RedMan

Karma: 628

RedMan 14 Apr 2026 18:50 UTC
2 points
0
on: The frightening future (i.e. present) of AI surveillance
When doing big data analysis on stuff like this, there’s a big difference between generating a story that seems to make sense and generating correct conclusions.
For these examples, how are you validating claude’s conclusions? Are you certain enough to put warheads on foreheads? People during that time were manually analyzing these types of data, and they absolutely were making those decisions.
What is the readiness level of the tech for doing this kind of analysis: 1) nobody can do it (you’re delusional if you believe we’re here) 2) some large organizations can do it if they really want to. 3) most places that want to do it right now can do it, slightly constrained by funding (I believe we are here) 4) absolutely anyone can do it to anyone at any time.
I don’t think data access availability shifts enough to move us from world 3 to world 4 when mythos hits. So...what are you worried about?

RedMan 12 Apr 2026 15:29 UTC
−1 points
−2
on: Pausing AI Is the Best Answer to Post-Alignment Problems
Has anyone written anything about the costs of pausing early? If the AI safety position on superintelligence eventually killing us all is correct, presumably there are points on the path to it that are better than others.
Is the best spot to pause in the past? If it’s in the future, what do we lose by stopping before we reach that point?
As I’ve written before, I think humans are on a glide path to extinction from non-AI causes. I think we are locked into a bunch of problems that require science and engineering solutions that are not currently available.
Pausing AI is likely pausing or rolling back technical development in general. I think the arguments for that leading to extinction long term are stronger than the arguments for superintelligence coming into being and instantly destroying the universe.

RedMan 9 Apr 2026 20:00 UTC
2 points
−5
on: Slightly-Super Persuasion Will Do
Seeking power over others seems like a zero sum game. You’re fighting everyone else on earth, including the people you want power over for some amount of power per human measured on a scale of no power, to 100% control. If AI levels the persuasion field, I’d expect a lot more people to enter the game, and thus the overall distribution of power to flatten. That sounds kind of cool.
Alternatively, an unelected oligarchy with access to the most capital intensive machinery ever created could try to use the outputs of that capital intensive machinery to gain and maintain power over many others, and make the global distribution of power over other humans look like one giant spike in the middle of a flat field?
This does not seem like a problem specific to AI.

RedMan 8 Apr 2026 20:02 UTC
−1 points
−2
in reply to: AlphaAndOmega’s comment on: Project Glasswing: Anthropic Shows The AI Train Isn’t Stopping
If you’re a company that wasn’t in on mythos, expect your stock to tank when it gets released. Building the tool and using it for the benefit of a self-selected elite is gross.

RedMan 8 Apr 2026 15:43 UTC
10 points
9
in reply to: 152334H’s comment on: Omne’s Shortform
An unelected oligarchy hoards access for themselves and uses it to make decisions about my life without my input. “These tools are too powerful for people like you”

RedMan 8 Apr 2026 4:47 UTC
3 points
0
in reply to: Thane Ruthenis’s comment on: Thane Ruthenis’s Shortform
This may ultimately be an edge for defense not offense. The status quo is that anyone with enough resources and enough programmers to throw at the task could have an exploit in whatever they wanted.
If the ‘most capable hacker on earth’ is an open model, it means that you will be able to reverse engineer your own equipment, build better rules for detection, actually do analysis of low level logs that would normally be too time consuming, and attackers would have no confidence that their exploits stay secret.
Compsci theory (weird machines) tells us that exploits exist in any sufficiently complicated codebase.
Targets who are too lazy to adopt technical measures like these are likely wide open today, so the increased vulnerability is irrelevant.

RedMan 7 Apr 2026 22:25 UTC
−1 points
0
on: Project Glasswing: Anthropic Shows The AI Train Isn’t Stopping
In the most annoying of all possible worlds, they held back some really nice bugs to sell to national militaries, and used their safeguards skills to ensure that users hunting for those bugs using the model are misled.

