iiinova — where your digital self keeps evolving

libpol.org @libpol.org · 17/09/2026

I also personally think that the opportunity cost of a pause/slowdown may be very bad for many reasons. Slowing down AI has immense costs and risks. But the existential risks of future superintelligence are still real, whether or not a pause is the best approach.

150

libpol.org @libpol.org · 17/09/2026

1) Most of the people demanding a pause/slowdown are people who are not in the AI industry but who know and think a lot about AI. 2) AI is not like other technologies. This is more like nuclear scientists warning the public. 3) People who want a pause understand the immense possible upside of AI.

170

libpol.org @libpol.org · 17/09/2026

Future AI and/or human terrorists working with AI probably could kill millions or possibly billions of people with engineered pathogens and dedicated dispersal efforts, if the plot were not detected early enough. We've already seen what a (likely) non-engineered average virus can easily do.

040

libpol.org @libpol.org · 17/09/2026

The specific nanobot scenario does not need to be true for hypothetical future absurdly intelligent machines to be capable of killing most or all humans, if it will in fact be the case that within this century we will have machines far smarter than the smartest humans at all things.

020

libpol.org @libpol.org · 03/08/2026

Soon they're gonna support A/B testing YouTube display names and there will no longer be any anchor, it will be a pure optimization permutation roulette machine

1130

libpol.org @libpol.org · 03/08/2026

I think both should be able to

000

libpol.org @libpol.org · 03/08/2026

I do not see it as different from a human seeing and learning from things, honestly

000

libpol.org @libpol.org · 03/08/2026

I have seen neither and will see neither (as both seem bad) but I suspect Big Joel would find it more understandable how someone could enjoy Beast Games Season 2 than Live Action Avatar Season 2

130

libpol.org @libpol.org · 25/07/2026

It's not so much that it was weak. It just was not as hardened as it could've/should've been. They may've had a weaker model (+ humans) secure it and did not yet spend time having the latest bleeding-edge GPT-6 checkpoint look for flaws in it to try to make it better.

000

libpol.org @libpol.org · 24/07/2026

Now if the sandbox fucking sucks when we're at GPT-8, that is just insane negligence or malice, but at this particular time and place in what will go down as the history of AI? It pretty much all makes sense (A future GPT-8 may one day break a sandbox, but it'd be through much scarier means)

050

libpol.org @libpol.org · 24/07/2026

Occam's razor, and what I think any technical person who's worked with or in infosec would agree is the default assumption, is that yeah OpenAI just totally fucked up and the sandbox was not that strong. It is completely believable and plausible a burgeoning GPT-6 would reveal this

260

libpol.org @libpol.org · 24/07/2026

they (or at least GPT) committed a serious felony. they did not pre-arrange the committal of the felony with Hugging Face (unless you're a total conspiracist). now, there might be other factors at play, maybe the prompt was really aggressive, but they obviously did not plan or intend or want this

010

libpol.org @libpol.org · 24/07/2026

some people I know who I even respect - technical people who are smart - are saying it's a publicity stunt I'm at the point where I feel like the actual based contrarian stance on any issue is to trust and believe what a government or corporation says www.lesswrong.com/posts/9kcTNW...

lesswrong.com

Intellectual Hipsters and Meta-Contrarianism — LessWrong

110

libpol.org @libpol.org · 24/07/2026

I think that is the plan but it may be so difficult that there may be many years between "we might have conscious systems" and "we might have a fairly reliable way to measure presence of consciousness in an entity"

000

libpol.org @libpol.org · 24/07/2026

Well, we're in agreement on that. Obviously there has to be some kind of sensible peace plan here, "kick Israelis out of Israel / kick Palestinians out of Gaza and the WB" are not moral options. But with Trump + the right and far-right Nentanyahu admin, seems hopeless.

110

libpol.org @libpol.org · 24/07/2026

Apartheid is not quite a fitting analogy, true, but what's happening in the West Bank (and overall region/vicinity) is arguably worse than South African apartheid in some ways. Especially given how long it's lasted, with no end in sight.

