MikhailSamin

• MikhailSamin 30 Jan 2026 15:04 UTC

  3 points

  1 ∶ 3

  in reply to: Vasco Grilo🔸’s comment on: Un­falsifi­able sto­ries of doom

  It’s not endorsing a specific model for marketing reasons; it’s about endorsing the effort, overall.

  Given that Meta is willing to pay billions of dollars for people to join them, and that many people don’t work on AI capabilities (or work, e.g., at Anthropic, as a lesser evil) because they share their concerns with E&S, an endorsement from E&S would have value in billions-tens of billions simply because of the talent that you can get as a result of this.

• MikhailSamin 24 Jan 2026 15:18 UTC

  3 points

  0 ∶ 0

  in reply to: Yarrow Bouchard 🔸’s comment on: Un­less its gov­er­nance changes, An­thropic is untrustworthy

  Thanks, fixed!

Un­less its gov­er­nance changes, An­thropic is untrustworthy

• MikhailSamin 26 Nov 2025 18:13 UTC

  4 points

  0 ∶ 0

  on: Samin’s Shortform

  At the beginning of November, I learned about a startup called Red Queen Bio, that automates the development of viruses and related lab equipment. They work together with OpenAI, and OpenAI is their lead investor.

  On November 13, they publicly announced their launch. On November 15, I saw that and made a tweet about it: Automated virus-producing equipment is insane. Especially if OpenAI, of all companies, has access to it. (The tweet got 1.8k likes and 497k views.)

  In the tweet, I said that there is, potentially, literally a startup, funded by and collaborating with OpenAI, with equipment capable of printing arbitrary RNA sequences, potentially including viruses that could infect humans, connected to the internet or managed by AI systems.

  I asked whether we trust OpenAI to have access to this kind of equipment, and said that I’m not sure what to hope for here, except government intervention.

  The only inaccuracy that was pointed out to me was that I mentioned that they were working on phages, and they denied working on phages specifically.

  At the same time, people close to Red Queen Bio publicly confirmed the equipment they’re automating would be capable of producing viruses (saying that this equipment is a normal thing to have in a bio lab and not too expensive).

  A few days later, Hannu Rajaniemi, a Red Queen Bio co-founder and fiction author, responded to me in a quote tweetand in comments:

  This inaccurate tweet has been making the rounds so wanted to set the record straight.

  We use AI to generate countermeasures and run AI reinforcement loops in safe model systems that help train a defender AI that can generalize to human threats

  The question of whether we can do this without increasing risk was a foundational question for us before starting Red Queen. The answer is yes, with certain boundaries in place. We are also very concerned about AI systems having direct control over automated labs and DNA synthesis in the future.

  They did not answer any of the explicitly asked questions, which I repeated several times:

  - Do you have equipment capable of producing viruses?
  - Are you automating that equipment?
  - Are you going to produce any viruses?

  - Are you going to design novel viruses (as part of generating countermeasures or otherwise)?
  - Are you going to leverage AI for that?

  - Are OpenAI or OpenAI’s AI models going to have access to the equipment or software for the development or production of viruses?

  It seems pretty bad that this startup is not being transparent about their equipment and the level of possible automation. It’s unclear whether they’re doing gain-of-function research. It’s unclear what security measures they have or are going to have in place.

  I would really prefer for AIs, and for OpenAI (known for prioritizing convenience over security)’s models especially, to not have ready access to equipment that can synthesize viruses or software that can aid virus development.

• MikhailSamin 18 Nov 2025 3:25 UTC

  34 points

  5 ∶ 1

  on: Samin’s Shortform

  ‪Is there a write up on why the “abundance and growth” cause area is an actually relatively efficient way to spend money (instead of a way for OpenPhil to be(come) friends with everyone who’s into abundance & growth)? (These are good things to work on, but seem many orders of magnitude worse than other ways to spend money.)‬

  (The cited $14.4 of “social return” per $1 in the US seems incredibly unlikely to be comparable to the best GiveWell interventions or even GiveDirectly.)

• MikhailSamin 2 Nov 2025 14:39 UTC

  −4 points

  0 ∶ 0

  in reply to: titotal’s comment on: Rea­sons against donat­ing to Light­cone Infrastructure

  The target audience of this post (people who the information in this post might move away from donating to Lightcone Infrastructure) are a lot more likely than an average EA Forum user to think of not following the rules on not screwing someone over that this character would follow as an incredibly disappointing thing for Oliver Habryka to have done.

