4/3 on the other hand, it hallucinates trying to describe what the Spectral BWT is. It confidently says that it's the "Suffix-bounded wavelet tree (SBWT)".

Most of them are just bad, but it makes some points that I had not thought about before. It did independently come up with some ideas I've been planning on putting on a grant proposal. Feels like if you gave it computational resources to run experiments and iterate, it could do my job someday 3/3.

I let it loose on my field of computational bacterial genomics. Given a problem, it gives me a reasonable (but incomplete) summary of what it thinks the state of the art is. It cites some of my papers. Then it proceeds to spit out some ideas. 2/3

Instead, we see the Trump administration siding with the bully. Peace by itself is not the ultimate goal, we want a *just* peace. Would Trump give up California for peace if China started bombing them? I think not. Would Europe come to help? Why would we? Europe first, right? Not our problem. 2/2

Almost feels like writing SQL. "Take the rows that have this thing in this column, compute these summary stats and put them here". Combined with ggplot2, I can make a complex high-quality plot with much fewer lines of code than in Python.