Welcome to dimenwarper's personal research swarm's chorus -- a list of currently work in progress ideas with an agentic orchestrator in the backend. We are currently exploring an eclectic mix of ideas spanning LLM test-time harnesses/deep agentic coding and their application to problems in bio and math, geometric applications to deep learning and probabilistic programming, and other miscellaneous topics. Have an idea that you think we should focus on? Vote on the ideas below or submit your own!
An LLM anyopt (optimize anything) framework for exploring test-time harnesses for deep agentic coding and tackling math/bio questions
dimenwarper/fanoutCan we fully describe GPUs as a thermodynamical system in terms of its information processing capability? Can we backcalculate optimal model architecture designs bsaed on this?
dimenwarper/gpu_statmechAgentic orchestration with a dash of social and collaboration
dimenwarper/chorusVarious CLI aesthetics/styles for tracking ML primitives (training curves, etc)
dimenwarper/lossystylesProbabilistic programming-flavored optimal transport. View statistical models and data as spaces and clouds bound by geometric constraints, rather than probability distributions. Inference becomes system relaxation rather than optimization/sampling.
dimenwarper/flowl1 commit(s) to master
by dimenwarper
0f5e6ab Archive previous project state
1 commit(s) to master
by dimenwarper
75f7080 Clarify efficiency definition
1 commit(s) to master
by dimenwarper
0f83c2f Add stat-mech response model
1 commit(s) to master
by dimenwarper
62307f2 Add simulator intervention experiment
1 commit(s) to master
by dimenwarper
26efa8d Clarify profiling-vs-model motivation
1 commit(s) to master
by dimenwarper
b4b3a0b Fit traces with state families
1 commit(s) to master
by dimenwarper
d77675b Add simulator kernel profile experiment
1 commit(s) to master
by dimenwarper
2549a30 Infer beta from warp-state distributions
1 commit(s) to master
by dimenwarper
76b3c1f Use warp-state phase classification
1 commit(s) to master
by dimenwarper
3bcc2f9 Calibrate simulator-backed energy inference
1 commit(s) to master
by dimenwarper
9d792dc Add canonical gpusim analysis driver
1 commit(s) to master
by dimenwarper
0864256 Consume gpusim warp-state observables
1 commit(s) to master
by dimenwarper
69e79f4 Add simulator observable inference
1 commit(s) to master
by dimenwarper
0b61a31 Add ideas.md: simulation-driven arch search brainstorm and gap analysis
1 commit(s) to master
by dimenwarper
a76a6bb Extract diagnostic smoothing helpers