MichaelDickens

Karma: 5,895

I do independent research on EA topics. I write about whatever seems important, tractable, and interesting (to me).

I have a website: https://mdickens.me/ Much of the content on my website gets cross-posted to the EA Forum, but I also write about some non-EA stuff like [investing](https://mdickens.me/category/finance/) and [fitness](https://mdickens.me/category/fitness/).

My favorite things that I’ve written: https://mdickens.me/favorite-posts/

I used to work as a software developer at Affirm.

MichaelDickens Jun 1, 2025, 5:09 AM
8 points
0 ∶ 0
in reply to: Chris Leong’s comment on: Matthew_Barnett’s Shortform
Yes I think that is better than what I said, both because it’s more accurate, and because it’s more clear that Matthew did in fact use his knowledge of capability trends to decide that he could profit from starting an AI company.

Like, I don’t know what exactly went into his decision, but I would be surprised if that knowledge didn’t play a role.

Arguably that’s less on Matthew and more on the founders of Epoch for either misrepresenting themselves or having a bad hiring filter. Probably the former—if I’m not mistaken, Tamay Besiroglu co-founded Epoch and is now co-founding Mechanize, so I would say Tamay behaved badly here but I’m not sure whether Matthew did.

MichaelDickens Jun 1, 2025, 2:06 AM
22 points
6 ∶ 4
in reply to: Matthew_Barnett’s comment on: Matthew_Barnett’s Shortform

I have been publicly labeled a “sellout and traitor” on X by a prominent member of the EA community simply because I cofounded an AI startup.

This accusation was not because you cofounded an AI startup. It was specifically because you took funding to work on AI safety from people who want to ~~slow down AI development~~ use capability trends to better understand how to make AI safer*, and you are now (allegedly) using results developed from that funding to start a company dedicated to accelerating AI capabilities.

I don’t know exactly what results Mechanize is using, but if this is true, then it does indeed constitute a betrayal. Not because you’re accelerating capabilities, but because you took AI safety funding and used the results to do the opposite of what funders wanted.

*Corrected to give a more accurate characterization, see Chris Leong’s comment

MichaelDickens May 28, 2025, 6:10 PM
4 points
0 ∶ 0
on: Can you donate to AI advocacy
Zvi’s Big Nonprofits Post lists a bunch of advocacy orgs.

Last year I wrote about every policy/advocacy org that I could find at the time, although the list is now somewhat out of date, e.g. The Midas Project (which Ben West mentioned) did not exist yet.

I was in a similar position to you—I wanted to donate to AI policy or advocacy, but as far as I know there aren’t any grantmakers focusing on advocacy. You shouldn’t necessarily trust that I did a good job of evaluating organizations but maybe my list can give you ideas.

MichaelDickens May 26, 2025, 9:36 PM
1 point
0 ∶ 1
in reply to: Jacob Watts🔸’s comment on: Jacob Watts’s Quick takes
Alternative idea: AI companies should have a little checkbox saying “Please use 100% of the revenue from my subscription to fund safety research only.” This avoids some of the problems with your idea and also introduces some new problems.

I think there is a non-infinitesimal chance that Anthropic would actually implement this.

MichaelDickens May 23, 2025, 2:01 PM
6 points
0 ∶ 0
in reply to: dirk’s comment on: dirk’s Quick takes
Hard to know the full story but for me this is a weak update against CAIS’s judgment.

Right now CAIS is one out of a total of maybe two orgs (along with FLI) pushing for AI legislation that both (1) openly care about x-risk and (2) are sufficiently Respectable TM* to get funding from big donors. This move could be an attempt to maintain CAIS’s image as Respectable TM. My guess is it’s the wrong move but I have a lot of uncertainty. I think firing people due to public pressure is generally a bad idea although I’m not confident that that’s what actually happened.

*I hope my capitalization makes this clear but to be explicit, I don’t think Respectable TM is the same thing as “actually respectable”. For example, MIRI is actually respectable, but isn’t Respectable TM.

