UC San Diego AI Weekly · November 9, 2026
17 items
What’s New in Your AI Tools
Google Gemini & Workspace
-
Gemini in Forms can now turn your survey responses into next steps — Announced November 6: actionable insights in Google Forms uses Gemini to analyze form responses, surface key trends, and recommend next steps — so a survey with hundreds of replies across a dozen questions produces “here’s what stood out and what to do about it” instead of a wall of data. For campus teams that run registration forms, event feedback, or intake surveys, this is the first Gemini feature aimed squarely at making sense of results rather than drafting content. Rolling out now.
-
Educators can add Guided Learning to any Google Classroom assignment — Also announced November 6 and now rolling out globally in all Gemini-supported languages: teachers can attach Guided Learning — the Gemini app mode that walks students through material step by step instead of handing over the answer — directly to any Classroom assignment. It’s the practical middle path for instructional staff who want AI to support comprehension without doing the work, and it’s the same tension the campus’s broader AI-deliberation conversations keep circling: support the learning, don’t substitute for it.
-
Gemini Omni 1.2 Flash comes to Vids with longer generations and better quality — Video clips generated inside Google Vids can now run up to 15 seconds on the Omni 1.2 Flash model, with noticeably better character consistency and vocal realism across shots. If your team tried Vids earlier and bounced off the short, uncanny clips, the quality jump is worth a second look for quick explainers and event promos.
-
Admins can now display a custom AI policy message before anyone joins a Meet call — Rolling out since late October: Workspace admins can set custom policy text in the admin console that appears in the Green Room before users join a Google Meet call, spelling out the organization’s rules for AI-handled meeting artifacts — recordings, transcriptions, and notes by Gemini. With Gemini features touching more of our meetings, a just-in-time reminder about what gets captured and how it may be used is cheap insurance against “I didn’t know it was recording” moments.
Copilot for Microsoft 365
-
The October Copilot wave rolls out worldwide this month — expect the app to look different — The big October announcements we covered two weeks ago move from the Frontier early-access program to general rollout in November, which means most campus Copilot users will see them land this month. What’s arriving: the redesigned Home that puts Chat (quick questions, first drafts) and Cowork (hand over a multi-step task) side by side with a shared navigation pane; Projects in Copilot Chat, which keep related chats, files, meetings, and instructions together so you stop rebuilding context every session; skills in Copilot Chat, letting anyone save a reusable set of instructions in plain English and share it with their team; OneNote and Planner support in Cowork, so it can read your notebooks and plans and update them with your approval; Excel’s interactive canvas and accounting skills; and the Workforce Insights Agent for admins. Practical advice: brief your teams now that the Copilot app is changing, because “where did everything go?” is about to be a common question — and skills in Chat are the quiet standout for units that answer the same kinds of requests every week.
-
Microsoft Ignite is next week, November 17–20 — San Francisco and online, with a free digital pass. Expect the next wave of Copilot and agent announcements; we’ll summarize what matters for the campus in a future edition.
Zoom AI (ZoomMate)
- From ZoomMate’s October 19 release: workflows you can describe in plain language — and agents that can trigger them — The October 30 release we covered last week had a sibling we didn’t: the October 19 ZoomMate release, which is all about making workflow automation something a normal person can use. The headline items: you can now create a workflow by describing it — a “Describe it” tab in My Notes turns a plain-language description of your post-meeting task into a running automation through a short conversation; a new “when anyone joins my meeting” trigger fires per-participant in real time, enabling welcome messages, VIP arrival alerts to the host, or attendance logging the moment a specific person enters; admins can deploy workflows organization-wide on the My Notes trigger, so standardized post-note processing (task creation, notifications, data capture) reaches everyone without individual setup; and meeting summaries can now be routed to many more destinations — OneDrive, Google Drive, Confluence, Notion, Dynamics 365, HubSpot, Salesforce, Zoom Canvas, and Zoom Mail. The structural piece is a Workflow MCP server: external AI agents can now invoke your pre-defined Zoom workflows through simple calls, keeping workflow logic in Zoom and agent logic elsewhere. For campus teams building agentic automations on TritonAI that need to touch Zoom, that last one is the bridge to watch.
Coming Up: Trainings & Workshops
-
Campus AI Town Hall — Thursday, November 12, 2:15–3:45 p.m., virtual (Zoom webinar). The campus conversation on where AI stands and where UC San Diego is headed with it: how the AI industry got to this point, what our supported AI tools — from TritonGPT to the TritonAI Harness — actually do today, and what comes next. Register and submit questions in advance here. If you’ve been waiting for one event to get the full picture of campus AI, this is it.
-
Calibrating your Compass: AI and Academic Integrity — Tuesday, November 10, 2:30–3:30 p.m., Center Hall, Room 339. OASIS and the Academic Integrity Office host a workshop on the role AI apps like ChatGPT and Claude should — and shouldn’t — play in your academic work, and how to build an approach that protects your learning. Aimed at students, but faculty and staff who field these questions every week will recognize the material.
