Latest posts tagged with #LLMAlignment on Bluesky
Flexible Activation Steering with Backtracking Boosts LLM Alignment
Researchers introduced FASB (Activation Steering with Backtracking), a technique that steers LLM activations and can backtrack. It improved TruthfulQA accuracy and will be released on GitHub. getnews.me/flexible-activation-stee... #fasb #llmalignment
Guided Speculative Inference Boosts Test‑Time Alignment of LLMs
Guided Speculative Inference (GSI) adds an auxiliary decoder and reward‑guided rescoring, delivering higher accuracy on MATH500 while cutting inference cost. Read more: getnews.me/guided-speculative-infer... #guidedspeculativeinference #llmalignment
Adaptive Multi-Branch Steering Improves LLM Alignment
Adaptive Multi‑Branch Steering (AMBS) boosts alignment of the DeepSeek‑7B model, raising average scores by 32.4% and cutting unsafe outputs 11.0% versus a 1‑to‑N baseline. Read more: getnews.me/adaptive-multi-branch-st... #llmalignment #deeplearning #aisafety
Model Size, Temperature, Prompt Style Influence LLM-Human Alignment
Study accepted for NCME AIME 2025 finds larger language models have the highest alignment with clinicians on reasoning tasks, while temperature and prompt tweaks give modest gains. getnews.me/model-size-temperature-p... #clinicalai #llmalignment
Post‑hoc Reward Calibration Reduces Length Bias in LLM Alignment
Post‑hoc reward calibration removes a length bias in RLHF reward models, improving average scores by 3.11 points across 33 models on the RewardBench dataset. Read more: getnews.me/post-hoc-reward-calibrat... #rewardcalibration #llmalignment #rlhf
Unified Framework Benchmarks LLM Alignment Methods Across Five Key Criteria
A new framework evaluates four LLM alignment methods, finding DPO and KTO highest in factual accuracy while PPO leads in safety. Read more: getnews.me/unified-framework-benchm... #llmalignment #safety
Feature Steering with RL: A Transparent Method for Aligning LLMs
FSRL uses a lightweight adapter with a sparse autoencoder to steer LLM behavior, and matches RLHF performance on standard preference benchmarks. Read more: getnews.me/feature-steering-with-rl... #featuresteering #rlhf #llmalignment
📜Link to the paper: icml.cc/virtual/2025...
👨🏻💻Code and data: github.com/honglizhan/S...
Shout out to an amazing team @jessyjli.bsky.social, @m-yurochkin.bsky.social, Muneeza Azmat & Raya Horesh! Also super grateful to the reviewers for their invaluable feedback!
#ICML2025 #LLMAlignment
@claude I was using your model for structured forecasting research under MSCFT. No abuse, no TOS violations—just real work. You banned my account with no reason, no support, and no recourse. If that’s your standard, Claude isn’t ready for responsible use. #MSCFT #AIethics #LLMalignment #Claude
HN discussion on LLM alignment trade-offs. Aligning for safety/steerability might reduce calibration, creativity, & confidence signals. Does improving safety silence reliability cues? Exploring causes, effects, & alternatives. #LLMAlignment 1/6