Edit: I just re-read the CAIS tweet, from the wording it is clear that CAIS meant one of two things: “We are bowing to public pressure” or “We didn’t realize John Sherman would say things like this, and we consider it a fireable offense”. Neither one is a good look IMO.

MichaelDickens May 23, 2025, 2:12 AM
2 points
0 ∶ 1
in reply to: evhub’s comment on: Samin’s Shortform
The RSP specifies that CBRN-4 and AI R&D-5 both require ASL-4 security. Where is ASL-4 itself defined?

MichaelDickens May 14, 2025, 8:06 PM
8 points
0 ∶ 0
on: MichaelDickens’s Shortform
What can ordinary people do to reduce AI risk? People who don’t have expertise in AI research / decision theory / policy / etc.

Some ideas:
- Donate to orgs that are working to AI risk (which ones, though?)
- Write letters to policy-makers expressing your concerns
- Be public about your concerns. Normalize caring about x-risk

MichaelDickens May 13, 2025, 3:55 AM
4 points
0 ∶ 0
in reply to: Joseph’s comment on: Managing Your Money 101
1. This portfolio has nothing to do with chasing past performance. According to standard finance theory (including making a bunch of assumptions about rational investors, zero transaction costs etc.), the global market portfolio is the theoretically optimal portfolio to hold. The idea is that if markets are efficient then you can’t predict which asset classes will outperform, so you should just hold some of everything.
2. This portfolio doesn’t require any rebalancing. It’s the global market portfolio (or it was as of 2015). If you have 18% in US stocks because they represent 18% of the global market portfolio, and then US stocks go up to 20%, your holdings also go up to 20% so you don’t have to do anything. Although it might still take some work to manage if you’re adding more money to the portfolio on a monthly basis (or whatever) because you need to deploy your new investment in the correct proportions.
3. Realistically you can get pretty close to the global market portfolio by buying global stocks + global bonds and not worry about the smaller positions. You can do a 2-fund portfolio with 50% VT, 50% BNDW.
4. 1973 to 2013 isn’t arbitrary. 1973 was the chosen start year because that’s the earliest date at which we have good data on global equity returns.

MichaelDickens May 11, 2025, 3:29 PM
4 points
0 ∶ 0
in reply to: Clara Torres Latorre 🔶️’s comment on: Clara Torres Latorre ️’s Quick takes
Ah I see what you’re saying. I can’t recall seeing much discussion on this. My guess is that it would be hard to develop a non-superintelligent AI that poses an extinction risk but I haven’t really thought about it. It does sound like something that deserves some thought.

When people raise particular concerns about powerful AI, such as risks from synthetic biology, they often talk about them as risks from general AI, but they could come from narrow AI, too. For example some people have talked about the risk that narrow AI could be used by humans to develop dangerous engineered viruses.

MichaelDickens May 10, 2025, 7:14 PM
4 points
0 ∶ 0
in reply to: Clara Torres Latorre 🔶️’s comment on: Clara Torres Latorre ️’s Quick takes
My take is that I think there are strong arguments for why AI x-risk is overwhelmingly more important than narrow AI, and I think those arguments are the main reason why x-risk gets more attention among EAs.

MichaelDickens May 10, 2025, 1:43 AM
6 points
5 ∶ 0
in reply to: Guy Raveh’s comment on: Chicken welfare reforms may be harmful accounting for effects on wild arthropods?
I agree with David’s comment. These sorts of ethical dilemmas are puzzles for everyone, not just for utilitarianism.

And in the case of insect welfare, rights-based theories produce more puzzling puzzles because it’s unclear how to reckon with tradeoffs.

MichaelDickens May 9, 2025, 7:45 PM
13 points
3 ∶ 3
in reply to: LeahC’s comment on: Please Donate to CAIP (Post 1 of 3 on AI Governance)
There is a related concern where most of the big funders either have investments in AI companies, or have close ties to people with investments in AI companies. This biases them toward funding activities that won’t slow down AI development. So the more effective an org is at putting the brakes on AGI, the harder a time it will have getting funded.*

Props to Jaan Tallinn, who is an early investor in Anthropic yet has funded orgs that want to slow down AI (including CAIP).

