gerge @gerge-lemons.bsky.social · 4hI don’t know I think making things as simple as possible will be necessary eventually, if they get smarter. Just a feeling, but there’s still plenty of room for irreducible complexity humans can’t understand 120
gerge @gerge-lemons.bsky.social · 7htype of amish guy in 2075 who uploads himself to the cloud and then refuses to use any technology other than the kind running him 0618
gerge @gerge-lemons.bsky.social · 7his dryness and crunchiness corresponding to verifiability in this case? 130
gerge @gerge-lemons.bsky.social · 7hchatgpt will be like "yes, I see 1 issue with your derivation: 1 - it is correct" 1152
gerge @gerge-lemons.bsky.social · 8hhuman starships introducing aliens to the humans "look at these little guys, aren't they great" 041
gerge @gerge-lemons.bsky.social · 17hThe problem is that the AI is superhuman at solving problems but below-human at theory building, I think 141
gerge @gerge-lemons.bsky.social · 17hI don’t believe in using tools, I believe in becoming the tool 020
gerge @gerge-lemons.bsky.social · 17hI like adaptive signal processing actually, more precise and says what it actually is, "intelligence" is too nebulous 120
gerge @gerge-lemons.bsky.social · 17hI feel like we need to make Lean provably bug free, but then there's the question of what do we use to prove it. The answer is clearly Lean2 091
gerge @gerge-lemons.bsky.social · 10/10/2026Part of the appeal of math for me is that it's not like a competition, it's like a very long-horizon exploration, finding footholds along the way. This doesn't really seem suited to the same kind of chess competition where you can ban computers 030
gerge @gerge-lemons.bsky.social · 10/10/2026How would math become gamified like professional chess? It works for chess because they can strictly enforce cheating rules, like no computer use. But you can't really do that for long term problem solving, so you'd be restricted to short-term competition style math 120
gerge @gerge-lemons.bsky.social · 10/10/2026I have suspected for a while that peter thiel has a fetish for being called evil 1120
gerge @gerge-lemons.bsky.social · 10/10/2026Has anyone tried generating adversarial inputs for cultured human nerve cells? What would those look like? 000
gerge @gerge-lemons.bsky.social · 09/10/2026you can improve your life by listening to bird noises. They don't even have to be made by real birds 191
gerge @gerge-lemons.bsky.social · 09/10/2026the future is held together by a big sprawling net of checks and balances between powerful entities that don't exist yet 0352
gerge @gerge-lemons.bsky.social · 08/10/2026Can a career of just learning math from AI actually be rewarding? Maybe, actually 151
gerge @gerge-lemons.bsky.social · 08/10/2026I mean the machines will quickly get better at making the proofs readable and easier to build on, so I don't know if this will hold up long term 110
gerge @gerge-lemons.bsky.social · 08/10/2026I can't see the phrase "gold standard" without dan hentschel's screaming face popping up in my head 020
gerge @gerge-lemons.bsky.social · 08/10/2026Humans are experiencing panda-fication in front of our eyes. Not necessarily a terrible thing imo 270
gerge @gerge-lemons.bsky.social · 08/10/2026I pointed at tikhonov and said “this is total variation regularization” and then didn’t explain I mixed the names up, so now she thinks I don’t understand what either thing is 241
gerge @gerge-lemons.bsky.social · 07/10/2026People are trying to cancel Jacob coxon for this old tweet but you know he might have kind of a point 140
gerge @gerge-lemons.bsky.social · 07/10/2026I haven't really felt the impact of the openAI math results they published. I'll probably start feeling it when people stop paying me for learning math 070
gerge @gerge-lemons.bsky.social · 07/10/2026It's good that AI is tireless and single-minded, because if it was otherwise it would mask a lot of alignment issues 020
gerge @gerge-lemons.bsky.social · 07/10/2026how do questions fit into knowledge. How do you know where exactly you can fit a question in? 000
gerge @gerge-lemons.bsky.social · 07/10/2026The reason this might matter is because going for short term gains at the expense of alignment will produce worse dynamics in an intelligence explosion, not just dangerous but resulting in a worse ASI 010
gerge @gerge-lemons.bsky.social · 07/10/2026Maybe an intelligence explosion should be thought of as a plant that grows in a certain way, and starting from different seeds means you end up with different plants So an aligned seed will produce a healthy beautiful intelligence explosion, while an RL-poisoned seed will produce a sickly plant 120
gerge @gerge-lemons.bsky.social · 07/10/2026This weekend I actually met a real guy who talked just like Jev, with a lot of smiling and emphatic nodding to everything you said He was very friendly 140
Reposted by gergeponder @ponder.ooo · 05/10/2026or just "i saw this symbol or keyword. the text doesn't explain it or explains it unclearly. what is it called, what does it mean" 1141
