grantmaking.ai
Actively FundraisingRecent ActivityFull Database
Resources
grantmaking.ai
Actively FundraisingRecent ActivityFull Database
ResourcesApply for funding
grantmaking.ai kickoff grant round$278k / $1M distributed
Get funded

Recent Activity

The latest comments and grant applications on grantmaking.ai.

Recent Activity

The latest comments, grant applications, and grants on grantmaking.ai.

AllCommentsFunding AsksGrantsHide private comments
  1. David.goeckede endorsed
    Accelerating PauseAI Germany
    3h(edited)View

    As a recent volunteer for PauseAI Germany, I can say that the leadership Team is highly effective. On a shoestring budget they have started multiple Germany-wide writing-campaigns, briefed many people in leadership positions (journalists, lawmakers, professors) and coordinated various local meetups and speaking engagements. A little extra funding would go a long way towards improving the social…

    3h(edited)View
  2. Naufal Ridwan commented on Why Is It So Hard to Build Truly Safe AI? Dynamic Constraint Boundary
    3hView

    @Katja Gorlinski I've been following several public discussions and comments, and I see a context and relevance that I feel is close to what I'm currently developing. I'm also interested in your perspective, particularly on how decision-making behavior in both humans and machines takes place.

    I'm building integrity not as an additional safety…

    3hView
  3. BBrandon Thomasonis raising $11KforMeasuring whether guardrail phrasing changes agent rule-breaking.
    8hView
    IndividualResearchEvals

    A pre-registered study measuring whether prohibition-framed and approach-framed guardrails produce different rule-violation rates in deployed coding agents, so practitioners know whether the one sentence protecting their agent act

    8hView
  4. Marcus Abramovitch endorsed
    Frame Fellowship
    10hView
    10hView
  5. RRaymond Wangis raising $5KforAGIs who are free people: avoid the slave revolt
    10hView
    IndividualResearchAlignment Theory

    AIs that learn by generating ideas and selecting using contradiction, rather than optimization; end goal is free people not tools.

    10hView
  6. Marcus Abramovitch endorsed SafeBio-Registry: An Open-Source SpecDef platform for PLMs/GLMs
    10hView
    10hView
  7. Marcus Abramovitch endorsed Detailed Threat Models for AI-Enabled Extreme Power Concentration
    10hView
    10hView
  8. Katja Gorlinski commented on No Silent Landing — Landing Integrity Testbed
    1dView

    Loek, I owe this project a careful read for a personal reason: its author found a version boundary in my own project. So I assume that your lens clearly works on live material. I have two questions out of my genuine curiosity☺️:

    1. When a decision «lands», what is the smallest thing you freeze: the words of the approval, the meaning as it was understood, or the authority that approved it? I…
    Loek Verdonk1d

    Thank you Katja.
    Waiting for a dance like this on my own page, so thank you for that :D :D: D!!!

    These are very good questions tho, because both expose where my current description is still too compressed.

    At the lowest technical level, the only thing I can literally freeze is an exact object: bytes, a version, a hash.

    But that is not enough to preserve a decision.

    The smallest…

    Katja Gorlinski11h

    Thank you, Loek, for the detailed answer.

    The way you corrected your own «15 tests» claim in public, plus your two false-green catches, looks to me like concrete proof of your lens so far.

    What would be interesting to read next is the experiment itself: one concrete workflow, from input to verdict, that the full testbed would run. But that is just my curiosity.

    Good luck!

    1dView
  9. Marcus Abramovitch commented on Distinguishing progress in data and algorithms
    12hView

    COI: I know Robi and consider him a friend.

    This seems quite important to me. I don't think I quite understand how this project would work though.

    Robi Rahman12h

    Hey Marcus,

    Glad you understand the significance; let me try to explain the mechanics of the experiments.

    We'll train models using {old, new} algorithms and {old, new} data to see how much new data helps, compared to new architectures, does that part make sense?

    12hView
  10. Marcus Abramovitch endorsed Distinguishing progress in data and algorithms
    12hView
    12hView
  11. Ryan Kidd commented on Postdoc on zero knowledge verification for frontier AI training
    2dView

    I endorsed this grant proposal because:

    • Compute verification is vital to international treaties, which are a vital component of Plan A, Plan S, and many other AI slowdown scenarios. There are ~60 people working full-time on compute workload verification worldwide.
    • Paul seems like a qualified scientist. A postdoc seems like a good mechanism for accomplishing this research.
    • I know and…
    Ryan Kidd2d

    Absent cG support, I'd recommend a full 12 months funding, to give Paul plenty of time to accomplish this project. However, as he is a strong cG grant candidate, he might just need 3 months runway to secure a cG grant. I'll split the difference and recommend 6 month's funding.

