All of Parker_Whitfill's Comments + Replies

Arxiv link here https://arxiv.org/abs/2507.23181

We are still working on getting a more official version of this on Arvix, possibly with estimates for $\lambda$ and $\varphi$. 

When we do that, we'll also upload full replication files. But I don't want to keep anyone waiting for the data in case they have some uses for it, so see here for the main CSV we used: https://github.com/parkerwhitfill/EOS_AI

On 2), the condition you find makes sense, but aren't you implicitly assuming an elasticity of substitution of 1 with Cobb-Douglas?

Yes, definitely. In general, I don't have a great idea about what $Z(\cdot )$ looks like. The Cobb-Douglas case is just an example. 

Yep. We are treating $L$ as homogenous (no differentiation in skill, speed, etc.) I'm interested in thinking about quality differentiation a bit more. 

In complete generality, you could write effective labor as 

$L=Z(H,A{K}_{inf},A{K}_{train})$.

That is, effective labor is some function of the number of human researchers we have, the effective inference compute we have (quantity of AIs we can run) and the effective training compute (quality of AIs we trained). 

The perfect substitution claim is that once training compute is sufficiently high, then eventually we can spend the inference compute on running some AI that substitutes for human researchers. Mathematically, for some $x$, 

$Z(H,A{K}_{inf},x)=H+\frac{A{K}_{inf}}{c}$

w... (read more)

Here is a fleshed out version of Cheryl's response. Lets suppose actual research capital is $qK$ but we just used $K$ in our estimation equation. 

Then the true estimation equation is 

$\ln \frac{qK}{L}=\sigma \ln \frac{\gamma }{1-\gamma }+\sigma \ln \frac{w}{r}$

re-arranging we get 

$\ln \frac{K}{L}=\sigma \ln \frac{\gamma }{1-\gamma }-\ln q+\sigma \ln \frac{w}{r}$

So if we regress $\ln \frac{K}{L}$ on a constant and $\ln \frac{w}{r}$ then the coefficient on $\ln \frac{w}{r}$ is still $\sigma$ as long as q is  independent of $w/r$. 

Nevertheless, I think this should increase your uncertainty in our estimates because there is clearly a lot go... (read more)

Note that if you accept this, our estimation of $\sigma$ in the raw compute specification is wrong. 

The cost-minimization problem becomes 

${min}_{H,K}wH+rKs.t.F(AK,H)=\overline{F}$.

Taking FOCs and re-arranging, 

$\frac{K}{H}=\sigma \frac{\gamma }{1-\gamma }+\sigma \ln \frac{wA}{r}$

So our previous estimation equation was missing an A on the relative prices. Intuitively, we understated the degree to which compute was getting cheaper. Now A is hard to observe, but let's just assume its growing exponentially with an 8 month doubling time per this Epoch paper. 

Imputing this guess of A, and estimating via OLS w... (read more)

I spent a bit of time thinking about this today.

Lets adopt the notation in your comment and suppose that $F(\cdot )$ is the same across research sectors, with common $\lambda$. Let's also suppose common $\sigma <1$. 

Then we get blow up in $A_{cog}$ iff 

$\{\begin{array}{cc}{\varphi }_{cog}+\lambda >1& \text{if}{\varphi }_{cog}\le {\varphi }_{exp}\\ max\{{\varphi }_{cog},{\varphi }_{exp}+\lambda \}>1& \text{if}{\varphi }_{cog}>{\varphi }_{exp}\end{array}$

The intution for this result is that when $\sigma <1$, you are bottlenecked b... (read more)

Also, updating this would change all the intelligence explosion conditions, not just when $\sigma <1$. 

Yep, I think this gets the high-level dynamics driving the results right. 

Thanks for the clarification. We updated the post accordingly. 

This is a good point, we agree, thanks! Note that you need to assume that the algorithmic progress that gives you more effective inference compute is the same that gives you more effective research compute. This seems pretty reasonable but worth a discussion. 

Although note that this argument works only with the CES in compute formulation. For the CES in frontier experiments, you would have the $\frac{A{K}_{res}}{A{K}_{train}}$ so the A cancels out.^[1]

1. ^{^}

   You might be able to avoid this by adding the A's in a less naive fashion. You don't have to train larger models

Thanks for the insightful comment. 

I take your overall point as the static optimization problem may not be properly specified. For example, costs may not be linear in labor size because of adjustment costs to growing very quickly or costs may not be linear in compute because of bulk discounting. Moreover, these non-linear costs may be changing over time (e.g., adjustment costs might only matter in 2021-2024 as OpenAI, Anthropic have been scaling labor aggressively). I agree that this would bias the estimate of $\sigma$. Given the data we have, there sho... (read more)

Great paper, as always Phil. 

I'm curious to hear your thoughts a bit more about if we can salvage SWE by introducing non-standard preferences. 

Minor quibble: "There is then no straightforward sense in which economic growth has historically been exponential, the central stylized fact which SWE and semi-endogenous models both seek to explain" 

I agree that there is no consumption aggregate under non-homothetic preferences, but we can still say economic growth has been exponential in the sense that GDP growth is exponential. Perhaps it is not a ... (read more)

People often appeal to Intelligence Explosion/Recursive Self-Improvement as some win-condition for current model developers e.g. Dario argues Recursive Self-Improvement could enshrine the US's lead over China. 

