Lilian Weng, co-founder of Thinking Machines Lab, posted a cryptic message on July 27 calling her decision 'hard and sad,' without specifying its nature, following the July 15 release of the company's…
Thinking Machines Lab released Inkling, an open-weights multimodal foundation model family with 975B total parameters and 41B active, under the Apache 2.0 license. The model supports up to 1M token co…
Mira Murati's Thinking Machines Lab released Inkling, a 975-billion-parameter open-weight mixture-of-experts model that activates only 41 billion parameters per task and natively handles text, image, …
Thinking Machines Lab, an AI startup founded by former OpenAI executives including Mira Murati, has released its first open-weight model, Inkling, a 975-billion-parameter model trained on audio, video…
At the AI Engineer World's Fair 2026, AI engineering has matured beyond building agents to engineering the systems around them, according to conference coverage by Richard MacManus. The shift is exemp…
OpenAI launched GPT-5.6 as a three-model family — Sol, Terra, and Luna — with Terra performing above Anthropic's Fable tier at half the output tokens and a quarter of the cost. The model introduces an…
OpenAI's head of safety systems, Johannes Heidecke, is leaving the company after an internal restructuring merged safety and research teams under a single leader, marking the second time in two years …
OpenAI has restructured its safety organization, placing safety teams under the VP of Research after a wave of leadership departures, including CTO Mira Murati and Chief Research Officer Bob McGrew. T…
OpenAI's head of safety systems, Johannes Heidecke, is leaving the company amid a reorganization that integrates safety and research teams. The departure follows the launch of GPT-5.6, which showed co…
Lilian Weng published a survey on harness engineering for self-improving agents, documenting how agents that optimize their own scaffolding face provenance and trust issues. The Darwin Gödel Machine p…
Lilian Weng, cofounder of Thinky, published a summary of 35 papers on harness engineering for recursive self-improvement (RSI), arguing that harness improvements will remain essential even as core mod…
An engineer argues that the frontier of AI productivity lies not in new models but in an architecture built around memory, planning, and tools. The post distinguishes three levels of memory—in-context…
Scaling laws predict training loss as model size, dataset size, and compute scale, but their practical application is sensitive to hyperparameter choices like learning rate. Lilian Weng's post highlig…