matters.

A comment on DeepMind’s AlphaCode Conquers Coding, Performing as Well as Humans. Parent comment

Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by ren_engineer on Thursday December 15, 2022 @02:07PM (#34004086)
I think we are hitting the flywheel stage where AI is going to be able to start improving its own architecture and performance. Biggest thing holding AI back right now is performance efficiency. Will be interesting to see how AI is used to come up with chip designs and maybe even improvements on its own model. AI can effectively bootstrap itself and improve itself so it can then improve itself again
[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by godshatter on Thursday December 15, 2022 @02:45PM (#34004628)
> AI can effectively bootstrap itself and improve itself so it can then improve itself again

Does this not frighten anyone else? Or am I alone in this?

[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by acuozzo on Thursday December 15, 2022 @03:41PM (#34005429)
You are not alone.
[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by ren_engineer on Thursday December 15, 2022 @03:53PM (#34005589)
it's classic sci-fi Skynet type stuff. AI for example might come up with amazing new nuclear fusion reactor designs so it has access to more energy to do more processing. Could be great, could turn out horribly
[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by kristiandupont on Thursday December 15, 2022 @04:26PM (#34006002)
You are definitely not alone. I am really excited about ChatGPT but it also has me ruminating about the consequences.
[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by status200 on Thursday December 15, 2022 @05:14PM (#34006563)
It has been pondered on and raised terror under the name "intelligence explosion"

https://en.wikipedia.org/wiki/Technological_singularity#Inte...

[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by jimbokun on Thursday December 15, 2022 @05:44PM (#34006852)
Terrified.
[ Parent | on HN ]
Re:DeepMind’s AlphaCode Conquers Coding, Performing as Well ... by goatlover on Friday December 16, 2022 @07:47AM (#34013964)
The LLMs don't perform accurate math, particularly floating point, so I'm not sure how they going to be improving their own architectures, since the neural nets rely on weights, not textual tokens. These models are all designed, fine-tuned, adjusted and fed data by humans.

You act as if AIs are independent agents in the that have goals to modify themselves. Human engineers are doing that. The models don't have any independent goals. They just respond to prompts.

[ Parent | on HN ]