6+ day, 22+ hour ago   (146+ words) YC Paper Club put harnesses center stage: Prime Agent, OpenJarvis and QM show how self-improving scaffolding beats fixed prompts. Prime Agent hit 95.5% on ARC-AGI using self-improving harnesses on Claude Opus, up from 30% on the private set without scaffolding. The demo…...

1+ week, 5+ hour ago   (400+ words) OpenAI, WAN-IFRA and AIRPPU will train and fund AI pilots for ten Ukrainian regional newsrooms, with API credits and a Masterclass plus Catalyst. Ten Ukrainian regional newsrooms will get OpenAI API credits and hands-on help building AI workflows under the…...

1+ week, 2+ day ago   (321+ words) Databricks’ Proteus hit 1.8, 5.2x on Qwen 3.5 122B kernels by specializing for runtime shapes, but only after fixing reward-hacking in validation. Databricks says its Proteus harness generated Qwen 3.5 122B kernels that ran 1.8 to 5.2 times faster than the best vLLM implementations. Profiles of the companies…...

1+ week, 5+ day ago   (595+ words) ElevenLabs brings its full API to the terminal with agents-as-code, versioned configs and agent-first tooling, tightening its lead over voice and video tooling rivals. ElevenLabs CLI v1 arrived Aug 24, and it does something simple but ambitious: it puts the entire ElevenLabs…...

2+ week, 33+ min ago   (589+ words) Google DeepMind's Zubin Ghahramani argues AI needs explicit Bayesian uncertainty, not just scale, to be trustworthy in the real world. Google DeepMind laid out the mathematics of AI uncertainty this week, with Cambridge professor and frontier AI co-lead Zubin Ghahramani…...

2+ week, 2+ hour ago   (735+ words) Origin builds Axis, an AI that designs regulatory DNA sequences controlling gene expression in disease cells. It outperforms Google DeepMind's AlphaGenome by 6.7% and is building a proprietary validation dataset that could create a multi-year lead in cell and gene therapy…...

2+ week, 2+ day ago   (582+ words) Valgo builds probabilistic simulation tools that let insurers price risk for autonomous systems. Stanford PhD founders who co-authored the textbook on safety validation, paired with a 12-year actuary, are tackling the hardest unsolved problem in physical AI deployment. Insurance is…...

2+ week, 3+ day ago   (408+ words) StartupHub.ai LLMs Fail to Write Fast Multi-GPU Kernels Simran Arora from Together AI discusses the challenges of multi-GPU kernel development and why current LLMs struggle to optimize them, despite ongoing research. In the ever-evolving landscape of AI hardware, optimizing…...

2+ week, 3+ day ago   (159+ words) StartupHub's Data Trust & Reputation score for Physicalloop is 26 out of 100, based on site security posture and privacy practices. Physicalloop was founded in 2026. Ship of Theseus is developing a fleet of biologics based on the homeobox (HOX) family of transcription factors…...