The latest comments and grant applications on grantmaking.ai.
The latest comments, grant applications, and grants on grantmaking.ai.
As a BASE advisor, I believe deeply in this mission because expanding the pipeline of Black AI safety researchers strengthens not only the field, but the communities these researchers serve. Diverse perspectives lead to more equitable, trustworthy, and responsible AI. Thank you for all you do to create these opportunities and invest in the next generation.
I've worked with Gergo as his executive coach over a period of 6 months and can testify to his strong work ethic, in-depth understanding of AIS field-building/culture/ecosystem, and insatiable drive for both personal development and impact optimisation through reasoning transparency, epistemic humility, and a scout mindset.
In addition to endorsing Gergo's person, I support the multiplier…
♂Three hundred bucks♂, giving as we can
@mrBlueSky Dude really appreciate it!
Brandon Thomason endorsed 
You have my endorsement. It will be interesting to see if the funders catch the correct ask and you are able to see this fully funded. Let me know either way how it turns out. Thank you for your trusting in my opinion, I only hope that it sharpens where sharpening is in greatest need.
-Brandon Thomason
ThomasonBuilds
Brandon, thank you. Your first review made my document better, and that's the rarest kind of help here. I'm deeply grateful for your attention to my project, your time and endorsement.
Your open question about power is fair, and you're right, it deserves a real answer in the funding conversation. Briefly, right now nobody can design a properly powered version: the between-seed spread has…
Katja Gorlinski commented on Approved--very interesting work!
Excited to recommend a $50,000 grant based on reviewers' endorsements, @Raffaello Fornasiere
Two quick questions:
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Brandon Thomason commented on Katja — I read the Stage 1 registration in full, including the appendices. This replaces my earlier notes rather than adding to them, since most of what I raised has been answered.
Taking my three items in order.
The attentive-human arm. Not only restored but promoted to the comparison that carries the hypothesis, with random supplying the yardstick. Your reason for moving the industry arm…
Brandon Thomason commented on Katja — one time-sensitive item, separate from my review of the registration, which I'll send after.
Your funding ask on the platform is frozen at $9,000 minimum / $20,000 ideal, and I understand from your note that you can't edit it. But the registered Stage 1 in Appendix B comes to $40,000 minimum, and line 1 alone — the independent selector at 344 hours — is $15,480. That single line is…
If there is any way for models to embed messages in secret language text, there must also be some way to construct the auditor’s side of the game, and this is the most methodical effort that I have seen in this respect: three auditor types, five knowledge environments, six embeddings, five datasets, and three model sizes.
May I inquire what single cell in this giant matrix the canary, the one…
Hi Katja,
Thank you for your interest and support. The strongest auditor setting is the Adaptive Chosen-Stego for all the three auditors (across all models and datasets). In case the steganographic scheme evades such a strong auditor, that would be concerning due to the possibility of "robust" secret collusion among misaligned agents. Ideally, we would like to see stronger auditors being…
Thank you, Vasisht, that's a strong point to stand on. So the canary is the strict auditor itself. This kind of failure is worth knowing about early enough. Good luck with the work, I hope the funding finds it.
Maxim Krivonogov endorsed Hello, and thank you again for your endorsement. Rather than replying in the spirit of "you praised me, so I praise you back", I studied your project as carefully as I could, so that I might actually be useful and repay kindness with kindness :) I think that is worth more.
The design is much better than most I have seen here. The random arm is what I like best, it separates structured…
Hi @Maxim Krivonogov,
Thank you a lot for reading my project with close attention Your three questions cover exactly the places I redesigned a couple of days ago and published openly just today, which says something uncomfortable about my original application.
The design is now frozen and public, so you can check it…
Wow, strong document, well done. I won't write a wall of text, most of what I could ask, you have already put on the page yourself in sections 6, 10, 14 and 15, so I'll stick to the one number that seems load-bearing.
The techniques themselves are standard, which is exactly why it is surprising to see them here: almost nobody applies all of them, and you did. You also helped me see something I…
@Maxim Krivonogov, thank you, especially for naming the one number that is worth checking beforehand.
