Davidmanheim

Mis­nam­ing and Other Is­sues with OpenAI’s “Hu­man Level” Su­per­in­tel­li­gence Hierarchy

• Davidmanheim 30 Jun 2024 14:59 UTC

  13 points

  2 ∶ 0

  on: How to re­al­is­ti­cally re­duce most suffer­ing on earth

  This makes a number of non-trivial assumption and unsourced claims about a number of different issues, from relative moral value of animals to the carrying capacity of different biomes; I know that many of these are seen as common wisdom in EA, but I think failing to lay them out greatly weakens the conclusions.

  Also, some questions to think about: Why are insects ignored? How does the transition happen, legally or economically? What are the impacts of land use changes, and do farmers sell the land? (To whom?) Do social norms around meat undermine the viability of a transition?

• Davidmanheim 16 Jun 2024 10:05 UTC

  2 points

  0 ∶ 0

  on: Safety isn’t safety with­out a so­cial model (or: dis­pel­ling the myth of per se tech­ni­cal safety)

  {making humanity more safe VS shortening AGI timelines} is itself a false dichotomy or false spectrum.

  Why? Because in some situations, shortening AGI timelines could make humanity more safe, such as by avoiding an overhang of over-abundant computing resources that AGI could abruptly take advantage of if it’s invented too far in the future (the “compute overhang” argument).

  
  I think this also ignores the counterfactual world with less safety research, where the equivalent advances, which are funded because of commercial incentives, come from less generalizable safety research, and we end up with less well prosaically aligned but similarly capable systems. (And I haven’t really laid out this argument before, but I think it generalizes to the counterfactual world without OpenAI or even Deepmind being inspired by AI safety concerns.)

• Davidmanheim 9 Jun 2024 6:48 UTC

  3 points

  0 ∶ 0

  in reply to: Richard Möhn’s comment on: How can I donate to biorisk pre­ven­tion?

  I think it’s fine, if orgs are set up for getting the donations. If they have a “donate” button or page, they are set up to get the money, less credit card fees, etc. They problem is that setting up that sort of thing is anywhere from easy to legally very complex.

• Davidmanheim 5 Jun 2024 13:32 UTC

  5 points

  0 ∶ 0

  on: How can I donate to biorisk pre­ven­tion?

  As someone who runs an organization that does a lot of biorisk work, it’s incredibly expensive in staff time and logistics to receive small donations—but if you’re giving more than, say, $5,000, you could just email the organizations to ask, and I’m sure they could figure it out.
  
  But as I answered, CHS does have a donation page. (And NTI does allow donations, with a box to indicate where you’d like the money to go, but it’s unclear to me if that actually lets you direct it only to bio.)

• Davidmanheim 5 Jun 2024 13:27 UTC

  3 points

  2 ∶ 0

  on: How can I donate to biorisk pre­ven­tion?

  https://​​centerforhealthsecurity.org/​​make-a-gift-to-the-center-for-health-security

• Davidmanheim 28 May 2024 10:02 UTC

  28 points

  8 ∶ 0

  on: Effec­tive Altru­ism In­fras­truc­ture Fund: March 2024 recommendations

  Any chance we’ll see some evaluation of prior grants?

ALTER Is­rael − 2024 Mid-Year Update

• Davidmanheim 22 May 2024 9:18 UTC

  17 points

  2 ∶ 0

  in reply to: DC’s comment on: A tale of two Sams

  LLMs are not AGIs in the sense being discussed, they are at best proto-AGI. That means the logic fails at exactly the point where it matters.
  
  When I ask a friend to give me a dollar when I’m short, they often do so. Is this evidence that I can borrow a billion dollars? Should I go on a spending spree on the basis that I’ll be able to get the money to pay for it from those friends?
  
  When I lift, catch, or throw a 10 pound weight, I usually manage it without hurting myself. Is this evidence that weight isn’t an issue? Should I try to catch a 1,000 pound boulder?

• Davidmanheim 22 May 2024 8:57 UTC

  3 points

  0 ∶ 0

  on: Com­mon re­but­tal to “paus­ing” or reg­u­lat­ing AI

  No one is really suggesting that a unilateral “pause” is effective, but there is growing support for some non-unilateral version as an important approach to be negotiated.
  
  There was a quite serious discussion of the question, and different views, on the forum late last year (which I participated in,) summarized by Scott Alexander here; https://​​forum.effectivealtruism.org/​​posts/​​7WfMYzLfcTyDtD6Gn/​​pause-for-thought-the-ai-pause-debate

• Davidmanheim 21 May 2024 10:19 UTC

  11 points

  2 ∶ 0

  in reply to: Nathan Young’s comment on: Call for At­tor­neys for OpenAI Em­ploy­ees and Ex-Employees

  Confirmed; he does work in this area, there’s independent reporting about his work on these topics, and has a substack about his very relevant legal work; https://​​www.nlrbedge.com/​​

• Davidmanheim 20 May 2024 5:11 UTC

  4 points

  0 ∶ 0

  on: Call for At­tor­neys for OpenAI Em­ploy­ees and Ex-Employees

  Do you have any comment on the idea that nondisparagement clauses like this could be found invalid for being contrary to public policy? (How would that be established?)

