Recipes for Steering and Scaling LLMs via Sampling
Source
arxiv.org
Author
Jiajun He, Zongyu Guo, Jos\'e Miguel Hern\'andez-Lobato, Yuanqi Du
Date
Why it matters
A principled alternative to Best-of-N for raising output quality at inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → time — steer toward a reshaped model distribution with no reward model or external supervision, and better scaling behavior than MCMC baselines.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
articleAccelerating LLM Inference via Vector Index Based Output EmbeddingsMartin Loretz, Sepp Hochreiter
articleWhy Better Models Can Create Riskier Systems: Evidence from LLM Agents in Financial MarketsJillian Ross, Eric So, Zoe De Simone, Charles Pozniak, Andrew W. Lo
articleA Year in LLM Serving: Workload Evolution, Caching and Load-BalancingWilliam Nixon, Jon Durbin, Florian Standhartinger, Haryadi S. Gunawi, Juncheng Yang