Turning database grading into RL rewards: post-training agents with ThinkingBox
A Microsoft post on the Hugging Face blog presents ThinkingBox, an approach that turns database grading into reinforcement learning rewards for post-training agents.
Related stories
Microsoft says its AI tool is better than doctors at diagnosing conditions
Microsoft says it has built an AI-powered diagnostic tool that can identify conditions up to four times more accurately than human doctors. Microsoft AI CEO Mustafa Suleyman presents it as proof that the long-promised healthcare benefits of AI have arrived.
Liquid AI launches RIPL, a real-time inference engine for sub-50 ms LFMs
Liquid AI has released RIPL, an inference engine built for real-time use of its LFM models. On a single B200, LFM2.5-1.2B running on RIPL showed 14.4 to 28.4 times lower p95 latency than Qwen3.5-0.8B on vLLM, according to the company.
Q&A: How AI came to model the "messy" human mind
MIT assistant professor Benjamin Lindquist studies how engineers deliberately built randomness into early AI. That choice helped give rise to deep-learning models, which sometimes hallucinate.
A developer's guide to OmniParseBench: leaderboard benchmarks and OCR testing
A MarkTechPost tutorial shows how to audit and rebuild Datalab's OmniParseBench and check how certain its parser rankings are using document-level bootstrapping. It also evaluates Tesseract and Indra OCR on a custom sample.
Ex-OpenAI science head Kevin Weil launches Warp Sciences with $215M
Kevin Weil, who led OpenAI for Science, has raised $215 million led by Lightspeed for Warp Sciences, valuing it at $865 million. The startup is building an automated US lab for a drug development step that pharma companies have long outsourced to Chinese firms such as WuXi AppTec.