I totally agree. A broad base of knowledge and strong tastes are also my hallmarks of a good researcher. Extra points if their broad base is in something others in their field do not understand.

“To discover something in mathematics is to overcome an inhibition and a tradition. You cannot move forward if you are not subversive.” –Laurent Schwartz

At best, MITGSU is super incompetent. They listed a total of 4 different prices for dues and defaulted me to getting significantly overcharged.

Perplexity seems to periodically and without consulting you change the model you're using to "Best." I think it's a cost saving thing, but then it should be labeled "Cheapest." I notice immediately because output quality is way worse. Many AI companies do this 👀👎.

GPT5 is shockingly good. Not different in kind than previous models but just better. Better at math. Better at coding. Congrats to the team at OpenAI.

Even if AI progress suddenly halted today (it hasn't and won't), we will still see a decade of extraordinary technological progress as we figure out how to build reliable, useful software out of this incredible new material. The future is bright.

I'm shocked that math.mit.edu/ennui/ still works after years with no fixes. I find it very surprising that web technology is this stable.

GPT 5 can go from BNF to working AST and generate correct elements of that grammar (incl typing constraints). It still makes mistakes translating operational semantics to an implementation, but it's shockingly good. It's by far the best model for my research so far.

I think that operational semantics is becoming more relevant... not a strong statement, but still. The ability to concisely express what you want to do, including all subtleties has more value when implementation is more opaque and less difficult.

Successful vibe coding requires great git hygiene and a clear understanding of the desired structure of the generated program.