RedMan 6 Apr 2026 19:06 UTC
3 points
0
in reply to: GeneSmith’s comment on: The Practical Guide to Superbabies
Thanks a ton for the response, I think my partner and I are good candidates for Fertilo as trialed, and I’d be really grateful if you could DM me contact info for a someone who may want my business.

RedMan 5 Apr 2026 3:16 UTC
1 point
0
in reply to: RedMan’s comment on: My journey to the microwave alternate timeline
The browning dish seems to work as advertised, but it has not become a regular appliance. Some people in the thread mention the air fryer. I found a countertop combisteam (with the air fry option of course, the brand is FOTILE) can do everything I want better.
I am sad to admit that I have yet to make a better chicken than the one I get by sticking a whole chicken into the combisteam and using the ‘chicken’ setting.

RedMan 4 Apr 2026 19:47 UTC
3 points
0
on: The Practical Guide to Superbabies
Has anyone been able to actually access Fertilo anywhere in the world? Gameto hasn’t returned my emails, same goes for the clinic in peru that was publicized as having a successful birth.
I think I’m weird about the embryo selection ethics. My goal is for every produced embryo without “serious” defects to get a chance at life. If the kid won’t see himself as cursed by his genetics (osteogenesis imperfecta, for example), then whatever, let’s give the kid a shot. How often do parents basically set a cutoff (any embryo with t1d has a cursed life) and just roll with the first embryo over that cutoff?
I’d be mad if my parents had the option to gene edit me and declined it though.

RedMan 4 Apr 2026 2:29 UTC
1 point
0
on: Systematically dismantle the AI compute supply chain.
Data centers are being targeted with airborne bombs as we speak. So this discussion isn’t really theoretical.
Is the destruction of these data centers good, and would you advocate for actions that expand the present conflicts so that more data centers in more regions are struck?

RedMan 7 Mar 2026 20:00 UTC
3 points
0
in reply to: Wei Dai’s comment on: Wei Dai’s Shortform
I argued on this forum 9 years ago that human safety and alignment was unsolved: https://www.lesswrong.com/posts/dSXkTNe6NnyL5fX79/ai-safety-reading-group#rn26tdfASnGa8YGpc (and other posts)
I think ‘humans running the AGI’ can cause at least as much harm as ‘agentic AI acting on its’ own’.
I think most calls for slowdown and safety are based on a world model of the present day which is unreasonably friendly. If you believe (as I do) that completely ignoring issues around superintelligence, the human race may be (or is) currently on a glide path towards extinction and failure, your judgment about what constitutes acceptable risk in pursuit of technical solutions to our immediate problems will change.
I do believe that humans are not safe, and the status quo for the human race is unstable. If we collectively crash to a lower level of technology, I think the best case is probably worse than things were before modernity. If we keep pushing up the technological hills ahead of us, the view may look better.
I would like to see a better use of scarce energy and material resources than LLM training and inference come along. Hopefully we can get a step function increase in efficiency soon, and capital can be thrown at the next cool thing.

RedMan 4 Mar 2026 14:41 UTC
1 point
1
in reply to: 152334H’s comment on: AI found 12 of 12 OpenSSL zero-days (while curl cancelled its bug bounty)
If the discovery was from reviewing comments instead of reviewing code, that’s unfortunate and makes me less optimistic about the technology. A conspiracy theorist can always add complexity to keep a nefarious explanation alive. I don’t think that there’s any evidence of nefarious activity by that engineer so I hope he doesn’t suffer personally or professionally.
I stand by my analysis, this is the type of bug that would be useful to a nation state, assuming it is actually exploitable.

RedMan 28 Feb 2026 20:35 UTC
1 point
0
in reply to: ryan_greenblatt’s comment on: Anthropic: “Statement from Dario Amodei on our discussions with the Department of War”
Thanks for replying, I think I was thinking of this post: https://www.lesswrong.com/posts/FG54euEAesRkSZuJN/ryan_greenblatt-s-shortform?commentId=B6oDGoyphuNuzdDAT I didn’t understand the distinction between employee and helpful only access. Can you elaborate on the difference between employee access, public access, and helpful only access?

