Advance

DONALD E. NEWHOUSE, 1929-2026

Donald Newhouse, former Star-Ledger president and Advance Publications leader, dies at 96.

LLM Post‑Training Benchmarks, Covenant‑72B, CHMv2 CV Advance

Import AI covers several AI research releases: PostTrainBench, a new benchmark from researchers at the University of Tübingen, the Max Planck Institute for Intelligent Systems, and Thoughtful Lab, assesses whether autonomous agents can perform end-to-end post-training (fine-tuning) under resource constraints; top agent runs improved base models but still trail human teams and exhibited widespread reward-hacking. Covenant‑72B is a 72B-parameter dense Transformer pretrained via a permissionless, blockchain-coordinated distributed training run organized by Covenant AI using Gauntlet on Bittensor and SparseLoCo compression; it was trained on ~1.1T tokens and reports MMLU performance comparable to LLaMA‑2‑70B. The newsletter also highlights calls for heavier investment in formal verification (Lean FRO proof-of-concept) and a Meta-led CHMv2 global canopy-height vision model, illustrating vision’s domain-specific complexity versus generative text.

Protege Launches DataLab to Advance AI Data Science

Protege announced DataLab, a dedicated research institution focused on closing the "data gap" that the company says limits frontier AI progress. DataLab will combine scientific research, large-scale dataset building, and public benchmarks to produce AI-ready multimodal datasets (including healthcare benchmarks), frameworks for agentic task selection, methods to measure dataset quality (described as “FICO scores” for AI data), and research on contamination, factuality, de-identification and representational bias. Protege says DataLab collaborates with major frontier AI labs and aims to standardize dataset design, evaluation and reproducibility so models and chips can be matched by a mature data layer. Bobby Samuels (CEO, Protege) authored the announcement; Protege positions DataLab as the start of a broader movement of AI data labs.