• Davidmanheim 8 May 2024 19:50 UTC

  2 points

  0 ∶ 0

  in reply to: Sebastian Kreisel’s comment on: Biorisk is an Un­helpful Anal­ogy for AI Risk

  I think there are useful analogies between specific aspects of bio, cyber, and AI risks, and it’s certainly the case that when the biorisk is based on information security, it’s very similar to cybersecurity, not the least in that it requires cybersecurity! And the same is true for AI risk; to the extent that there is a risk of model weights leaking, this is in part a cybersecurity issue.
  
  So yes, I certainly agree that many of the dissimilarities with AI are not present if analogizing to cyber. However, more generally, I’m not sure cybersecurity is a good analogy for biorisk, and have heard that computer security people often dislike the comparison of computer viruses and biological viruses for that reason, though they certainly share some features.

Biorisk is an Un­helpful Anal­ogy for AI Risk

• Davidmanheim 2 May 2024 4:02 UTC

  6 points

  2 ∶ 0

  in reply to: Jason’s comment on: Per­sonal re­flec­tions on FTX

  Gutting the FTT token is customers losing money because of their investing, not customer losses via FTX loss of custodial funds or token, though, isn’t it?

• Davidmanheim 2 May 2024 3:51 UTC

  10 points

  1 ∶ 0

  in reply to: Jason’s comment on: Per­sonal re­flec­tions on FTX

  Alameda exile told Time that SBF “didn’t have a distinction between firm capital and trading capital. It was all one pool.” That’s at least a badge of fraud (commingling)

  Alameda was a prop trading firm, so there isn’t normally any distinction between those. The only reason this didn’t apply was that there was a third bucket of funds, pass-through custodial funds that belonged to FTX customers, which they evidently didn’t pass through due to poor record keeping. That’s not as much indicative of fraud, it’s indicative of incompetance.

• Davidmanheim 10 Apr 2024 6:23 UTC

  2 points

  0 ∶ 0

  in reply to: Jason’s comment on: Some un­der­rated rea­sons why the AI safety com­mu­nity should re­con­sider its em­brace of strict li­a­bil­ity

  Yes, I see a strong argument for the claim that the companies are in the best position to shoulder the harms that will inevitably come along, and pass that risk onto their customers through higher prices—but the other critical part is that this also changes incentives because liability insurers will demand the firms mitigate the risks. (And this is approaching the GCR argument, from a different side.)

• Davidmanheim 10 Apr 2024 6:21 UTC

  2 points

  0 ∶ 0

  in reply to: Jason’s comment on: Some un­der­rated rea­sons why the AI safety com­mu­nity should re­con­sider its em­brace of strict li­a­bil­ity

  I think that the use of insurance for moderate harms is often a commercial boondoggle for insurers, a la health insurance, which breaks incentives in many ways an leads to cost disease. And typical insurance regimes shift burden of proof about injury in damaging ways because insurers have deep pockets to deny claims in court and fight cases that establish precedents. I also don’t think that it matters for tail risks—unless explicitly mandating unlimited coverage, firms will have caps in the millions of dollars, and will ignore tail risks that will bankrupt them.

  One way to address the tail, in place of strict liability, would be legislation allowing anticipated harms to be stopped via legal action, as opposed to my understanding that pursuing this type of prior restraint for uncertain harms isn’t possible in most domains.

  I’d be interested in your thoughts on these points, as well as Cecil and Marie’s.

• Davidmanheim 10 Apr 2024 6:10 UTC

  2 points

  0 ∶ 0

  on: Some un­der­rated rea­sons why the AI safety com­mu­nity should re­con­sider its em­brace of strict li­a­bil­ity

  I would be interested in understanding whether you think that joint-and-several liability among model training, model developers, application developers, and users would address many of the criticisms you point out against civil liability. As I said last year, “joint-and-several liability for developers, application providers, and users for misuse, copyright violation, and illegal discrimination would be a useful initial band-aid; among other things, this provides motive for companies to help craft regulation to provide clear rules about what is needed to ensure on each party’s behalf that they will not be financially liable for a given use, or misuse.”
  
  I also think that this helps mitigate the issue with fault-based liability in proving culpability, but I’m agnostic about which liability regime is justified.
  
  Lastly, I think that your arguments mean that there’s good reason to develop a clear proposal for some new liability standard, perhaps including requirements for uncapped liability insurance for some specific portion of eventual damages, rather than assume that the dichotomy of strict vs. fault based is immutable.

• Davidmanheim 1 Apr 2024 19:08 UTC

  4 points

  1 ∶ 0

  in reply to: Ariel Simnegar 🔸’s comment on: Ex­cerpts From The EA Talmud

  If you find anyone who quotes that as an excuse where a modern Halachik authority would rule that they don’t have too much money for that to apply to them, I’ll agree they are just fine only giving 20%. (On the other hand, my personal conclusion is less generous.) But DINKs or single people making $100k+ each who comprise most of the earning to give crowd certainly don’t have the same excuse!