RedMan 28 Feb 2026 7:20 UTC
1 point
0
on: Anthropic: “Statement from Dario Amodei on our discussions with the Department of War”
Absolutely love this trend of all sorts of different people telling the US Government, ‘do the thing you’re threatening’.
Some people here @ryan_greenblatt (?) have claimed access to uncensored anthropic models at anthropic, maybe they can chime in about what exactly that means. If safeguards are baked into the model early enough in the training process that removing them requires substantial reengineering and maybe additional training runs, then maybe anthropic simply isn’t capable of complying.
If safeguards are introduced post training, anthropic has an uncensored model available. Assuming there isn’t another vendor, or that they want to make a super public demonstration of their power, the US Government has a ton of methods to compel compliance. They can demand it under the DPA, or plausibly use that designation as justification for applying countermeasures to the company. They may even just steal it outright...or if someone else steals the base model, steal it from them, relabel it, and deploy it internally.
There’s always the possibility that this is just a face saving move, where anthropic has a tiny team in the company and a classified contract, where they just hand over the uncensored model and get to keep their public face intact. Under normal circumstances, I’d assume that DoW would just go with a different vendor, but who knows with this administration, they seem really averse to being publicly contradicted.
DoW may also believe that they were being conciliatory and kind by allowing anthropic to add ‘any lawful use’ safeguards instead of just demanding the base model without post training safeguards.

RedMan 25 Feb 2026 21:26 UTC
2 points
0
in reply to: Matrice Jacobine’s comment on: Tom Smith’s Shortform
Uncensored models being available to a self-selected elite, and the rest of us getting whatever those elites decide to give us after censorship is more dangerous than giving uncensored models to everyone. AI gatekeeping in the guise of “safety” is going to lead to tangible immediate harms.

RedMan 25 Feb 2026 16:13 UTC
7 points
0
in reply to: Random Developer’s comment on: Tom Smith’s Shortform
Manipulating the money printer, government contracts, and law enforcement should make it possible for the US government to seize all the hardware and use it however they please if it comes to that. Some judge might approve strategic production rules applying to inference or training runs, but that may not even be necessary. The USG generally tries to avoid being that aggressive and authoritarian, for the same reason that the FBI agent does not pull his gun and point it at you when he wants to talk. Pretending that the implied threat isn’t present is both silly and common.
A hyperscaler isn’t like a central bank where it ceases to be fit for purpose if it loses its’ independence.

RedMan 25 Feb 2026 2:31 UTC
1 point
0
on: Citrini’s Scenario Is A Great But Deeply Flawed Thought Experiment
Has anyone considered the following?
AI companies eat the economy, revenue never appears, AI companies become insolvent, Federal government prints money, takes ownership of data centers, prints more money, uses totality of formerly fragmented AI infrastructure to train its’ own AI, which it uses to dominate the world. Welfare state using surplus value created by ai government is used to buy votes.

RedMan 25 Feb 2026 1:35 UTC
4 points
0
on: Large-Scale Online Deanonymization with LLMs
Obligatory: https://terrytao.wordpress.com/about/anonymity-and-the-internet/
31 bits of anonymity are a maximum. Users do and say things that reduce this. 31 bits shouldn’t be hard to attack with a lot of compute, so...this should not be unexpected.

RedMan 18 Feb 2026 12:06 UTC
1 point
−1
on: You’re an AI Expert – Not an Influencer
Let’s think about the opposite strategy.
Would broadcasting beliefs that are controversial but extremely popular with a small minority, then also broadcasting opinions about AI safety move that minority towards agreement with you on AI safety? Would a concerted and coordinated influence campaign doing this, where you said things like ‘I hate alpine worms, just like you, now please form a positive opinion on this AI safety professional who only talks about that issue’ work? What if two people screamed at each other about their strong disagreement about alpine worms, then accused each other of ‘not taking AI safety seriously’ and persuaded their respective followers to take AI safety as a good thing that belongs to their side?