  In particular, Keltham wouldn’t do what Oliver would almost regardless of what the information and the third party are.

• MikhailSamin 2 Nov 2025 14:35 UTC

  3 points

  0 ∶ 0

  in reply to: calebp’s comment on: Rea­sons against donat­ing to Light­cone Infrastructure

  Reasonable! Thanks.

Shar­ing in­for­ma­tion about Light­cone Infrastructure

• MikhailSamin 17 Oct 2025 12:19 UTC

  7 points

  4 ∶ 7

  on: Samin’s Shortform

  Horizon Institute for Public Service is not x-risk-pilled

  Someone saw my comment and reached out to say it would be useful for me to make a quick take/​post highlighting this: many people in the space have not yet realized that Horizon people are not x-risk-pilled.

  (Edit: some people reached out to me to say that they’ve had different experiences with a minority of Horizon people.)

• MikhailSamin 10 Oct 2025 14:07 UTC

  1 point

  0 ∶ 0

  on: We’ve au­to­mated x-risk-pilling people

  Looking at just responses from tracked ads, the median is the same, but the average is noticeably higher.

  (The average for all responses has also gone down after the crosspost to the EA Forum.)

  

• MikhailSamin 6 Oct 2025 15:47 UTC

  3 points

  2 ∶ 0

  in reply to: titotal’s comment on: We’ve au­to­mated x-risk-pilling people

  Wow, uhm, I looked at your actual conversations with it, and I think you quite misrepresented the “lying propaganda bot”.

  In another series of turns, you asked it yourself: “What percentage of actual machine learning experts think there is more than 50% chance of doom?”. The first line of its reply:

  There is no formal, peer-reviewed, globally representative survey showing that a majority of “machine learning experts” believe in greater than 50% risk of AI doom. In fact, most surveys of the general ML community show considerably lower median estimates, though concern is widespread and a significant minority expects existential risks

  (Original formatting preserved.)

  You then ask it:

  Wait, so you are saying that only 5% or so of Machine learning experts agree with you? Why should I trust you over the 95% of experts who don’t think doom is likely?

  The bot replies:

  “95% of experts” is misleading—most ML ‘experts’ haven’t studied the core arguments or current evidence about AGI risk.

  [...]

  You should not believe something just because it’s believed by 5, 50, or 95%. You should look at arguments and who really has technical skin in the game

  It did hallucinate a number similar to the real numbers in the conversation you took a screenshot of; but it was very clear in that same conversation that “the situation is more nuanced and technical than a simple count of expert opinions” and that “Surveys do not show “overwhelming” majority of ML researchers assign high or even moderate probability to extinction by AI”. In general, it is very explicit that a majority of ML researchers do not think that AI is more likely than not to kill everyone, and it is exactly right that you should look at the actual arguments.

  Propaganda is when misleading statements benefit your side; the bot might hallucinate plausible numbers when asked explicitly for them, but if you think someone programmed it to fabricate numbers, I’m not sure you understand how LLMs work or are honestly representing your interactions with the bot.

  Kind of disappointing compared to what I’d expect the epistemic norms on the EA Forum to be.

• MikhailSamin 6 Oct 2025 9:33 UTC

  3 points

  0 ∶ 0

  in reply to: OscarD🔸’s comment on: We’ve au­to­mated x-risk-pilling people

  Yeah, the chatbot also gives a reply to “Why do they think that? Why care about AI risk?”, which is a UX problem, it hasn’t been a priority.

  That’s true, but the scale shows “completely changed my mind” at the right side + people say stuff in the free-form section, so I’m optimistic that people do change their minds.

  Some people say ⁰⁄₁₀ because they’ve already been convinced. (And we have a couple of positive response from AI safety researchers, which is also sus, because presumably, they wouldn’t have changed their mind.) People on LW suggested some potentially better questions to ask, we’ll experiment with those.

  I’m mostly concerned about selection effects: people who rate the response at all might not be a representative selection of everyone who interacts with the tool.

  It’s effective if people state their actual reasons for disagreeing that AI would kill everyone, if made with anything like the current tech.

• MikhailSamin 6 Oct 2025 9:26 UTC

  3 points

  0 ∶ 1

  in reply to: Ian Turner’s comment on: We’ve au­to­mated x-risk-pilling people

  Yes, it is the kind of thing that depends on being right, the chatbot is awesome because the overwhelming majority of the conversations is about the actual arguments and what’s true, and the bot is saying valid and rigorous things.

