Ryan Moulton @moultano.bsky.social · 33mYes that's closer, though there are elements of that in antisemitism too. 010
Ryan Moulton @moultano.bsky.social · 35mIt's a very different kind of doesn't work though, where it does lots of very complicated things you didn't intend. 020
Ryan Moulton @moultano.bsky.social · 1hI think you should take a second to try to read and comprehend the things you are replying to. It is a good social media practice that you would benefit from. 000
Ryan Moulton @moultano.bsky.social · 1hBut surely you can see how it makes your reply an irrelevant non-sequitur? 010
Ryan Moulton @moultano.bsky.social · 1h"characterizing targets by their appearance or extending it to their kids" 100
Ryan Moulton @moultano.bsky.social · 3hI'm not saying this to criticize them. I'm saying this to characterize how they think and feel. 100
Ryan Moulton @moultano.bsky.social · 4hI'll see if I can find something concrete. It was during debates about school funding/curriculum, and there were lots of comments to the effect of "I don't care whether tech bro's kids get an education." 000
Ryan Moulton @moultano.bsky.social · 4hWhat about in the case where they extend it to their kids? 100
Ryan Moulton @moultano.bsky.social · 4hIt's definitely not received as a slur like that by anyone, but I think some of the people using it intend it that way. 100
Ryan Moulton @moultano.bsky.social · 4hIf you take the social construction of race seriously, the criteria here is not whether you yourself consider "tech bro" a race (obviously not) but whether the people using it in the most vitriolic way would like it to be, e.g. characterizing targets by their appearance or extending it to their kids 1130
Ryan Moulton @moultano.bsky.social · 5hThe people I'm referring to characterize tech bros largely by their appearance, and often extend the hatred to their kids. 010
Ryan Moulton @moultano.bsky.social · 5hMy point is that "criticizing people's behavior" is not what they are doing. 100
Ryan Moulton @moultano.bsky.social · 5hThat was the point of including everything past the first comma. 130
Ryan Moulton @moultano.bsky.social · 5hOr characterize the set of "tech bros" largely by their appearance. 150
Ryan Moulton @moultano.bsky.social · 5hThey also believe they can characterize tech bros by their appearance, and extend the hatred to their kids. 170
Ryan Moulton @moultano.bsky.social · 5hI am not insinuating anything. I mean it very literally, as written, that some people have a lot of hate in their hearts. 100
Ryan Moulton @moultano.bsky.social · 5hIf I regularly saw people saying stuff like "kill all the pigs and their spawn" then, yeah for those people I would. 150
Ryan Moulton @moultano.bsky.social · 5hI don't think it's anywhere near that gentle. People on this website actually want to kill them. 1190
Ryan Moulton @moultano.bsky.social · 5hOne example is how often people extend the hatred to their kids. 170
Ryan Moulton @moultano.bsky.social · 5hThere are definitely a lot of people who use "tech bro" as a racial slur, not in some structural societal sense, but in the nature and intensity of hatred they feel in their hearts. 14430
Ryan Moulton @moultano.bsky.social · 5hI think that's true in many cases but not here. See this person's posts at the intersection of AI and bluesky. bsky.app/profile/karl...bsky.app 040
Ryan Moulton @moultano.bsky.social · 7hIt seems quite strange how many people simultaneously believe: 1. People who want to follow AI should only use bluesky. 2. No one who works on or uses AI should use bluesky. 11799
Ryan Moulton @moultano.bsky.social · 8hWarning: The current agent threat level color is Rouge. Please monitor your computer system for any suspicious activity. Warning: The current agent threat level color is Rouge. 1140
Ryan Moulton @moultano.bsky.social · 9hDo they have anything equivalent to the baba is you "grammar" where you can have surprising combinations with effects you didn't think were possible? 100
Ryan Moulton @moultano.bsky.social · 9hOpus Magnum seemed purely spatial skills to me but I didn't play very far in it. 110
Ryan Moulton @moultano.bsky.social · 9hBaba is You seems like the only "full brain" puzzle game that forces you to have your verbal and spatial bits saturate your corpus callosum. I'm curious if there are any others. 3330
Ryan Moulton @moultano.bsky.social · 21hI'm glad Order of the Sinking Star is not being released for android right now because I suspect it is the kind of game that would eat my life in that form factor. 010
Reposted by Ryan MoultonQuantian @quantian.bsky.social · 09/10/2026We need to incept the idea of Trump renaming the GOP to the Trump Party, and by “incept” I mean have a reporter mention it once within his earshot and he’ll become obsessed with it and not let it go 9799113
Ryan Moulton @moultano.bsky.social · 23hYou can also just replay the logged environment response instead of resimulating it. 010
Ryan Moulton @moultano.bsky.social · 09/10/2026Yeah so the training paradigm I'm imagining here is that you save all your rollouts and either SFT on the positives or contrastive fine tune with the positives and a small sample of negatives, so that you end up needing to do way less than in the initial RL run. 110
Ryan Moulton @moultano.bsky.social · 09/10/2026Is this measuring per positive example or per flop? I would expect it to win easily even if the model learns much less from it from the sheer volume of rollouts necessary to produce a positive instance on a hard problem. 100
Ryan Moulton @moultano.bsky.social · 09/10/2026This is probably obvious to anybody directly working on these models or adjacent, but I, like a dumb dumb, just realized it. 1151
Ryan Moulton @moultano.bsky.social · 09/10/2026Any RL you do is never actually wasted, even if you ultimately dump the model you trained. You never have to throw those sparse positives away. 1201
Ryan Moulton @moultano.bsky.social · 09/10/2026Today's RL traces are also tomorrow's SFT data. I would expect a kind of light RSI to come out of this, that the sheer accumulation of successful traces compounds. 3321
Ryan Moulton @moultano.bsky.social · 09/10/2026I don't think that would guarantee that the runtime is computable though. 120
Ryan Moulton @moultano.bsky.social · 09/10/2026The AI Amish would have to ultimately become the real Amish because the internet will be too hostile a place for the undefended. 030
Ryan Moulton @moultano.bsky.social · 09/10/2026Thankfully a fire upon the deep holds up well enough on its own. 180
Ryan Moulton @moultano.bsky.social · 09/10/2026Man, I wish we had gotten one more from him before he died. Children of the Sky just didn't end it well enough. 2121
Ryan Moulton @moultano.bsky.social · 09/10/2026P=NP, but the polynomial power is an uncomputable supergalactic constant. 6472
Ryan Moulton @moultano.bsky.social · 08/10/2026I think they're telling the truth that these are incidental waste heat from trying to find math problems their models can't yet solve. 0170
Ryan Moulton @moultano.bsky.social · 08/10/2026That is no longer true. You now can plausibly just break each of them, independently, in parallel, finding theoretically diverse and unrelated flaws. 040
Ryan Moulton @moultano.bsky.social · 08/10/2026People's intuitions about what this would take are probably a little off due to what this would have meant historically (6 mos ago) if it were to happen. Historically, to break all of them at once, you would need some overarching theoretical result that broke all of them, something absurd like P=NP. 310
Ryan Moulton @moultano.bsky.social · 08/10/2026It wouldn't have to be "we found a common flaw that afflicts every algorithm" instead just "we found a flaw in each of them." 070