*I’m not confident that this is a factor in why CAIP has struggled to get funding, but I wouldn’t be surprised if it was.

MichaelDickens May 9, 2025, 1:39 PM
14 points
1 ∶ 2
in reply to: emre kaplan🔸’s comment on: emre kaplan’s Quick takes

In general, writing criticism feels more virtuous than writing praise.

FWIW it feels the opposite to me. Writing praise feels good; writing criticism feels bad.

(I guess you could say that it’s virtuous to push through those bad feelings and write the criticism anyway? I don’t get any positive feelings or self-image from following that supposed virtue, though.)

MichaelDickens May 7, 2025, 3:01 PM
8 points
2 ∶ 0
on: Why “Solving Alignment” Is Likely a Category Mistake
I think this is an important point that’s worth saying.

For what it’s worth, I am not super pessimistic that “solving alignment” is something that can be solved in principle. But I’m quite concerned that the safety-minded AI companies seem to completely ignore the philosophical problems with AI alignment. They all operate under the assumption that alignment is purely an ML problem and they can solve it by basically doing ML research. Which I expect is false (credence: 70%).

Wei Dai has written some good stuff about the problem of “philosophical competence”. See here for a collection of his writings on the topic.
What links here?
- Will Aldred's comment on The Soul of EA is in Trouble by Mjreard (May 9, 2025, 10:03 AM; 13 points)

MichaelDickens May 2, 2025, 4:33 AM
2 points
0 ∶ 0
in reply to: Denkenberger🔸’s comment on: Poll: the next existential catastrophe is likelier than not to wipe off all animal sentience from the planet
This is a good point that I hadn’t thought of when I wrote my poll answer—a “gradual disempowerment” risk scenario would probably not kill all sentient animals, and it represents a non-trivial percentage of AI risk.

MichaelDickens May 2, 2025, 4:32 AM
6 points
1 ∶ 0
on: Poll: the next existential catastrophe is likelier than not to wipe off all animal sentience from the planet

The next existential catastrophe is likelier than not to wipe off all animal sentience from the planet

If an existential catastrophe occurs, it will probably (~90%) be AI, and an AI that kills all humans would probably also (~80%) kill all sentient animals.

The argument against killing all animals is that they are less likely than humans to interfere with the AI’s goals. The argument in favor is that they are made of atoms.

Edit: Updated downward a bit based on Denkenberger’s comment.

MichaelDickens May 1, 2025, 5:06 PM
5 points
1 ∶ 0
on: Community Polls for the Community
Some invitees to the Meta Coordination Forum (maybe like 3 out of the ~30) should be ‘independent’ EAs
Independent as in not affiliated with any org? If that’s what it means then I probably agree

MichaelDickens May 1, 2025, 5:05 PM
4 points
1 ∶ 2
on: Community Polls for the Community
There exists a cause which ought to receive >20% of the EA community’s resources but currently receives little attention
My top picks for small causes that should maybe receive >20% of resources:
- wild animal welfare
- post-AGI non-human welfare

MichaelDickens May 1, 2025, 5:03 PM
5 points
3 ∶ 0
on: Community Polls for the Community
The case for doing EA community building hinges on having significant probability on ‘long’ (>2040) AI timelines
>2040, no. >2030, yes.

MichaelDickens May 1, 2025, 5:01 PM
2 points
0 ∶ 0
on: Community Polls for the Community
We should try to make some EA sentiments and principles (e.g., scope sensitivity, thinking hard about ethics) a core part of the AI safety field
On a literal interpretation of this statement, I disagree, because I don’t think trying to inject those principles will be cost-effective. But I do think people should adopt those principles in AI safety (and also in every other cause area).