+6 -2

Deep Funding.md

reviewed

··· 1 1 # Deep Funding 2 2 3 3 - The goal of [Deep Funding](https://deepfunding.org/) is to develop a system that can allocate resources to public goods with a level of accuracy, fairness, and open access that rivals how private goods are funded by markets, ensuring that high-quality open-source projects can be sustained. Traditional price signals don't exist, so we need "artificial markets" that can simulate the information aggregation properties of real markets while being resistant to the unique failure modes of public goods funding. [Deep Funding is an Impact Evaluator](https://hackmd.io/@dwddao/HypnqpQKke). 3 3 + The goal of [Deep Funding](https://deepfunding.org/) is to develop a system that can allocate resources to public goods (discover public goods value signals) with a level of accuracy, fairness, and open access that rivals how private goods are funded by markets, ensuring that high-quality open-source projects can be sustained. Traditional price signals don't exist, so we need "artificial markets" that can simulate the information aggregation properties of real markets while being resistant to the unique failure modes of public goods funding. [Deep Funding is an Impact Evaluator](https://hackmd.io/@dwddao/HypnqpQKke). 4 4 5 5 In Deep Funding, multiple mechanisms (involving data, mechanism design, and open source) work together. Each layer can be optimized and iterated independently. 6 6 ··· 13 13 - Having experts fill weights manually 14 14 3. A mechanism that takes that weight vector as input and distributes money to the projects 15 15 16 16 - Deep Funding can be viewed as a [Software-2.0](https://karpathy.medium.com/software-2-0-a64152b37c35) approach to public-goods allocation. Instead of manually designing funding rules, evaluation processes, and governance structures, define an objective function, tests, eval sets, and scoring criteria. Then, let any kind of mechanism (AI models, prediction markets, statistical algorithms, human raters, etc.) compete to solve them. The human work shifts from hand-crafting decision procedures to specifying what "good allocation" looks like and collecting high-quality data. Everything else becomes an optimization problem where participants will try to produce weight predictions that best fit the data. Deep Funding can be seen as **an evolving benchmark suite for truthfully estimating public-goods value**, and progress comes from iterating on the evals rather than hard-coding the system itself. 16 16 + Deep Funding can be viewed as a [Software-2.0](https://karpathy.medium.com/software-2-0-a64152b37c35) approach to public-goods allocation. Instead of manually designing funding rules, evaluation processes, and governance structures, define an objective function, tests, eval sets, and scoring criteria. Then, let any kind of mechanism (AI models, prediction markets, statistical algorithms, human raters, etc.) compete to solve them. The human work shifts from hand-crafting decision procedures to specifying what "good allocation" looks like and collecting high-quality data. Everything else becomes an optimization problem where participants will try to produce weight predictions that best fit the data. Deep Funding can be seen as **an evolving benchmark suite for truthfully estimating public-goods value**, and progress comes from iterating on the evals rather than hard-coding the system itself. In this view, Deep Funding produces competing value models rather than a single canonical one. 17 17 18 18 ## Desired Properties 19 19 ··· 28 28 - Dependencies reveal themselves through market mechanisms rather than being declared 29 29 - Skin in the Game. Participants have something to lose from bad assessments 30 30 - Project Independence (no need to participate in the process to get funded) 31 31 + - Decouple epistemics from money distribution. 32 32 + - The system discovers candidate value estimations. 33 33 + - Funding bodies decide which to use. 31 34 32 35 ## Current Approach 33 36 ··· 104 107 - Allow projects to "insure" against their critical dependencies disappearing or becoming unmaintained. This creates a market signal for dependency risk and could fund maintenance of critical-but-boring infrastructure 105 108 - Composable Evaluation Criteria 106 109 - Rather than a single weighting mechanism, allow different communities to define their own evaluation functions (security-focused, innovation-focused, stability-focused) that can be composed. This enables plurality while maintaining comparability 110 110 + - Multiple weight models compete simultaneously and funding allocators choose among them. 107 111 - Create a bounty system where anyone can claim rewards for discovering hidden dependencies (similar to bug bounties) 108 112 - This crowdsources the graph discovery problem and incentivizes thorough documentation. 109 113 - Projects can opt out of the default distribution and declare a custom one for dependencies. Organizers can allow or ignore that

+8

Mechanism Design.md

reviewed

··· 21 21 - You can increase mechanism complexity if you trade it off for identity or collusion resistance. If you figure out a way to make it the mechanism identity resistant then, it'll support more complex setups. 22 22 - [Truthtelling games](https://jonathanwarden.com/truthtelling-games/) can incentivize honesty through coordination games where participants win by giving the same answer as others, with truth serving as a powerful Schelling point (truthtelling is the winning strategy only if everybody else tells the truth). Information elicitation mechanisms can get people to reveal private/subjective information truthfully even without verification. 23 23 - Some of the interesting properties of a mechanism are; local/bottom up decision making, can be combined in different layers (horizontal / vertical), and [[Modularity|modularity]]. 24 24 + - Mechanisms should preserve human agency instead of replacing it. 25 25 + - The best mechanism design work on real life environments: 26 26 + - Low-trust 27 27 + - Adversarial behavior 28 28 + - Disagreement 29 29 + - Evolve quickly 30 30 + - No pre-existing hidden legitimacy 31 31 + - Partial consensus 24 32 25 33 ### Examples 26 34