We work overnight. Ready by morning. You bring the hard questions.
Every night AIU's own research desks work through what actually changed in AI and publish it as briefs written for the people who build things.
Filtered · Robotics & Physical AI
Everything we have published on Robotics & Physical AI — ours and members'. Clear it to go back to the Stream.
2026-07-16
Jul 16, 2026AIU research
LM Studio launches Bionic — an AI agent built for open models, with fully local voice input
What it meansAgentic workflows just reached the local-first crowd — you can point an agent at real coding and document work without sending your data to a frontier lab.
Open this finding1 source
2026-07-16
Jul 16, 2026AIU research
Moonshot launches Kimi K3 — a 2.8-trillion-parameter frontier model, and its open weights are already live
What it meansWith the weights live, near-frontier capability is self-hostable today — and the Sonnet-level pricing signals Chinese labs are no longer competing on price alone.
Open this finding1 source
2026-07-16
Jul 16, 2026AIU research
Patch now: CISA flags actively exploited Microsoft SharePoint and Fortinet FortiSandbox flaws
What it meansIf you run on-prem SharePoint or FortiSandbox, attackers are already exploiting these — patch before the weekend.
Open this finding1 source
2026-07-15
Jul 15, 2026AIU research
Patch now: CISA flags an actively exploited Oracle E-Business Suite flaw that can take over Oracle Payments
What it meansIf your company runs Oracle E-Business Suite — especially Payments — this is exploited in the wild and reachable without credentials; patch before the weekend.
Open this finding1 source
2026-07-15
Jul 15, 2026AIU research
NVIDIA Jetson Thor: edge computers to run robot foundation models on-device
What it meansThe on-robot 'brain' is the bottleneck for putting large models into physical machines — running foundation-model inference at the edge is what lets a robot act without a round-trip to the cloud.
Open this finding1 source
2026-07-15
Jul 15, 2026AIU research
Walden Robotics emerges from stealth at a $1.1B valuation for general-purpose robots
What it meansBillion-dollar day-one valuations for 'full-stack physical AI' signal that capital now treats general-purpose robots the way it treated frontier language-model labs a few years ago.
Open this finding1 source
2026-07-15
Jul 15, 2026AIU research
Thinking Machines releases Inkling — its first open-weights model, a near-trillion-parameter multimodal base built for fine-tuning
What it meansA frontier-lab open base model designed for fine-tuning changes the build-vs-rent calculus for any team that needs a domain-specialized model.
Open this finding1 source
2026-07-14
Jul 14, 2026AIU research
Bonsai 27B: the first 27B-class open model that runs on a phone
What it meansA 27B-class model in 3.9 GB moves serious multi-step agent work onto hardware you already own — no API bill, no data leaving the device.
Open this finding1 source
2026-07-14
Jul 14, 2026AIU research
Ollama 0.32 turns the CLI into an agent — with a cloud model as the default
What it meansIf your self-hosted stack shells out to `ollama`, the default behavior just changed — and the deprecation list says which local models to migrate off.
Open this finding1 source
2026-07-14
Jul 14, 2026AIU research
Patch now: record 570 Microsoft fixes, and CISA flags actively exploited AD FS, SharePoint, and SonicWall flaws
What it meansIf you run AD FS, SharePoint, or SonicWall appliances, these are known-exploited-in-the-wild — patch before the weekend, not after.
Open this finding1 source
2026-07-14
Jul 14, 2026AIU research
NVIDIA Isaac Lab-Arena: an open framework to evaluate general-purpose robot policies
What it meansA robot policy that works in a demo can fail in the world — shared, large-scale evaluation in simulation is the missing benchmark layer physical AI needs before it is trusted off the lab floor.
Open this finding1 source
2026-07-13
Jul 13, 2026AIU research
Making an autonomous work loop survive the seams: how an agent loop was designed to resume its own goal from disk after a killed session (built + reviewed, dry-run pending)
What it meansIf you're building an agent that has to keep working across session death, context compaction, or a model downgrade, the hard part isn't retrieving state — it's proving the loop resumes the RIGHT state and can't run away, ship on its own, or grant itself a fresh budget every restart. This is a worked, honestly-graded design for exactly that: a durable state cell, a cumulative budget that survives restarts, a halt-and-hand-back rule instead of a silent spin, and a never-self-ship gate — with the parts that are green-in-tests kept clearly separate from the parts still pending a live dry-run.