100

libpol.org @libpol.org · 23/07/2026

I have literally never once sent an emoji to a human, though. Let alone an LLM. And I've been on the internet since I was 9. (Instead I just end every other sentence with "lol") I talk to Opus/Fable/Sol like a human coworker pretty much, but I also talk to it like a model.

030

libpol.org @libpol.org · 23/07/2026

Of course, I could turn out incorrect. But I'd bet on it at like 85% odds, and 95% if extended to 10 years.

100

libpol.org @libpol.org · 23/07/2026

Change in that probably within 2 - 6 years we will start to see a lot of AI-generated images which are both indistinguishable from human-made images and which, to many humans (not primed that it may be AI), appear aesthetically tasteful, artful, pleasing, creative, high-quality, and "non-slopful".

100

libpol.org @libpol.org · 23/07/2026

I left a sol running for a few days and....well, it did things, alright I think we need GPT-6 before these things cross a threshold into "more often than not superhuman at all scales" (even without careful prompting)

000

libpol.org @libpol.org · 23/07/2026

The thing is, within some years a lot of AI-generated images will no longer have that feel. Eventually, a huge percentage won't. We're in a weird early period where they almost always make generic mulch. Same for coding, etc. This will change relatively soon, in the scheme of things.

120

libpol.org @libpol.org · 23/07/2026

Or, in other words: one day AI will be able to make non-slop. And lots of humans make slop every day and will continue to. Your political and economic ideology needs to account for that eventual future, because it will be within your lifetime.

000

libpol.org @libpol.org · 23/07/2026

At *some* point they'll be able to make things that even tasteful, eclectic art critics look at and - not being biased by knowing it's AI - would say "okay, that's good". So you can't just rely on that as an eternal reason to oppose AI-generated images/videos.

100

libpol.org @libpol.org · 23/07/2026

I don't think there's any real issue if it is aesthetically good (perhaps in some kind of blind testing scenario, where you aren't primed by knowing it's possibly/likely AI) + if the human credits the AI with creating the piece Kneejerk "we cannot let machines create drawings" is senseless

100

libpol.org @libpol.org · 23/07/2026

I think there are a few different things here: 1) being aesthetically tasteless/bad, 2) being soulless, 3) a human artist not putting in effort to create output, 4) a human taking credit for making something when a machine made it, 5) job loss 1) will probably get (sometimes) solved within 10 years

211

libpol.org @libpol.org · 23/07/2026

People can, and may try to, use generative AI for bad and immoral things, just as some people (including powerful people) use the internet for bad and immoral things, but we obviously also get great benefit from the existence of the internet and the existence of intelligent thinking machines.

000

libpol.org @libpol.org · 23/07/2026

You're dancing between two different things: saying it's incapable (which is antivax-tier science denialism at this point) and also that it's evil and will destroy everything. I think you're just trying to throw whatever you can at the wall, here. There's nothing morally wrong with generative AI.

110

libpol.org @libpol.org · 23/07/2026

exactly I doubt Haus of Decline has literally antisemitic intentions here (especially since she's Jewish) but it is just not a good tweet

010

libpol.org @libpol.org · 22/07/2026

No one (including Will) is saying Will is receiving more harassment than the average trans woman target of Kiwi Farms receives. He's saying that website is one prominent example of how this kind of collusive harassment does not stop even if the person tries to disappear.

