![[session7_social planner 2025-04-13-11.svg]] %%[[session7_social planner 2025-04-13-11|🖋 Edit in Excalidraw]]%% 2025-04-03 [[tan_zhixuan]]'s curation | Paper | Xuan's Curation | | --------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [📜 Pragmatic instruction following goal assistance via cooperative language-guided inverse planning](https://arxiv.org/abs/2408.12022) | "Agents reasoning about other agents to take decisions (in this case, to provide assistance) [...] People often give instructions whose meaning is ambiguous without further context, expecting that their actions or goals will disambiguate their intentions. This paper introduces cooperative language-guided inverse plan search (CLIPS), a Bayesian agent architecture for pragmatic instruction following and goal assistance." | | [Understanding Epistemic Language with a Bayesian Theory of Mind](https://arxiv.org/abs/2408.12022) | "I've also started doing more work on the belief side of things." The paper introduces a cognitive model of epistemic language interpretation called language-augmented Bayesian theory-of-mind (LaBToM) that translates natural language into an epistemic "language-of-thought" and evaluates these translations against inferences from a probabilistic generative model. | | [📜 Planning with theory of mind for few shot adaptation in sequential social dilemmas](https://openreview.net/forum?id=Y8OaqdX5Xt) | "A nice example of planning with theory-of-mind used in the context of multi-agent environments to better coordinate (or compete) with other agents." The paper proposes PToM, a multi-agent algorithm enabling few-shot adaptation to unseen policies in sequential social dilemmas through a hierarchical approach combining opponent modeling and Monte Carlo Tree Search. | | [A unifying framework for observer-aware planning and its complexity](https://proceedings.mlr.press/v161/miura21a.html) | "There's also a bunch of work in the I-POMDP literature and offshots (like this one) that are about planning with theory of mind." This paper introduces Observer-Aware MDP (OAMDP) as a less complex framework than I-POMDP for producing observer-aware behaviors, analyzing their relationship and establishing complexity. | | [Planning with Theory of Mind (Mark Ho)](https://probcomp.slack.com/files/UN3TMRLMD/F08JPFLLG06/planning_with_theory_of_mind.pdf) | "This paper by Mark Ho is also a nice introduction to the idea of using a theory of mind to make plans (e.g. planning to influence other people's actions by changing their beliefs and/or desires)" | | [Group-level theory of mind for efficient coordination (Kleiman-Weiner)](https://onlinelibrary.wiley.com/doi/full/10.1111/tops.12525) | "I really like this paper by Max Kleiman-Weiner on using group-level Bayesian theory-of-mind to solve tasks together -- it's one of the inspirations for my work on pragmatic instruction following." | [[📜Chandra25_dsppl_rr_memo]] 2025-02 as i was preparing for NBER entrepreneurship camp [[day1 david(entfinance)]] using https://claude.ai/chat/b22b9b97-4b74-4697-a33a-5a507d22ccd6 ![[Pasted image 20250212083605.png]] 1. ☝️one and still. "Entrepreneurial strategy under uncertainty is not a one-and-done plan – it’s an evolving hypothesis." 🌙 could you connect this # ✉️ choosing tech envelope? sketch a feedback loop between entrepreneur and society using the p(phi, theta) model and imagine how sigma_phi is from future; how as time passes, s curve ✉️envelope becomes concrete on which the next model builds on that lowers s_phi as theta is my execution capability 2. 👴's portfolio. "an entrepreneur might update their belief that a technology is not viable after one trial and stop investing, even though a social planner with a portfolio of many such trials might continue to fund experimentation to eventually unlock a breakthrough." 🌙 this reminds me of social planner (warren buffet) survives yet another day exceeding the lifespan of ⭐️startups 3. 🔬🔭"entrepreneur can make data-driven adjustments" 🌙 can we think of a good analogy related to coordinating magnifying lens (현미경 배율조정)? if initial strategy was to target a lead user group and the pilot experiments reveal unexpectedly strong interest from a larger segment, the entrepreneur updates their market beliefs and can scope out (switching lens from 100x to 10x) " could you frame startup pivoting as balancing supply and demand uncertainty via experiment choice? THIS IS RELATED TO "allocate attention in rational and dynamic and goes far beyond the schedulers and caching systems that exist and operating systems and hardware, but not forever. Maybe one of you will make a decision theoretic schedule for the GPU." comment from # vikash's probabilistic program 4. ⛏️mining for test quantities 🌙 i wish you to read my draft below and imagine how my "Proposed Solution: Mining Exchangeable Patterns" can be connected with your "👥In effect, the industry as a whole is conducting a portfolio of experiments."