The latest comments and grant applications on grantmaking.ai.
The latest comments, grant applications, and grants on grantmaking.ai.
A community tool which accelerates how people find others in the AI space based on mutual interests and goals. Also a "radar" which scans Slack channels (e.g. Blue Dot's) for relevant opportunities.
[Final report]
@chi-nguyen It looks like this report is missing the text; I'm curious what the latest on this project is!
Hello @Aishwarya Balwani,
It is inspiring to meet a scientist who works on a well-known, complicated problem.
I was really impressed by your design, that your falsification logic is built into the method itself, so masks that hold across independently trained adapters mean a reusable substrate, and masks that only work for one run mean an…
Hello @Katja Gorlinski, thanks so much for your endorsement!
Our webpage for the shared EM subspaces work is: https://darturi.github.io/shared-em-subspaces/.
OpenReview links for the NeurIPS workshops are:
It was my pleasure, because your project is really worthy :)
And it is a pleasure to talk with expert whose work has been formally recognised, and thank you for the links.
Your observation catches my mind: that different narrow tasks end up travelling through the same directions in the weights, as if they were finding one shared corridor rather than each carving its own. To me that reads…
The grantmakers were excited about this project and we are not funding mostly because the grant round is limited by 50k per grant, and I didn't want to fund it for an amount that won't take the project off the ground. I hope it will end up funded by the Lightcone Commons
Thanks for the feedback, Anton. I really appreciate the consideration.
Given the €50k funding cap, would you be open to supporting a reduced scope proof of concept instead? I could adjust the proposal to focus on validating the core assumptions and delivering an initial milestone within the available budget.
If that’s of interest, I’d be happy to revise the proposal accordingly.
Christos
Hi Tomáš,
We'd like to fund this for $50k (our maximum). Logistical questions:
Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?
Please confirm your commitment to post quarterly updates on how the project is going
I'm very…
That is great news - thank you!
1. Did you receive funding from anywhere since submitting this application, or has the funding ask changed for any other reason?
No to both.
2. Please confirm your commitment to post quarterly updates on how the project is going
We do. Say 16 Sept latest, though I hope we'll have a shareable prototype much sooner.
*3. How will…
The first phase is mapping relevant literature (+LW/AF, blog posts, ...) to alignment techniques (a rich taxonomy of techniques is WIP), models mentioned/used/..., and orgs&labs. The first goal is presenting this as a technique/literature survey for both alignment researchers and experts in policy, govt and journalism. I agree that this already seems useful [@Pip…
Hello, @Maxim Krivonogov
It was very interesting to read your project.
I strongly support your observation: a judge you can bend by rephrasing is not a judge at all. That is a real oversight problem, and, in my humble opinion, you stated it plainly and measured it with effect sizes instead of adjectives.
Love your discipline in design:…
+1 earlier reply
Maxim, this reply shows that my endorsement did not get weaker today. It got its proof.
My congrats on receiving your first external review!
In my opinion, what you did is rare. You learned that part of your own result was overstated, you fixed it, and published the correction before the end of a funding round, when honesty can cost real money. My respect :)
Good to see that discipline…
And I'd be grateful for any kind of support in return for my project , especially your critique :)
@Katja Gorlinski Of course, give me time to study the project thoroughly and give a reasoned assessment!
@Maxim Krivonogov Thank you, take your time 😊
I really like the project @Katja Gorlinski. The one critique that comes to mind is scope — it all runs on one model, so a win tells you the curation moved that model's character, not that it would move another one trained differently. Seeing it hold across a couple of model families is the part I'd be most curious about.
Good luck with it.
@Abraham Asseffa, thanks for your kind wishes, and you've read my further plans for when the project gets funded, because cross-model replication is the logical next step, and I'm curious to see how this holds across different model families too.
Thanks for confirming this direction.
After reading the proposal, the discussion, and the thoughtful feedback from Ryan and others, I decided to support this project because I believe it addresses an area that deserves much more attention as AI systems become increasingly collaborative.
This is one of the more interesting proposals I've read on multi-agent alignment because it focuses on something that often receives less…
@rosely-ai Hey David, are you actually "supporting" as in planning on donating money? Or more just voicing support? Also, how did you find this project in the first place?
FWIW: while we're okay with some amount of AI assistance among users and posting comments, and also some amount of self promotion, I think your case might be across the line.
Hi @Anastasia Wei and team,
I respectfully endorse this project, and I read it slowly before that.
It is difficult not to notice that the field keeps building on results nobody has re-run, including the one you address. I also remember that publication where Anthropic showed that models trained on explanations of why a model's behaviour is…
I endorsed this grant proposal because:
Excited to see this recommendation
@Akshyae Singh, please let me know:
Private comment. Only shown to approved funders and grant reviewers.
Private comment. Only shown to approved funders and grant reviewers.
Approved--it sounds like BAISH has successfully gotten many people working in AI safety!
