I might be messing up the technical terms here, but i'm actually using the heightmap approach. The PARC clips carry heightfields as what I called "terrain", so I baked them into a single global heightfield. Just one huge collision trimesh. From there, it's still the same game of mapping each motion to the correct spawn location. PARC's uses mostly blocky stepped geometry. So while there are a handful of resampling artifacts, it mostly works. I used 8000 unique motion/terrain combos, all baked in without any nasty slowdown to simulation time!

Just resampling model responses on WildChat and grading them predicts Petri alignment audits remarkably well! Just compare those two plots. WildChat resampling even captures the Grok trend towards increasing misalignment. Wild!

As @EricBigelow et al. show in the amazing Forking Paths paper, during CoT models can keep track of multiple future answer options. Adapting a resampling approach inspired by @uzaymacar et al.'s Thought Branches, we find similar uncertainty dynamics for behavioral evals. (3/n)

Prima Materia Audio’s Radiocore 4471 for Windows turns clean audio into unstable radio transmissions, signal artifacts, static, dropouts, and degraded textures for producers, sound designers, and experimental resampling workflows. ▶️ audiopluginguy.com/news-prim… #APGNews

a quant at Bridgewater told me something that flipped how i think about backtesting "a backtest is one path through history. one. we need to see ten thousand before we trust anything" that tool is Monte Carlo simulation. invented in 1946 by Stanislaw Ulam while working on nuclear weapons at Los Alamos and most quants don't trust it this video covers 6 reasons why in 90 seconds the core problem: Monte Carlo assumes your returns are independent and normally distributed real markets have fat tails, volatility clustering, and serial correlation simulate without those and your risk estimates are dangerously optimistic a standard Monte Carlo says your max drawdown is 15% add fat tails and it's 35% add correlation between consecutive losses and it's 48% same strategy. three completely different risk profiles depending on which assumptions you feed the simulation this is why quant desks don't run vanilla Monte Carlo they run it with GARCH volatility, regime-conditional distributions, and bootstrap resampling from actual historical sequences retail runs a backtest, sees profit, goes live quant desks run 10,000 simulations, find the 5th percentile outcome, and ask: can i survive this path for 14 months without changing anything? if the answer is no, they don't trade it. regardless of the median outcome > Monte Carlo method: 1946, Los Alamos > used at every major quant desk since the 1980s > the 6 problems: explained in this video, free, 90 seconds > most retail traders have never run a single simulation on their own strategy your backtest is one story the market could have told you there are 9,999 others. some of them end very differently full breakdown in the video below

We address by prompting another LLM to simulate results of agents’ tool calls. Tool simulation can be made quite faithful by giving the simulator access to the original trajectory and a time-matched codebase. With all of those affordances, we found that resampling responses on OpenAI internal Codex traffic results in trajectories that the model can’t distinguish from the real ones!

unfortunately some nerd will say that the resampling to C "ruins" the sound in that case

Prince Broken rare Hip Hop Jazzy Rap Session First Step session Freek Banning dec 2025 recovery and resampling to HQ ! youtu.be/J84vJQvXAwA?si=4RVJ… via @YouTube

Discovery Pro 8.21 is here. Cleaner offline renders with new Ultra Wave Resampling, lower CPU across the engine, analog-style filter envelopes that breathe, and right-click quick menus on every knob. Less friction, more music. Free for all 8.x users. discodsp.com/discoverypro/

iq_tool: A Command Line Tool for Resampling, Filtering, Shifting and Correcting IQ Data Streams rtl-sdr.com/iq_tool-a-comman…

Here’s the quick breakdown of last week's @PolarGrid changelog: 🧠 Enhanced Intelligent Routing: Traffic dynamically routed by model and real-time load. 🎙️ Streaming STT: Real-time partial transcripts as audio arrives. ⚡ 2x Speed: In-process audio resampling cut inference times in half. 📊 Live Dashboards: Deep-dive usage analytics powered by live data. Dive into the full changelog details here: polargrid.ai/changelogs/week…

1. ISL is a tour-de-force in Data Science foundations. The book covers: - Statistical Learning - Linear Regression - Classification - Resampling - Model Selection vs Regularization - Nonlinear Models

Evening Log 02 : • Revisited the statistical foundations of bootstrap resampling. • Analyzed how averaging weakly correlated models suppresses variance. • Reinforced the mathematical basis of bagging and its impact on generalization.

📝Estimator Statistics from Simulation-Free Dirichlet Block-Bootstrap Resampling 👥by Tillmann Rosenow 📖brnw.ch/21x3mgp #BlockBootstrap; #Resampling; #NonParametric; #Parametric