Sober Sapien

A sober look at complex systems - tech, probability, philosophy, and human behavior.

[Sober Sapien @SoberSapien]

checkers took almost 2 decades to solve and only had 10^20 states.
chess has ~10^43 positions and 10^120 variations. even 52! (shuffling a deck of cards) is ~10^68 and that already breaks human intuition.
chess isn't just hard, it’s practically unsolveable coz storing that game tree literally needs more atoms than the observable universe. engines dominate humans coz of good heuristics, not coz the game is solved.

[Sober Sapien @SoberSapien]

@Clearpreso Try taking 10 hours of flights in basic economy

[Sober Sapien @SoberSapien]

@signulll When you achieve flow state in a cafe, every other noise feels like white noise. It's quite therapeutic tbh.

[Sober Sapien @SoberSapien]

@MizzyMogs Midtown madness

[Sober Sapien @SoberSapien]

@signulll indian: sir, I'll work entire weekend and would finish the work by Sunday morning.

[Sober Sapien @SoberSapien]

@theo Tbh gemini 7 flash has been pretty decent.

[Sober Sapien @SoberSapien]

@signulll Personal ai agents change the dynamic though. You can have very different thought processes based on the kind of questions you ask the bots. It's active thinking rather than the passive shit feed from social media, and hence quite refreshing

[Sober Sapien @SoberSapien]

Your comment is so far from reality. Men can assume women have it on easy mode because they reduce a woman's entire existence to dating perks. Navigating career ceilings, societal double standards, biological clocks, periods, and then the expectation to carry both the family and the workforce is anything but easy.

[Sober Sapien @SoberSapien]

@KaiLentit How did they manage to convince Magnus Carlsen to perform in this video?

[Sober Sapien @SoberSapien]

@vodka_boy Very simple. Just keep running the model through a fixed internal eval set every day/week. If you notice a drop in performance with the same model, you know something fishy happened to that model.

[Sober Sapien @SoberSapien]

@Thom_Wolf The bot looks amazing. And great design. But tbh I'm still not sure from the video, what is its utility? Is there an easy way to look at some list of what all it can do?

[Sober Sapien @SoberSapien]

@zdogmode Garbage input, garbage output. If the person doesn't understand how to design well, they won't be able to differentiate bad design code from good one. Hence they can't iterate on it well enough, unlike someone who knows coding well.

[Sober Sapien @SoberSapien]

Writing manual code now is like taking stairs 6 floors when there is already an elevator.

[Sober Sapien @SoberSapien]

@signulll Would anyone buy louis vuitton if everyone else went blind tomorrow?

[Sober Sapien @SoberSapien]

While some words of reassurance from bots can feel good, we don't really crave that, we crave the fact that another human chose to spend finite time giving you some words of comfort. If you knew every reply on your feed was generated by an LLM, the psychological reward would be zero. Validation requires proof of conscious attention.

[Sober Sapien @SoberSapien]

@sattyyouneed Anything built on human status games.
We solved chess 25 years ago, and professional chess is more popular now than ever. Nobody buys tickets to watch engines play engines.

[Sober Sapien @SoberSapien]

@anah_sahh Empathy, ethics, learning how to learn, manners, logical thinking, and how to find something that motivates you in life.

[Sober Sapien @SoberSapien]

There are very few people in the world right now who know how to run a LLM on a laptop locally. There are millions (~billion) who know how to run Chatgpt, Claude, Gemini, etc. Distribution matters. Companies distributing stuff to millions/billions need to prioritize safety. Safety research is not pointless. It helps us understand the issues and helps educate people about them. By the time these local LLMs reach a mass, people are already educated.

[Sober Sapien @SoberSapien]

@creatine_cycle Depends on how heavy the silence is.

[Sober Sapien @SoberSapien]

Most debates aren't disagreements over facts, they're unrecognized disagreements over definitions.