I know Roman from a chat group about AI Safety, where he consistently contributes insightful and well-reasoned messages! The project proposal is great and potentially very impactful if it pans out. Gladly endorsed!
Rooting for Roman!
Private comment. Only shown to approved funders and grant reviewers.
I have known Roman for almost six years, since our first undergraduate year. While AI safety is not my focus area, my conversations with Roman convinced me that everyone, including myself, should pay attention to this problem.
Roman has consistently demonstrated his devotion to the AI alignment field. For a long time, his dream was to discover a new type of fundamental particle, and he was…
As someone who also runs comms for an AI safety field-building org myself iaS, I know first-hand how hard this space is to communicate well, and I am glad to back this project. ARC's five-workstream approach research, framework development, field visibility, outreach and format innovation, tackles the coordination and terminology gaps at the right level: infrastructure, not just messaging. The…
As someone building AI safety infrastructure for the Spanish-speaking world myself, I've seen firsthand how much groundwork BAISH has laid, and how far a small team of committed people can go with the right structure. I've learned a great deal from them, and expect to keep learning as I am doing the same on the Spanish-speaking side of the field. Their track record on placements and fellowships…
I think people's understanding of China is very poor and further understanding of China would be great for improved communication, less scary thoughts of China as the "other", etc.
As such, I am recommending we fully fund the remainder of this ($22000-$2350=$19650).
I just need these two questions answered first.
Hi Marcus!
I am so happy to read that you find this idea valuable!
No haven’t received any funding since I sent the application . The funding ask hasn’t changed.
yes happy to post quarterly updates.
Our mission at Encode Africa is to strengthen Africa's footprint in the global AI Safety ecosystem by fostering a collaborative community where diverse actors contribute meaningfully towards shaping safe and ethical AI.
Making a small donation as a sign of support; I continue to be excited for games as a method of imparting interesting lessons (eg universal paperclips, maybe Manifold). Pip also has a good track record of supporting work in the xrisk space, through Ashgro and now https://beacongcr.org/
I wasn't actually able to demo the game myself -- I'm on Mac, Sequoia 15.5,…
@Austin Thanks!
I am developing on PC and have only been able to test on one Mac build so far, but I'll work on making that more robust. I also need to go pay for some certificates to make it easier for people to get the game up and running.
I am probably culturally biased against mobile games so that is definitely a blindspot of mine, noted!
Noted on the AI generation and…
Hi @Logan Graves,
Excited to recommend the grant of $12,000!
Two quick questions before we distribute:
Hi Anton,
Delighted to receive the grant! Thank you for the recommendation.
(1) Funding situation has not changed — haven’t received any other funding and don’t need to update my funding ask.
(2) I will be happy to post quarterly updates on the project! I intend to have at least one paper out before the end of September, or else a blog post on why the hypotheses failed and related…
@Naufal Ridwan Thank you, I enjoyed your comment under my project. And both our projects are truly similar at the base: training-data structure is not an empty slot, but, likely, a move that writes structure.
Even though I can't find the bridge from neurobiology to silicon, because LTD in the dlPFC and how a model weights patterns are two…
+3 earlier replies
@Naufal Ridwan Thank you for the exchange of some interesting ideas. Your DCB sounds bright and positive - definitely wishing you a successful test of it.
I answered your question above, sorry if my answer is not enough - I tried my best.
My warm wishes.
I appreciate your willingness to engage in this discussion. However, I would like to note one thing: this discussion began with you posing a very difficult question about AGI turning away from humans a question that I answered using the framework I have. When I returned the same question to your project, I did not see a…
@Naufal Ridwan,
It seems we define the word "answer" differently, and that is fine.
You answered my question with a framework: a description of how DCB should behave, not yet tested against reality. I answered yours with an experiment: a design, published on this page, that will test whether my mechanism exists at all. You read that as a deflection because the experiment is unfinished. For…
Thank you for your response. I would like to clarify one thing about the definition of “answer” you mentioned.
When we talk about a hypothesis, we are talking about something that might happen in the future. Your question about AGI is a question about the future something that does not yet exist and has not yet happened.…
CaML does rigorous, empirical technical research in order to better understand AI behavior and motivations. Their research is currently outside the Overton window of most AI safety researchers, but I believe that the questions they are asking will become very relevant in the next few years as AI systems continue to improve in complex behavioral capabilities and autonomous deployment. I'm excited…
Approved--excited for this project!
Thanks Carol!
Using MUDs as a long-term, persistent AI evaluation environment for understanding model behavior and psychology.
omomehin emmanuel king commented on LEX-Aureon: Stateful Constitutional Runtime Governance for LLM's and AI AgentsSubmitted verification Loom video on 14th July as requested through mail. Happy to answer any questions or provide more details on the Project if needed
💪
Hi Nikita,
I found your project worthy of endorsement because of its logic, clear design and strict discipline.
I should admit that your project speaks to me directly as a business owner: I make the decision about switching or adding a new LM to my workflow once every 2-3 months, and it is always challenging. I can only dream about an automated, well-tested tool for making that choice…
