Debate Training Reduces Reward Hacking in RLAIF
A new paper from the GDM Amplified Oversight team shows that debate training reduces reward hacking in reinforcement learning from AI feedback (RLAIF), recovering about 45% of the gap between the peak…
A new paper from the GDM Amplified Oversight team shows that debate training reduces reward hacking in reinforcement learning from AI feedback (RLAIF), recovering about 45% of the gap between the peak…
An engineer has built the OpenTelemetry (OTel) Instrumentation Wizard, a Streamlit and Python tool that automatically injects production-grade OTel traces, metrics, and logs across multiple programmin…
A LessWrong post argues that common frames for understanding AI bargaining problems are confused and proposes mechanistic explanations for why powerful AIs might fail to coordinate, emphasizing the ne…
A developer argues that blockchain technology, despite its association with NFTs and crypto scams, may be the necessary infrastructure to solve the provenance crisis created by generative AI. The post…
IBM cut its full-year 2026 revenue growth forecast to 4%-5% after second-quarter revenue of $17.16 billion missed the $17.58 billion consensus and Z mainframe sales plunged 42%, which CEO Arvind Krish…
IBM lowered its full-year revenue forecast after warning that earnings would fall short of expectations, citing a slowdown in its consulting business. The company said it is turning to artificial inte…
A 13-year-old developer released 'The Sword of Ghix', a retro game created with AI-assisted tools, where players help Bob collect sword pieces to save the multiverse from King Krad.…
Advanced AIs may use credible commitments unavailable to humans when bargaining over resources, according to a new model based on program equilibrium. The model outlines a two-phase process where agen…
Atoms, a new vibe coding tool built by the team behind MetaGPT, uses a multi-agent AI system to handle the entire product lifecycle—from market research and app development to deployment and marketing…
The placeholder names `foo` and `bar`, known as metasyntactic variables, originated from early-20th-century comics and WWII-era military slang, not from computer science. Their exact etymology is cont…
Two probability puzzles involving two fair coins and partial information from Alice yield conflicting answers depending on how Bob interprets the statement "the left coin is Heads." The paradox arises…