AI is unlikely to speed up science by multiples acting solely as a co-pilot to humans. There are some early successes using current LLMs to autonomously research, plan & conduct experiments, including learning how to use equipment by reading documentation, with big potential rewards (& risks). - ThreadSky

About ThreadSky

emollick.bsky.social • 19 days ago

AI is unlikely to speed up science by multiples acting solely as a co-pilot to humans.

There are some early successes using current LLMs to autonomously research, plan & conduct experiments, including learning how to use equipment by reading documentation, with big potential rewards (& risks).

1 / 3

Comments

marcomilani-it.bsky.social•19 days ago

This is great possibility.
Risks? Going quickly towards the self replication condition that would allow intelligent entities to exist without any need of the human being.

korpivaara.bsky.social•18 days ago

How about bridging the challenges with data overload? In medicine the literature health and biology is doubling every ca 72days. No human can keep up with the new information but with developing Specialist Silos

bradleybirddos.bsky.social•18 days ago

LLMs have a difficult time following specific instructions. Until that's solved they won't be able to follow documentation.

pogue.omg.lol•19 days ago

This might interest you, Professor

https://www.reddit.com/r/selfhosted/s/yUt5EsNvv1

safavi.bsky.social•19 days ago

Soft disagree. A bottleneck in science is coordinating experts in different fields + aligning incentives. Soon, a researcher in field X can just simulate collaborations with experts in Y & Z.. these may “only” be copilots. As a bridge between disciplines, this can speed up things by multiples.

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply