AI: Rising the bar.

Starting with ourselves

How I feel about the end of Science as we know it:

  • A major part of  the work we do currently will be automated away.
  • What part of what we do is not possible to automate? 
  • Now I have a fun energetic colleague with super powers to answer the questions I care about.

"Judge a man by his questions rather than his answers."

And that is a good thing!

Elegance + Taste

  • In number of papers, I am less productive
  • Learning faster + Attempting more ambitious projects
  • Even though I am running four Claude instances in parallel

In the past the effort spent justified publishing

  • Should we also worry about addiction and possible anxiety inducing effects?
  •  Verification is expensive, I only pay it for the most promising results.

Risk of not learning from others mistakes?

How I feel about working with Claude:

 If we claim we care about trust and reproducibility, we have to start ourselves.

If we care about the arXiv being overflooded, we have to start ourselves.

Let's demonstrate it day to day when we interact with AIs by being the most playful

  1. Reproducibility

2. Overflooding 

3. Lack of deep understanding

The threat of AI in Science

humans

Code + data always public

Let's make papers the maximal version they can be.

  • A source of awe,  pulls people out of themselves and reminds them they're part of something larger.
  • Educating critical thinkers that can reason about a messy world
  • Is Astronomy different from Philosophy? 

No, and it's a good thing

What value does Astronomy add to society?

Requires sharing our knowledge actively

Focuses on human understanding

These may not become software engineers or data scientists.

How do we envision their future?

  • I care about the questions I care about. 
  • As opposed to engineering, it is very difficult to come up with a well defined metric to solve a given problem
  • The best scientists are often the ones that ask the best questions, rather than provide the best answers

Science has always been "art", AI just made it obvious

  • Beyond astrophysics: should we fully give up on open source AI?
  • Should we be thinking about the most ambitious projects we could work on with "a country of geniuses in a data center"?

Opportunities that we could seize

The Encyclopédie

The Human Genome Project

Wikipedia

Planck / SDSS