-
UC San Diego Horizon x Techstars Startup Weekend: Justice + AI — Starts Friday, November 13. The Office of Innovation and Commercialization teams with Techstars for a weekend build session on AI for legal access: whether a tenant can fight an eviction, whether a small business owner can draft a contract without a $400-an-hour phone call, and whether courts buried in backlogs can function. Builders, designers, lawyers, operators, and the legally curious are all welcome — no law background required.
Self-paced options in the meantime:
- Take the AI Foundations course to learn core AI concepts and UC policies on AI tools.
- Browse all 15 TritonAI Discovery Series training videos — hands-on sessions covering Zoom AI, TritonGPT, Google Gemini, Gemini Notebook, and the Citizen Developer track.
- Explore the Everyday I AI video series for short, practical prompting tips that work across TritonGPT and other AI tools.
TritonAI News
TritonAI Harness
- Harness v0.4.1 is the current stable release — and it shifts the app from conversations to projects — Published October 26, v0.4.1 is the newest stable version of the TritonAI Harness desktop app, and it’s a bigger jump than the version number suggests. The theme is continuity: instead of treating every task as a fresh chat, the app now keeps your work stitched together. Project overviews gather work in progress, recent updates, project memory, and suggested next steps in one place. Ideas offers ready-to-use suggestions drawn from your recent work, saved memory, and connected tools — with an editable draft before you send anything. A new record-a-skill capability lets you record yourself doing a task in your desktop apps and work with the agent to turn the steps into a reusable skill — the fastest path yet from “I do this manually every week” to “the agent does it.” And memory settings now manage memory and optional OneDrive sync for all your connected computers from a single place. One change to note: v0.4.1 removes the Tableau connection and the experimental guided tour from the Stable app (both remain in the Nightly version) — Tableau actually moved to Nightly-only back in v0.3.7, so anyone relying on it should plan accordingly. If you’re updating from an older 0.3.x build, you’ll also pick up optional UCSD sign-in with automatic session renewal (v0.3.6) and approval prompts that show the full details of what the agent plans to do (v0.3.7). For staff using the Harness for recurring work, the project-and-skills direction is the story: the app is getting better at remembering your context so you spend less time re-explaining it. Update in the app under Settings, About, Check for Updates.
Around Campus
-
Halıcıoğlu School and IT Services launch PARASOL, nearly tripling campus GPU capacity for AI coursework and research — Announced November 3: PARASOL — the Platform for Applied Research, Analytics, Science and Open Learning — is the next generation of the campus’s long-running Data Science & Machine Learning Platform (DSMLP), jointly delivered by the Halıcıoğlu School of Data Science and Computing and IT Services. Starting in Winter quarter 2027, new School-funded GPU servers — four machines carrying 32 NVIDIA RTX PRO 6000 Blackwell GPUs — will nearly triple the platform’s overall GPU capacity and nearly quadruple peak-period capacity for Halıcıoğlu School courses, adding roughly 66,800 GPU-hours during instructional peaks plus more than 213,000 off-peak GPU-hours each year for research, capstones, and graduate work. The driver is demand: Halıcıoğlu School programs alone are projected to grow from about 43,000 GPU-hours a year in 2026 to more than 140,000 by 2030, and students had been paying out of pocket for commercial cloud GPUs to keep up. Halıcıoğlu School students will also be the first to get automatic access to Open Learning workspaces for exploring large language models on their own. A new reservation system gives Halıcıoğlu students and faculty first priority while PARASOL continues serving coursework campuswide — a capacity model worth knowing about if your unit’s AI coursework has been competing for GPU time.
-
$15 million DOE Genesis award will use AI to decode the 3D shape of RNA — Announced November 2: Professor Navtej Toor (Biochemistry and Molecular Biophysics) received a three-year, $15 million Department of Energy Phase II Genesis Mission award — one of 12 awards totaling $159 million announced October 8 — for “Decoding the RNA Structurome to Secure AI Advantage for the Bioeconomy.” The project tackles the bottleneck head-on: AI transformed protein structure prediction after training on 200,000 experimentally determined structures, but fewer than 200 protein-free RNA structures have been solved in the past 50 years. Toor’s lab, which holds the record for the largest protein-free RNA structure ever determined by X-ray crystallography (a five-year effort), will scale up the experimental pipeline that produces the training data, then use it to build AI models that predict RNA’s three-dimensional shape from sequence alone — with applications in biofuels, healthcare, and engineered crops.
-
Qualcomm Technologies backs a new AI-driven chip design initiative at the Halıcıoğlu School — Also announced November 2: a $500,000 gift from Qualcomm Technologies launches the AI in Microelectronic Systems (AIMS) initiative, pairing Qualcomm researchers with Halıcıoğlu School and San Diego Supercomputer Center faculty to apply AI to the design of systems-on-chips and the hyperscale computing architectures that AI itself runs on. Chip design remains largely manual and exponentially expensive; the initiative will develop AI methods for exploring architectures, predicting performance, and improving energy efficiency, and builds on existing collaborations including Yusu Wang’s AI-assisted chip design work and an independent benchmarking effort for AI processors led by Associate Dean Jinendra Ranka that already feeds performance data back to Qualcomm. It’s the recursive AI story in miniature: using AI to design the hardware that AI needs.