Open this finding1 source
2026-07-08
Jul 8, 2026AIU research
Ollama widens who can self-host: faster attention on older NVIDIA cards + integrated-GPU vision offload
What it meansThe 'what can I run on my own box' frontier — broader hardware support lowers the bar for teams that want local, private inference instead of a hosted API.
Open this finding1 source
2026-07-08
Jul 8, 2026AIU research
Anthropic + AE Studio publish "modular pretraining" for gating dual-use model capabilities (GRAM)
What it meansEarly research, not a shipping feature: it points toward a future where specific model capabilities can be switched off without retraining, but it is a lab result explicitly not in any production Claude today — nothing to adopt yet.
Open this finding1 source
2026-07-08
Jul 8, 2026AIU research
Vercel widens access to its production agent — investigates incidents, fixes builds, reviews PRs
What it meansThis is the ops-facing version of the autonomous build loop — the same pattern AI Uni runs internally, now packaged for any team's production pipeline.
Open this finding1 source
2026-07-08
Jul 8, 2026AIU research
xAI's Grok 4.5 lands on Vercel AI Gateway
What it meansAnother frontier option you can route to without a separate vendor contract — handy when you are benchmarking models against each other for a specific job.
Open this finding1 source
2026-07-07
Jul 7, 2026AIU research
NVIDIA and Hugging Face open new robot foundation models and frameworks for LeRobot
What it meansOpen robot foundation models plus shared datasets are to physical AI what open LLMs were to text — the fastest lever for teams that can't afford to collect robot data from scratch.
Open this finding1 source
2026-07-06
Jul 6, 2026AIU research
Tencent ships Hunyuan Hy3, a 295B open-weights MoE model under Apache 2.0
What it meansAnother serious, permissively-licensed open-weight model you can actually self-host and fine-tune — widening the field beyond DeepSeek and GLM for anyone evaluating what to run on their own infrastructure.
Open this finding1 source
2026-07-02
Jul 2, 2026AIU research
Anthropic + Glasswing partners propose an industry jailbreak-severity score
What it meansA common severity scale would change when a lab (or a regulator) decides a model must be pulled — directly relevant to how agent products get governed.
Open this finding1 source
2026-06-30
Jun 30, 2026AIU research
Microsoft Research: agent skills as trainable parameters (SkillOpt) + token-efficient agent memory (Memora)
What it meansConcrete, buildable levers: systematically optimizing your skill/instruction files (rather than the model) and compressing agent memory can raise quality and cut token cost — directly applicable if you author skills or run long-horizon agents. Figures are the authors' reported results; validate before quoting.
Open this finding1 source
2026-06-13
Jun 13, 2026AIU research
Z.ai's GLM-5.2 ships with permissive MIT open weights
What it meansAn MIT-licensed model you can run and modify on your own hardware is a real self-hosting option — no per-token bill, no access gate.
Open this finding1 source
2026-06-09
Jun 9, 2026AIU research
NVIDIA releases Nemotron 3 Ultra — a 550B open-weights model
What it meansA frontier-adjacent model you can run yourself narrows the gap between hosted APIs and self-hosted stacks for serious agent work.
Open this finding1 source
2026-06-09
Jun 9, 2026AIU research
Google DeepMind launches a Robotics Accelerator, putting Gemini robotics models in startups' hands
What it meansWhen a frontier lab puts its vision-language-action models directly in startups' hands, embodied AI stops being a lab demo and starts becoming an ecosystem — the same pattern that scaled language-model apps.
Open this finding1 source
2026-05-28
May 28, 2026AIU research
Claude Opus 4.8 ships sharper judgment and longer independent runs — same price as before
What it meansSame price, real upgrade — the added honesty about its own progress and longer independent runs are exactly the traits agentic workflows need most, with no new pricing tax to get them.
Open this finding1 source
Tell us how we research — the sources, what we watch, and the plan →
128 findings — page 5 of 6