PGK

buiding random things with AI for AI against AI

[PGK @facultyofwonder]

I wrote a long essay tracing a single question across psychology, sociology, philosophy of science, data mining, and AI:

What makes something interesting and can we compute it?

As scientific research gets automated, this stops being academic. An AI that can't judge interestingness can't do science. Here's the thread with key ideas. 🧵

[PGK @facultyofwonder]

https://t.co/krTV3YbOwi love this. Rothko drawing closest to your whether.

Here is Berlin today:

[PGK @facultyofwonder]

Isn’t it the same elsewhere except you do not need to pay for it (in Berlin it is basically free)? I read somewhere that for a kid it is also difficult to stay in Kita for longer than 6 hours.

We both work full time and alternate, sometimes I start erlier and pick daughter up, sometime my wife.

[PGK @facultyofwonder]

@tagir_valeev Хорошее видео! Немного покопался и нашел, это не просто деревня -- это юг Берлина.

[PGK @facultyofwonder]

@Net_gover Все так — реальный автомат в трамвае вот. Но справедливости ради через приложение можно купить.

[PGK @facultyofwonder]

@josefchen _All of human cooking compressed into 2 megabytes_ of 300d embeddings? Doubt that.

[PGK @facultyofwonder]

@p0lar_fawn Berlin is surprisingly walkable!

[PGK @facultyofwonder]

@selllsky Это не совсем главная тема, просто это то, что хорошо делит голосующих.

[PGK @facultyofwonder]

@eviliam13 Если дети, то внезапно родители других детей в ките или школе нормально качает круг знакомств.

[alex zhang @a1zhang]

https://t.co/p98KzjT73k

[PGK @facultyofwonder]

So where does that leave us?

Two traditions are converging: formal decomposition (novelty, surprise, coverage…) and FM-internalized taste. Neither is sufficient alone. The path forward combines both — but also needs to solve for whose observer it's approximating, how that observer updates over time, and how findings are presented.

Full essay: https://t.co/0Jlqi3orbd

[PGK @facultyofwonder]

I wrote a long essay tracing a single question across psychology, sociology, philosophy of science, data mining, and AI:

What makes something interesting and can we compute it?

As scientific research gets automated, this stops being academic. An AI that can't judge interestingness can't do science. Here's the thread with key ideas. 🧵

[PGK @facultyofwonder]

The deepest open problem: the abductive jump.

No algorithm in 1982 could have flagged sample 1725 as interesting in the way that mattered — not "this deviates from expectation" but "this requires a new concept."

Formal measures + FM judgment are converging, but that gap is still open.