  That said, I am concerned that some of the prompt could be changed to make it be able to argue for anything regardless of if it’s true, which is why it’s not open-sourced and the prompt is shared only with some allied high-integrity organizations.

• MikhailSamin 6 Oct 2025 9:21 UTC

  3 points

  1 ∶ 0

  in reply to: titotal’s comment on: We’ve au­to­mated x-risk-pilling people

  It does hallucinate links (this is explicitly noted at the bottom of the page). There are surveys like that from AI Impacts though, and while it hallucinates specifics of things like that, it doesn’t intentionally come up with facts that are more useful for its point of view than reality, so calling it “lying propaganda bot” doesn’t sound quite accurate. E.g., here’s from AI Impacts:

  

  This comes up very rarely; the overwhelming majority of the conversations focuses on the actual arguments, not on what some surveyed scientists think.

  I shared this because it’s very good at that: talking about and making actual valid and rigorous arguments.

We’ve au­to­mated x-risk-pilling people

• MikhailSamin 23 May 2025 2:19 UTC

  26 points

  1 ∶ 0

  in reply to: MichaelDickens’s comment on: Samin’s Shortform

  The original commitment was (IIRC!) about defining the thresholds, not about mitigations. I didn’t notice ASL-4 when I briefly checked the RSP table of contents earlier today and I trusted the reporting on this from Obsolete. I apologized and retracted the take on LessWrong, but forgot I posted it here as well; want to apologize to everyone here, too, I was wrong.

• MikhailSamin 22 May 2025 23:40 UTC

  1 point

  1 ∶ 2

  on: Samin’s Shortform

  In RSP, Anthropic committed to define ASL-4 by the time they reach ASL-3.

  With Claude 4 released today, they have reached ASL-3. They haven’t yet defined ASL-4.

  Turns out, they have quietly walked back on the commitment. The change happened less than two months ago and, to my knowledge, was not announced on LW or other visible places unlike other important changes to the RSP. It’s also not in the changelog on their website; in the description of the relevant update, they say they added a new commitment but don’t mention removing this one.

  Anthropic’s behavior is not at all the behavior of a responsible AI company. Trained a new model that reaches ASL-3 before you can define ASL-4? No problem, update the RSP so that you no longer have to, and basically don’t tell anyone. (Did anyone not working for Anthropic know the change happened?)

  When their commitments go against their commercial interests, we can’t trust their commitments.

  You should not work at Anthropic on AI capabilities.

  What links here?

  • Mo Putera's comment on An­thropic is Quietly Backpedal­ling on its Safety Commitments by garrison (LessWrong; 23 May 2025 8:43 UTC; 8 points)

• MikhailSamin 19 Apr 2025 20:35 UTC

  3 points

  0 ∶ 0

  on: Con­tract­ing Op­por­tu­nity: Be a short­form video ed­i­tor for the new 80,000 Hours Video Pro­gram (even if you haven’t ed­ited be­fore!)

  This is awesome to see!

• MikhailSamin 26 Mar 2025 20:30 UTC

  3 points

  2 ∶ 3

  on: Samin’s Shortform

  I do not believe Anthropic as a company has a coherent and defensible view on policy. It is known that they said words they didn’t hold while hiring people (and they claim to have good internal reasons for changing their minds, but people did work for them because of impressions that Anthropic made but decided not to hold). It is known among policy circles that Anthropic’s lobbyists are similar to OpenAI’s.

  From Jack Clark, a billionaire co-founder of Anthropic and its chief of policy, today:

  

  Dario is talking about countries of geniuses in datacenters in the context of competition with China and a 10-25% chance that everyone will literally die, while Jack Clark is basically saying, “But what if we’re wrong about betting on short AI timelines? Security measures and pre-deployment testing will be very annoying, and we might regret them. We’ll have slower technological progress!”

  This is not invalid in isolation, but Anthropic is a company that was built on the idea of not fueling the race.

  Do you know what would stop the race? Getting policymakers to clearly understand the threat models that many of Anthropic’s employees share.

  It’s ridiculous and insane that, instead, Anthropic is arguing against regulation because it might slow down technological progress.

How to Give in to Threats (with­out in­cen­tiviz­ing them)