Did you try running it on some standard heuristic programming competitions? (i.e. atcoder's AHC or topcoder's marathons)

Would be nice to see it benchmarked vs something where the human effort can be easily quantified

Oh man, I almost missed it. His Jibaro & Witness are my favorite animations from LD&R.

Oops, the correct link is: sudokupad.app/qq3kyjvu5q?s... (hides the grid)

Also, check out the official post on Logic Masters: logic-masters.de/Raetselporta... that includes the FAQ

Link to (ugly) code: github.com/FakePsyho/cp...

This result is surprising to me, as it took ~4h + I didn't think it would be possible to get close to 1st without doing any of the fancy rail placement (either optimizing at the end, splitting existing rails for a new station etc.) and/or complex BS eval.

* BS is very slow, I'm grouping states by turn and eval is expected money at turn 800 + small bonus for partially fulfilled trips.

I have also added DP at the end (same as nika), during the contest I had greedy instead of DP and incorrectly assumed that it doesn't add too much

Not that I disagree with anything that you've mentioned, but all I had in mind is that AI training thing is a silly take.

No matter if you're a republican, democrat or moderate, it is in your best interest to get rid off EC. The only people that are happy with the system are full-time politicians as it gives them a monopoly.
Both parties are screwing you over with their joke version of democracy.

Thanks for the hints, I think I clicked "show me less..." thing few hundred times already, but so far no results :( Maybe it updates user preferences in batches and takes a few days to kick in.

I mostly post about interesting/unusual games, immersive entertainment and competitive programming. From time to time, I also vent my frustration about the terrible state of AI safety.

Please like, comment or subscribe 😅

PS. What do I do to remove all of that anime from my discovery tab