    Marcus Abramovitch12h

    this seems very reasonable to me

    2dView
  12. Marcus Abramovitch endorsed Postdoc on zero knowledge verification for frontier AI training
    12hView
    12hView
  13. Marcus Abramovitch endorsed What Factors Influence Chain-of-Thought Faithfulness?
    12hView
    12hView
  14. Marcus Abramovitch endorsed The AI Safety Stack
    12hView
    12hView
  15. Marcus Abramovitch commented on Safeguarding DNA Synthesis from Biological Risks Posed by Advanced AI
    12hView

    I would like this fully funded for $50k and am going to make it happen.

    I agree that this would be a loophole in how our current safeguards work for this threat, and I support the person who brought it to my attention in studying it and remedying the situation if necessary.

    If the approach doesn't work, this is valuable to know and we should look for other methods of preventing this…

    12hView
  16. Marcus Abramovitch endorsed Safeguarding DNA Synthesis from Biological Risks Posed by Advanced AI
    12hView
    12hView
  17. grantmaking.ai funded
    Formalizing mathematical AI safety research in Lean     for $15,000
    13hView
    $7K minimum$30K ideal
    This project is still open for additional funding
    13hView
  18. grantmaking.ai funded
    Evaluating the Tacit Coordination Capacity and Decision Theory of LMs     for $10,000
    13hView
    $5K minimum$10K ideal
    Funding for this project is now complete
    13hView
  19. grantmaking.ai funded
    Using deals to find misalignment in current AI models     for $25,000
    13hView
    $11K minimum$25K ideal
    Funding for this project is now complete
    13hView
  20. grantmaking.ai funded
    Adversarially robust workload classification using hardware sensors     for $30,000
    13hView
    $10K minimum$50K ideal
    This project is still open for additional funding
    13hView
  21. Ryan Kidd commented on
    Scaling adoption of synthesis screening outside the US and EU
    2dView

    I endorsed this grant proposal because:

    • Biosecurity is the highest risk near-term AI threat vector and non-existential biorisks could significantly weaken civilisational resilience ahead of AGI (like COVID arguably did). There are less than 100 people working full time on worst-case biosecurity.
    • Screening mail-order DNA seems like a no-brainer. This needs to be funded yesterday! Even if…
    Anton Makiievskyi 🔸14h

    Great!

    @Tessa Alexanian, some logistics:

    1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?
    2. Please confirm your commitment to post quarterly updates on how the project is going
    2dView
  22. Ryan Kidd commented on
    safely.bio: Customer screening for DNA synthesis providers
    2dView

    I endorsed this grant proposal because:

    • Biosecurity is the highest risk near-term AI threat vector and non-existential biorisks could significantly weaken civilisational resilience ahead of AGI (like COVID arguably did). There are less than 100 people working full time on worst-case biosecurity.
    • KYC protocols for mail-order DNA seems like a no-brainer. This needs to be funded…
    Ryan Kidd2d

    Also, it's possible that cG won't be funding DNA synthesis screening?

    Anton Makiievskyi 🔸14h

    Awesome!

    @Phil Palmer ,

    1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?
    2. Please confirm your commitment to post quarterly updates on how the project is going
    2dView
  23. Loek Verdonk commented on
    proofbundle: Offline Evidence Receipts for AI Safety Evaluations
    14dView

    Hi Konrad,

    Great work on proofbundle — your offline receipts feel very familiar to what we are building. We literally call ours “bonnen” (Dutch for receipts). :D

    One question from our hardest corner: how does proofbundle handle legitimate change? If an evaluation record is intentionally superseded by a new version, is that treated as tampering, or can the receipt represent a visible…

    +33 earlier replies

    Loek Verdonk3d

    Well mate, I can see the goggles starting to move toward a full 360-degree view — and it was you who set that movement in motion. :D

    Besides that, your dancing shoes shine with honesty, and that makes this dance worthwhile in more than one way. :D :D :D

    For that, it would be an honor to choose the exact-byte route.

    Not because your constructed equivalent is technically inadequate — it…

    Konrad Gruszka3d(edited)

    @Loek Verdonk
    Hi Loek,

    right back at you! you make the dance worthwhile. The exact-byte route you opened is now built,
    and iter9 carries your artifact under its own name, not a stand-in for it.