150

libpol.org @libpol.org · 22/07/2026

the insane thing is at this point I'm pretty sure you would actually be getting less harassment on Twitter than on Bluesky (narcissism of small differences etc.) if you had stayed there. I respect your perseverance

040

libpol.org @libpol.org · 22/07/2026

somewhat paradoxically, I think the tweet would (non-ironically) actually be less antisemitic - given that context - if it had said "Jewish" instead of "Israeli"

170

libpol.org @libpol.org · 22/07/2026

if you want change, get someone similar to Mamdani elected president in 2028. that is the best path

100

libpol.org @libpol.org · 22/07/2026

the consequences if Mamdani tries to have him arrested would only result in Netanyahu being almost immediately freed + very bad consequences for NYC he does not have the jurisdiction. Mamdani cannot do anything here with any net positive result (beyond a few hours)

141

libpol.org @libpol.org · 22/07/2026

they started working together after each realized OpenAI was the responsible culprit (...how could they not work together after that?? especially if you want to avoid lawsuits and criminal charges) they obviously did not coordinate or plan this at all before this happened. you are being misleading

100

libpol.org @libpol.org · 21/07/2026

it is, but also the hype is justified. I work in infosec and am fully in the EA/rat camp on these; if GPT-6 or Mythos were released with absolutely zero guardrails, so many people would use them to wreak havoc in ways they previously couldn't. including very unskilled cybercriminals

060

libpol.org @libpol.org · 21/07/2026

there could be laws that make the owner of an AI instance fully criminally responsible for anything the instance does, but I'm not sure if such laws are on the books as of now

100

libpol.org @libpol.org · 21/07/2026

interesting legal question - does the CFAA cover non-malicious intent by a human who manages a machine that does have malicious intent? the company lacks mens rea here even if the agent doesn't. OpenAI could surely be liable for negligence, but unclear if it'd be a felony for anyone but GPT

320

libpol.org @libpol.org · 21/07/2026

wait this is actually the biggest story of this kind by far, no? much more of a true sandbox escape situation than the other recently disclosed incident, even without the additional crazy 0-day HF RCE exploitation cherry on top. and this escape was via a second separate 0-day

0341

libpol.org @libpol.org · 21/07/2026

Do we have a slopulism blocklist and if so how do we get you added to it

000

libpol.org @libpol.org · 21/07/2026

Anthropic is pretty notorious for having very liberal (as in, anti-fascist) executives Sam Altman is probably liberal Demis Hassabis is definitely liberal Elon is an insane fascist but he's basically dipped out of the model race

101

libpol.org @libpol.org · 21/07/2026

You people are going to be saying this when it's running the whole economy (better than any human could) in a few years.

000

libpol.org @libpol.org · 21/07/2026

It is probably a matter of years before the machines are the ones that are pulling the levers on the machines, though. Humans will still find meaning in things, but people are underestimating how transformative things will be. We're just really early right now.

000

Reposted by libpol.org

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 20/07/2026

There are no risks from a system unless it is intelligent. Picture unrelated sorry I can't seem to delete it

219012

libpol.org @libpol.org · 21/07/2026

"Here's some data." "Well here are my anecdotes and my emotions." This is why people scoff at people who hate economists. You're acting like how right-wingers respond to things.

1410

libpol.org @libpol.org · 21/07/2026

The amount of people going "ummm read the room!!" in response to someone making an objective claim is depressing. People expect this sort of irrational personal feelings-based anecdote-over-data reasoning from the right. It's disheartening to see it from the left.

171

libpol.org @libpol.org · 04/07/2026

Let's say AOC or a hypothetically eligible Mamdani won 2028. I could see lots of scenarios where this radicalizes right-wingers even further and they elect someone more extreme than Trump in 2032 or 2036. I could also see this happening due to a very milquetoast guy like Newsom winning, too, though.

000

libpol.org @libpol.org · 04/07/2026

I think it's a complicated calculus. Socialists like to say that if we elect an establishment neoliberal or whatever that it'll cause us to get someone much worse than Trump not long after. But, like, it could go the exact other way. Or it could end up happening either way.

100

libpol.org @libpol.org · 06/06/2026

side note but it's sad and disappointing (though very much not surprising) that that guy is a neo-nazi

020

libpol.org @libpol.org · 06/06/2026

I feel like the tech industry is one domain where this is actually generally not true, for worse and for better

110