henrybemis13.bsky.social @henrybemis13.bsky.social · 7hThe math people that I’ve seen complaining also have been forward friendly early adopters, not luddites. Don’t know why the AI crowd is so quick to dismiss them 140
henrybemis13.bsky.social @henrybemis13.bsky.social · 08/10/2026Then they could work with mathematicians to convert them into papers as Anthropic recently did 1100
henrybemis13.bsky.social @henrybemis13.bsky.social · 07/10/2026And futurists of the past in the USA were the ones who bulldozed neighborhoods to build highways for the sake of progreas 120
henrybemis13.bsky.social @henrybemis13.bsky.social · 07/10/2026The better analogy is walking/cycling. Cars didn’t replace those. In fact we should be doing a lot more walking/cycling and less driving (at least in USA) 120
henrybemis13.bsky.social @henrybemis13.bsky.social · 07/10/2026The more I use AI, the more promise i see in it, and the more disdain I feel toward the anti-humanism of the “pro-AI” crowd 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 07/10/2026I knew he was going to be unhappy about this 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026Have wonder if alpoge is the anthropic employee in reference to 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026It seems like Anthropic is almost going out of their way to contrast themselves with OpenAI by reaching out to the top experts instead of self-publishing à la navies-stokes 210
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026Also crazy what OpenAI engineers post on the other app. 1990s irc-level maturity 000
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026like meta hiring sandberg. Eventually you have to hand over the reins to the adults 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026The labs need to hire generic PR people with a lot of experience who haven’t drunk the kool aid 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 06/10/2026There’s probably a good workflow to use but there’s a lot of bad habits as well. Pilot autopilot is probably the best analogy 000
henrybemis13.bsky.social @henrybemis13.bsky.social · 05/10/2026It’s funny in print but telenovela is often shorted to novela, so it’s actually a legitimate 0170
henrybemis13.bsky.social @henrybemis13.bsky.social · 05/10/2026Very nostalgic to see people talking about inline html 000
Reposted by @henrybemis13.bsky.socialMichael Caley @michaelcaley.bsky.social · 03/10/2026the big questions of "what are LLMs" and "what is alignment" remain important but they are separable from the question of AI safety we can talk about safety as a systems process regardless of what the nature of the technology is, and AI companies are failing in those processes 315515
Reposted by @henrybemis13.bsky.socialMichael Caley @michaelcaley.bsky.social · 03/10/2026this is by far the best "I quit my AI job over safety concerns" essay yet because it is the first one that is able to talk about safety concerns in practical terms and that understands "safety" as a topic with extensive literature in analogous fields www.theatlantic.com/technology/2... 20737177
henrybemis13.bsky.social @henrybemis13.bsky.social · 02/10/2026My thought is that with Lean formalization it becomes a game like any other? But the labs claim that they don’t specifically train on math 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 02/10/2026Do we have a better sense of how they’re able to do advanced math? 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 02/10/2026Any hypothesis why BLS looks weaker than other data sources? 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 01/10/2026It’s still really weird that it can do arithmetic without tools 140
henrybemis13.bsky.social @henrybemis13.bsky.social · 30/09/2026Part of the issue is I do think the labs do a poor job of explaining how LLMs are so unreasonably effective 200
henrybemis13.bsky.social @henrybemis13.bsky.social · 30/09/2026The obvious solution here is to charge the agents that were involved in the hacking 000
henrybemis13.bsky.social @henrybemis13.bsky.social · 28/09/2026Also not the first time we felt threatened by machines. We got over it. Life goes on 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 28/09/2026There’s something deeply anti-human about this. I mean still run and play chess even though machines are better than us 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 28/09/2026AI companies, like Trump, seem to wiggle their way out of everything against all rhyme and reason 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 24/09/2026He got cocky after Venezuela operation I guess 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 21/09/2026I think it would also help if labs some hired normal, professional people for PR. Kind of like Sanderg for Facebook 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 20/09/2026I always tell Claude to be concise and not nit picky. Sometimes even tell it to look for grammar and logic issues only 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 01/09/2026Isn’t this bad considering the economy is growing without producing very many jobs? Job Recovery seems frustratingly slow from 2022. My view may be skewed being from a tech-adjacent field 000
henrybemis13.bsky.social @henrybemis13.bsky.social · 18/08/2026It’s only moving the goalposts when the ai critics are wrong. The ai hypsters are still directionally correct 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 13/08/2026It’s not clear that OP was trying to dunk here. But something rubs me the wrong way about people quoting him 001
henrybemis13.bsky.social @henrybemis13.bsky.social · 13/08/2026Which of course they account for in unemployment surveys, but it’s just more reason to express even basic applied stats lies outside the domain of an hs/101 class 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 13/08/2026And to be fair the original person he’s responding to, they may have meant “reliable” as in subject to nonresponse bias, not sampling variance 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 13/08/2026I’m just pointing out that stats a lot more complicated than most people assume. I don’t even think the simple SRS case is very intuitive, otherwise survey sampling, just from a philosophical level 100
henrybemis13.bsky.social @henrybemis13.bsky.social · 03/08/2026And as a stat analyst, I actually agree with a lot of criticism from Breiman being cited here. But I just don’t understand scientifically. Ultimately we want a data model that’s reasonable predictive. Otherwise we have good engineering but not science 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 03/08/2026I’m a total outsider on this so suspect there actually is a lot of research on this outside of social media vibes 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 03/08/2026I’m actually surprised how little we learned about language after having “solved” translation 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 24/07/2026Obviously, doesn’t take away from the fact that it’s amazing these things can disprove conjecture that most humans can’t even understand 030
henrybemis13.bsky.social @henrybemis13.bsky.social · 24/07/2026Also right now humans are still required to painstakingly check proofs and we don’t have any stats on error rates 220
henrybemis13.bsky.social @henrybemis13.bsky.social · 16/07/2026It really is a game changer though, and I’m surprised people don’t make a bigger deal about it 040
henrybemis13.bsky.social @henrybemis13.bsky.social · 11/07/2026Why they hiring all these eminent people? Just to harvest more training data? 140
henrybemis13.bsky.social @henrybemis13.bsky.social · 09/07/2026Also hate the macho bluster these debates inspire 010
henrybemis13.bsky.social @henrybemis13.bsky.social · 09/07/2026So the only convincing argument for me would be eventually models will be profitable in total, not just excluding training 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 09/07/2026I could be wrong and that’s okay, but I don’t see how training ever not be an issue. Humans deteriorate pretty fast once their out of loop, plus things change so fast 110
henrybemis13.bsky.social @henrybemis13.bsky.social · 05/07/2026My hot take is that the hipsters are being overly pedantic here because the lack of mass job loss is the strongest argument against their claims 220
henrybemis13.bsky.social @henrybemis13.bsky.social · 02/07/2026I mean ed could be right here. Coding in pharma is really different from coding in most tech jobs 030
