IEEE Spectrum
18 Jul 2026
Sarah Downs, now a Ph.D. student at Texas A&M, created a force-based insertion algorithm that lets a robotic arm assemble satellite antennas in zero gravity without relying on cameras. Her work, done with NASA and the U.S. Air Force during her master’s, addresses peg‑in‑hole insertion and compensates for reaction torques in space.
VentureBeat AI
17 Jul 2026
A VentureBeat Pulse survey of 157 enterprises finds half deployed agents that passed internal evaluations but later failed in customer-facing production, and only 5% fully trust automated evals. Two-thirds are allowing or engineering zero-human-in-the-loop deployments despite fragmented, provider-led tooling and limited real-time output quality checks.
IEEE Spectrum
15 Jul 2026
Researchers recovered ELIZA’s original source code from MIT archives and find it was a versatile platform supporting many scripted personas beyond the famous 'Doctor' therapist. The code shows features like script editing, contextual memory, and teaching/tutor modes that complicate the traditional story of ELIZA as a simple pattern matcher.
TechCrunch AI
15 Jul 2026
Miles Wang, a researcher at OpenAI, is reported to be in funding discussions to launch an AI-driven drug discovery startup valued at about $2 billion. The talks suggest investor interest in using AI to accelerate breakthroughs in life sciences, according to TechCrunch AI.
IEEE Spectrum
14 Jul 2026
Security researcher Dave Kuszmar demonstrated multiple prompt-based exploits that bypass safeguards in major large language models, extracting instructions for weapons, drugs, and other harmful activities. He found the vulnerabilities across most commercial LLMs, says disclosure efforts to companies and agencies were largely ignored.
MIT Technology Review
14 Jul 2026
Anthropic reports a new technique that reveals aspects of its models’ internal reasoning, offering a glimpse into how answers are formed. The piece examines what this finding demonstrates and cautions about its limits for understanding broader ‘world models.’
MIT Technology Review
14 Jul 2026
Anthropic released research exploring a novel phenomenon in large language models, detailing specific behaviors the team observed and the experimental methods used. The article also highlights limitations and cautions, noting the findings do not prove broad new capabilities or conscious states in AI.
IEEE Spectrum
13 Jul 2026
X Square Robot outlines an open embodied-AI stack combining curated interaction data, an event-oriented world model (WALL-WM), and a pretrained action model (Wall-OSS-0.5) with a semantic action tokenizer. The company emphasizes data validity through physical replay and claims cross-embodiment transfer and deployable pretraining before fine-tuning, while noting most results come from its own robots and benchmarks.