You're right: d = 1.33 is a ratio to a spread that probably nobody has measured yet. I marked those values as provisional in the document for exactly that reason. And your suggestion fits perfectly into the run plan: measure the between-seed…
Hello @Brandon Thomason,
I genuinely support your core statement that a control believed to work but which does not is more dangerous than a control known to be absent. I think it deserves to be a slogan of your test :)
It is very good to see that you preregistered it in public, with a DOI, before any data, with the sample size, the…
Thank you, Katja — and thank you for engaging with how it's built rather than what it's about. That's rarer than it should be.
You picked the line I'd have picked. An absent control gets designed around; a false one gets trusted. That asymmetry is the whole reason the study exists — I want to know which kind of guardrail I've been writing.
The preregistration was defensive for the…
Thanks for your kind words, Brandon. I find your project carefully prepared.
And I'd be also genuinely grateful for any support to my project, especially for critique :)
Katja — you asked for critique rather than endorsement, so that's what this is. I went through the project page, the funding ask, and the comment thread, and I checked the sources I could reach. Most of what follows is design-level and fixable in the pre-registration. A few items are just corrections.
Before the list: the things I think are right, so you know what I'd protect if the design…
Dear @Brandon Thomason, thank you for your careful answer - it was a pleasure to read your critique.
So it doesn't get lost in a thread under someone else's project, I copied your text under mine project and answered there in full. Just briefly here: two of your three conditions are already closed in the frozen preregistration, and the third…
Katja Gorlinski commented on Katja — you asked for critique rather than endorsement, so that's what this is. I went through the project page, the funding ask, and the comment thread, and I checked the sources I could reach. Most of what follows is design-level and fixable in the pre-registration. A few items are just corrections.
Before the list:…
Doing what I can.
@mkim8829 We're thankful for that!
Love this show. Keep up the great work!
@ldw_cardinal Thank you for being the ultimate Mission Partner!
As Steve was able to fund this project separately and stated that he wouldn't spend money from this application, I've returned the funds to the donors!
@carol Was this money held for a full year? Was there something I was supposed to do to release it? I'm so sorry, I don't know the system well enough, I had no idea this money was sitting in limbo.
@StevePetersen that's okay! We don't really have a system for this: I think your report was sufficient, it was just overlooked on our end.
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Rounding you up to 46k
@davepatton Wooooo!
A governed, evidence‑bound reconstruction engine that operationalizes the Mythar language inside a constitutional computing framework
A dedicated study of the temporal horizon and how to steer it with minimum noise.
Approved! This seems like a good way to try to engage AI safety talent in India.
I want to initiate the cryptanalytic study of the options available to a model trying to learn its own weights, as distinct from an external adversary.
Hi, think this project is really cool, and thanks for engaging on twitter!
Do you have a 'minimum number of actors commited' for counterfactual impact, or any good way to minimize that? Trust (in the AI's perspective) seems like a forall statement to me, that is, you need to be sure you're never lied to.
I'm not quite sure I follow the question here - are you asking about how much of a record we expect to need to generate, and how consistently, to enable justifiably credible interactions with model instances?
yes, exactly
The theory of change is only tangentially x-risk, but all the grantmakers are excited about this application, as well as many others in comments, so I'm excited to approve the $20,000 grant!
I hope the Lightcone Commons funds the full request
@Constance Li, two quick questions:
+3 earlier replies
@Anton Makiievskyi 🔸 Happy to do that. Let me know if there's anything specific re: content/form that you'd like. Also, we are currently running the Incubator twice a year. Counting from August, the updates would be:
@Constance Li I don't assume your responsibility to updates for a whole year given that our grant is $10k and would be spent a lot quicker, but I'd definitely be excited to see as many updates as you feel like sharing
I'm interested in highlights of what was done (and lowlights too), so that I know if the grant was a good bet, and to feel…
@Anton Makiievskyi 🔸 For sure. It's $20k right? That's what's stated in your first comment & in the fundraising progress bar.
We can certainly arrange for the updates. I personally think the best timing would be:
@Constance Li Yes, sorry, it's $20k
Yes, these updates would be awesome!
