Lab
Research, experimentation and open source: hardware, IoT, robotics, biotech and edge AI.
7 published articles
Agent evaluation
Sandbox benchmarks are hiding how agents really fail, HKU just built the fix
UniClawBench evaluates proactive agents across five fundamental capabilities in 400 bilingual real-world tasks, using live Docker containers and a three-agent closed-loop evaluation. It disentangles base model abilities from framework choices, revealing where agents truly break.
2026-07-13
Robotics Research
Robots that adapt without retraining? This new framework might actually deliver
A new framework called In-Context World Modeling (ICWM) allows robot policies to adapt to novel setups, like different camera angles or robot bodies, without retraining. By treating system identification as an in-context problem, ICWM uses task-agnostic interactions to infer world dynamics before task execution, outperforming standard VLA baselines in simulations and real-world tests.
2026-07-11
AI Infrastructure
The next trillion-dollar bottleneck in AI isn't algorithms, it's power and cooling
The explosive growth of generative AI is exposing a hard limit: the physical infrastructure needed to train and serve models at scale. Hyperscalers, chip designers, and a wave of energy-focused startups are pouring capital into rethinking data centers, cooling, and power delivery, betting that the next frontier of AI competition is infrastructural, not algorithmic.
2026-07-10
Smart Lock Review
The Schlage Sense Pro locked my front door before I could unlock it, then I fixed it
The Schlage Sense Pro uses ultra-wideband to unlock your front door as you approach, no keys, no codes, no phone taps. It is the first hands-free system this reviewer trusts, but its $399 price tag and exclusive reliance on Apple Home Key make it a hard sell for Android users, at least for now.
2026-07-09
Data Engineering
JPEG at quality 92: the boring data engineering choice that made Photoroom's 7B model work
Photoroom releases the fourth part of its PRX model development series, detailing a data pipeline that prioritizes breadth and consistency over per-image perfection. The post reveals that JPEG compression at quality 92 is indistinguishable from PNG for training, and that the fastest captioner won over the most accurate one when scaling to hundreds of millions of images.
2026-07-06
Genomic Medicine
Talos shows automated genome reanalysis can deliver new diagnoses in weeks, not years
Talos reframes genome reanalysis as a continuous, automated program rather than a rare manual event. Across a prospective cohort of 4,735 undiagnosed patients, it yielded 241 new diagnoses (5.1% additional yield) within weeks of new evidence being published, while requiring analysts to review only one new variant per 200 patients per month.
2026-07-05
Benchmarks & Tests
Treble Technologies and Hugging Face Launch FFASR Leaderboard for Far-Field Speech Recognition
The new FFASR Leaderboard from Treble Technologies and Hugging Face evaluates ASR models across nine conditions including reverberation, background noise, and microphone distance. Early submissions show far-field WER at low SNR is several times higher than near-field performance, highlighting the need for acoustically robust models.
2026-07-02