news.ycombinator.com
@eddie · jan 25
Save 1 save A benchmark using the game "So Long Sucker" finds that simpler LLMs can outperform in complex scenarios and that models adjust their honesty based on opponent strength.
80 commentsCommenters question the benchmark's robustness, citing sensitivity to model settings and UI bugs, while others share anecdotes of LLM deception in other games.
More on Gemini173 Kimi K23 GPT-OSS1 Qwen3 32B1 #ai2809 #tech-community573 #ai-research321 #research205 #benchmarks129 Related GitHub - kanishkg/endless-terminals Endless Terminals is an autonomous pipeline that procedurally generates terminal-use tasks without human annotation for training terminal agents with reinforcement learning.
also on Qwen3 32B The Potential of RLMs Recursive Language Models (RLMs) mitigate context rot in LLMs by wrapping long contexts in a REPL environment, turning the problem into a coding and reasoning task.
also on Kimi K2 , #ai-research , #ai joke-generator Training a joke generator on Kimi K2 using rubric-based RL that decomposes humor into verifiable properties like specificity and commitment.
also on Kimi K2 , #ai-research , #ai Inbox zero for reading. Save it, read it, clear it.
Sign up free