This seems non-obvious to me. For example, suppose OpenAI trains GPT 6 which trains GPT 7 which trains GPT 8. Then a fast follower could take GPT 8 and then use it to train GPT 9. In this case, the fast follower has a lead and has spent far less on R&D (since they didn't have to develop GPT 7 or 8 themselves).  

I guess people are thinking that OpenAI will be able to ban GPT 8 from helping competitors? But has anyone argued for why they would be able to do that (either legally or technically)?

Is the alignment motivation distinct from just using AI to solve general bargaining problems? 

Here is a counterargument: focusing on the places where there is altruistic alpha is 'defecting' against other value systems. See discussion here

Roughly buy that there is more "alpha" in making the future better because most people are not longtermist but most people do want to avoid extinction. 

Good point, but can't this trade occur just through financial markets without involving 1 on 1 trades among EAs? For example, if you have short timelines, you could take out a loan, donate it all to AI Safety. 

Agreed with this. I'm very optimistic about AI solving a lot of incentive problems in science. I don't know if the end case (full audits) as you mention will happen, but I am very confident we will move in a better direction than where we are now. 

I'm working on some software now that will help a bit in this direction! 

Since it seems like a major goal is of the Future Fund is to experiment and gain information on types of philanthropy —how much data collection and causal inference are you doing/plan to do on the grant evaluations? 

Here are some ideas I quickly came up with that might be interesting. 

1. If you decided whether to fund marginal projects by votes or some scoring system—you could later assess what you think is the impact of funding projects by using a regression-discontinuity-design. 
2. You mentioned that there is some randomness in who you used as r

I'd say it's close and depends on the courses you are missing from an econ minor instead of a major. If those classes are 'economics of x' classes (such as media or public finance), then your time is better spent on research. If those classes are still in the core (intermediate micro, macro, econometrics, maybe game theory) I'd probably take those before research. 

Of course, you are right that admissions care a lot about research experience - but it seems the very best candidates have all those classes AND a lot of research experience. 

I would say an ideal candidate is a math-econ double major, also taking a few classes in stats and computer science. All put together, that's quite a few classes, but not an unmanageable amount. 

One case where this doesn't seem to apply is an economics Ph.D. For that, it seems taking very difficult classes and doing very well in them is largely a prerequisite for admissions. I am very grateful I took the most difficult classes and spent a large fraction of my time on schoolwork. 

The caveat here is that research experience is very helpful too (working as an RA). 

Is there a strong reason to close applications in January? 

I'm only familiar with the deadlines for economics graduate school, but for that you get decisions back from graduate school in February-March along with the funding package. Therefore, it would be useful to be able to apply for this depending on the funding package you receive (e.g. if you are fully funded you don't need to apply, but if you are given little or no funding, it would be important to apply) . 

I highly recommend cold turkey blocker, link here. It offers many of the features you listed above,  including scheduled blocking, blocking the whole internet, blocking specific URL or search phrases (Moreover, this can be done with regex, so you can make the search terms very general),  password-protected blocks, no current loopholes (if there are ones please don't post them, I don't want to know!) and the loopholes that used to exist (proxies) got fixed. 

Pricing seems better than freedom as it's $40 for lifetime usage. My only complaint is that there is no phone version. 

I'd still agree that we should factor in cooperation, but my intuition is then that it's going to be a smaller consideration than neglect of future generations, so more about tilting things around the edges, and not being a jerk, rather than significantly changing the allocation. I'd be up for being convinced otherwise – and maybe the model with log returns you mention later could do that. If you think otherwise, could you explain the intuition behind it?

I think one point worth emphasizing is that if the cooperative portfolio is a p... (read more)

What piece of advice would you give to you 20 year old self?

Strong upvote as a find EA book recommendations very useful, and I'd like to encourage more people to post recommendations.

As an aside, it could be worth noting which books are available as audiobooks.

My vague impression is that this is referred to as pluralism in the philosophy literature, and there are a few philosophers at GPI who subscribe to this view.

Thanks for the summary and the entire sequence of posts. I thoroughly enjoyed them. In my survey of the broader literature c) is mostly true and I'd certainly like to see more philosophical engagement on those issues.

"A pretty standard view of justice is that you don't harm others, and if you are harming them then you should stop and compensate for the harm done. That seems to describe what happens to farmed animals."

I think this only applies to people who are contributing to the harm. But for a vegan for is staunchly opposed to factory farming, they aren't harming the animals, so factory farming is not an issue of justice for them.

"Whether we seek to alleviate poverty directly or indirectly, we might suppose that such efforts will get a privileged status over very different cause areas if we endorse the justice view. But our other cause priorities deal with injustices too; factory farming is an unjust emergency, and an existential catastrophe would clearly be a massive injustice that might only be prevented if we act now. And just like poverty, both of these problems have been furthered by selfish and corrupt international institutions which have also contributed to our wealth