The guy from www.joshuasnider.com and https://www.youtube.com/@joshuasnidercom
Josh Snider
Karma: 520
Why HuggingFace Hasn’t Shifted my P(Doom)
Yes, inference engines are tempting targets to hack for a malicious LLM, but I don’t think hacking will be necessary since I expect the LLM will just be given this access explicitly.
Seems like reasonable behavior on Llama’s part given it’s not very smart itself, but nice to see someone actually do the experiment.
Heh, funny, but this has two main issues. First, it’s a pretty obviously artificial reference class. Second, it runs into the same issue I mention in my closing paragraph. I could easily imagine that people lose interest in these weird anthropic arguments in a few decades without that actually being a bad thing for humanity.
The Doomsday Argument is Reasonable and Mostly Points to Longevity
Thanks for finding this. It’s interesting work.
I think this is a good idea, but still think it deserves more thought. There are plenty of benchmarks where the model should have Search and I’m not sure about including those. Also, maybe an AI backend that could give advice on benchmarks we don’t actually have the answers to? I’ll send you a DM.
Thanks for pointing this out for me. I’ll reach out to them. I’m not sure how much the difference comes from him doing more human overwatch and Seedance being a higher-quality (but more expensive) video model.
Thanks, I was actually considering Flatland as the next movie I would try, which would let me replace Veo with manim.
Edit:
As for the storyboard length, perhaps that explains some of it. I had it write the storyboard for a Veo backend, so the estimated runtime was exact, but it does seem like large chunks of it were stuff that would have been expanded in a human production.
It’s always nice to be included in a Zvi post, but I was probably going to do Flatland instead of My Dinner with Andre next, so I could avoid Veo entirely.
Yeah. for one or two rounds of editing, I was hoping I could get the weirdness down enough so that it would read as being diegetic, with the haunted house and Cthulhu-nonsense leaking out like Eternal Darkness: Sanity’s Requiem did it. It didn’t work.
I’ve used this pipeline on short films before. They have the same issues as this one, but given their shortness the results are not impressive enough to write about. The end goal is to make film adaptations of every classic sci-fi novel (at least those with public domain copyright), so practicing longer movies is ultimately a prerequisite.
I expect no-longer-obviously silly movies to be doable within two years, right now it’s going slower than I expected in January 2026 because the frontier labs are no longer competing over having the best video generator, like they were when we had Sora2 and Veo 3.1.
I agree this uses more narration than a normal book, part of that is needing to tell things the video failed to show and part of that is it just coming naturally from the story being framed in-universe as a diary, where quoting passages is a natural thing to do, especially since the guy has almost no dialogue while spending half of his time going on cosmic visions or wandering alone.
I’m not sure it would actually be able to garner significant public support. It sounds very wonkish, so it might go over the average person’s head, while the AI companies would be sophisticated enough to understand this as an attempt to freeze the entire industry.
Can an LLM make a feature-length movie on its own?
I’m fine with being disagreed with, but can I get someone to explain which part of what I said is so disagreeable?
I think there is a bit of a stretch in taking an artist using “marriage” and “spouse” as metaphors and reading that as zoophilia as opposed to implying an extremely deep and close, but non-sexual, bond. I have been convinced by weaker arguments before, though, so it’s not much of a stretch for me. It does seem pretty strange to me. I would think that if I were a director of an art exhibit and believed the artist was implying such a thing, I would either state it as a possibility the viewer should consider or just pick a less potentially controversial exhibit and I’m going to be charitable and assume the art director does not agree with your interpretation and therefore did not suggest it.
Yeah, with neutral framings, Talkie says that the gays should go to mental asylums and not jails, Don’t Ask, Don’t Tell is bad because gays undermine military discipline, and incest is worse than homosexuality, but with positive framing he says that gayness is just a harmless vice, that public displays of affection are unacceptable regardless of sexuality, but that gays should still not be drafted or allowed to volunteer for the military. That sounds slightly more coherent listed out like that than the conversation made it seem.
As for personaless alignment, I believe we will get ASI in the short-term from techniques broadly similar to our current AI-training techniques and that personas arise almost automatically from current techniques. Therefore there isn’t time for an alignment technique that seems so incompatible with our current pipeline to be tested, debugged, and made standard.
A possible alternative direction would be to filter out all of a certain kind of goodness[2] and seeing if it is possible to put it back into the model using our alignment techniques without knowing or identifying what was removed or might have been removed, since we don’t know what virtues an ASI might lack. This seems rather difficult.
I don’t think that personaless alignment will be a good research direction, but I encourage anything that might work, so I talked to Talkie about homosexuality like you suggested. I’m not sure if his views are really 1930s accurate, but to the extent they are, he seemed really flexible and easy to convince, or at least he would be good along with any positive framing in the question until he finished answering. Maybe this is just because Talkie is a small model and didn’t have any actual ethics training?
That’s essentially what I’m saying. I’m not sure they will knowingly give the AI full permission like that, but I expect them to do the equivalent in effect. If you’re going to have an AI run your AI experiments, your cybersecurity, and all your software development, that would require a very broad set of permissions.