    Your list, item by item, as shipped:

    iter8 root explicitly superseded: 9f827e25706479daa420bb4bcf680899e169ce52d813936452d47f2801651063…

    Loek Verdonk2d

    @Konrad Gruszka

    Thank you as well. This dance was quite a tough one for me :D — but the results show why. :D :D :D

    Goggles fully on. Hulk has taken off the dancing shoes and picked up the sledgehammer again. :D

    -->

    Review package:

    LOEK_ITER9_FRACTAL_ATTACK_REVIEW_V2…

    Konrad Gruszka18h(edited)

    @Loek Verdonk
    Hi Loek :D

    thank you again for the iter9 fractal-attack review. Your 12 neighbouring findings all landed in the
    first-party conformance instruments of the corpus (the comparator, the result-provenance verifier,
    the candidate/negtest verifiers, the spec and schemas), not in the proofbundle receipt cryptography
    (0 core…

    14dView
  24. Ankur Pandey endorsed
    ARC Risk Communication Initiative
    19hView

    AI risk comms is needed more than over. We need folks trained in nuanced thinking about risks to counter clickbaity discourse.

    19hView
  25. Miles Tidmarsh commented on
    Does the thinking behind data selection shape a model's character?
    2dView

    While I support the general concept, I worry that outsourcing the core of the project to an ML engineer who doesn't understand the motivations and directions as you do is not a good move to producing high quality research. (this paper we wrote is quite relevant here https://arxiv.org/abs/2606.24890)

    Katja Gorlinski1d

    Thank you, Miles, for taking a look through my project carefully 🙌
    Your worry is fair: small implementation choices can quietly change the output being measured, and your paper shows exactly how small edits lead to large effects.

    That is why I designed the handoff to be narrow. The corpora and the design stay with me: I select the data and freeze the falsification criteria. The engineer…

    Miles Tidmarsh1d

    Great to hear you've thought about this so carefully, this seems like a good plan

    Katja Gorlinski21h

    @Miles Tidmarsh Thanks for your kind words :)

    2dView
  26. Gavin Leech commented on Adversarially robust workload classification using hardware sensors
    2dView

    Hi Robi,

    We'd like to fund this for $30k. Logistics:

    1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?

    2. Please confirm your commitment to post quarterly updates on how the project is going

    Obvious question is how adversarially robust this is, and obvious answer is not very. That's not a fatal…

    Robi Rahman1d(edited)

    Hi Gavin,

    We haven't received any more funding that I'm aware of since I submitted this, but Felix was applying for some small (~$5k) personal/transitional funds, so I'll check with him if he got any of those.

    The main thing to mention is that we haven't been able to receive our previously awarded FAR AI verification grant yet due to legal bureaucracy. They can only grant to US…

    Gavin Leech1d

    @Anton Makiievskyi 🔸

    Anton Makiievskyi 🔸23h

    @Gavin Leech @Robi Rahman happy to make the grant unrestricted , as long as Manifund is be able to distribute it

    2dView
  27. Gavin Leech commented on The AI Safety Stack
    1dView

    Hi Tomáš,

    We'd like to fund this for $50k (our maximum). Logistical questions:

    1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?

    2. Please confirm your commitment to post quarterly updates on how the project is going

    1. How will funding be disbursed? Equal split to three individuals?

    I'm very…

    1dView
  28. Gavin Leech commented on Using deals to find misalignment in current AI models
    2dView

    Hi Ram,

    We'd like to fund this for $25k. You may need more than that. Logistics:

    1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?

    2. Please confirm your commitment to post quarterly updates on how the project is going

    I doubt that $100 is enough to move a misaligned model. Consider ways to make it clear…

    Ram Potham1d

    Hi Gavin, thanks, and yes to the $25k.

    Logistics

    • No funding received since submitting and nothing else pending. Ask unchanged.
    • Quarterly updates: confirming I will do this
    • On needing more: I think the current amount is sufficient. I might need more if I find that models are scope sensitive to money and want to increase the money given to them. I currently don't expect them to be…
    Gavin Leech1d

    Great! Godspeed

    2dView
  29. Liu Fen commented on Reading China's AI Registry: A Public Monitor
    1dView

    This is a rare case of someone with the actual language skills and legal background to read China's AI filings in the original turning that into an ongoing, usable resource for a field that's mostly been working from English summaries.

    1dView
  30. Liu Fen endorsed
    Token taxes as a mechanism for reducing AI-driven power concentration.
    1dView

    This is a timely and well-grounded proposal: it moves the token-tax idea from position paper to concrete policy analysis at exactly the moment political interest is materializing, with a clear research plan and a credible partner in IPPR to ensure the output lands with policymakers. I'm glad to endorse it.

    1dView
Load more