This Month's Latest Tech News in Riverside, CA - Sunday August 31st 2025 Edition

Weekly commentary: Riverside's AI moment - from campus labs to county halls - UC Riverside and Google's UNITE system shifts the region from research novelty to practical defense, using transformer-based analysis of full frames (backgrounds, motion and all) to catch manipulations that traditional face-focused detectors miss; read UC Riverside's announcement UC Riverside develops tool to detect fake videos and SciTechDaily's clear explainer SciTechDaily explainer: how AI can spot fake videos.

The result matters locally: newsrooms, platforms and public agencies in Riverside now face a future where empty rooms and subtle motion glitches can reveal a lie, and practical upskilling - like Nucamp AI Essentials for Work 15-Week Bootcamp - helps reporters, managers, and civic staff learn prompt craft and tool use to verify what they see.

“Deepfakes have evolved. They're not just about face swaps anymore. People are now creating entirely fake videos - from faces to backgrounds - using powerful generative models. Our system is built to catch all of that.” - Rohit Kundu

1) UC Riverside + Google unveil UNITE deepfake detector (CVPR 2025) - At CVPR 2025 in Nashville, UC Riverside and Google introduced UNITE, a transformer-based, “one-model” approach that looks beyond faces to scan full frames, backgrounds, motion patterns and subtle temporal glitches so entirely synthetic scenes can be flagged just as reliably as face swaps; read the detailed ScienceDaily coverage ScienceDaily coverage of UNITE deepfake detector or consult the team's technical write-up on the paper arXiv paper “Towards a Universal Synthetic Video Detector”.

Built on SigLIP features and an “attention-diversity loss” that forces the model to monitor multiple regions, UNITE is aimed at newsrooms and platforms racing to catch manipulations where the only giveaway might be a background shimmer or a tiny motion inconsistency - an advance that could make tampered videos far harder to hide in plain sight.

“Deepfakes have evolved. They're not just about face swaps anymore. People are now creating entirely fake videos -- from faces to backgrounds -- using powerful generative models. Our system is built to catch all of that.” - Rohit Kundu

2) Veritone signs multi-year AI redaction deal with Riverside County Sheriff's Office - In a move to tame growing volumes of digital evidence, Veritone inked a multi-year agreement to deploy its Redact software (part of the iDEMS suite) across Riverside County's sprawling agency, automating the identification and obscuring of faces, license plates and other PII in bodycam, dashcam, CCTV and interview footage to cut down the hours staff spend on manual redaction; read Veritone's announcement Veritone press release on the Riverside County Sheriff's Office AI redaction agreement and learn about the redaction tool itself at the official Veritone Redact AI redaction software product page.

The contract highlights a practical balance many agencies now seek: faster public disclosure and FOIA compliance without sacrificing privacy, and a chance for deputies and analysts to focus on investigations instead of pixel-by-pixel blurring.

“This partnership with Riverside County reinforces our mission to empower public safety agencies with scalable, cost-effective, and impactful AI solutions. As the demand for transparency and timely evidence disclosure increases, Veritone Redact helps departments like RCSO to meet public expectations without compromising operational efficiency or privacy rights.” - Ryan Steelberg

3) UCR reinvents webinars with AI-driven participant synthesis - UCR's webinar design follows a clear industry rhythm: instead of one-way slide decks, modern sessions stitch together automated evidence extraction, smart adjudication and synthesized participant summaries so moderators can surface the strongest claims in real time; this mirrors tools like PICO Portal Ask AI tool for rapid data extraction (which promises to cut the typical 40–60 minute per-article slog), Cochrane AI methods in evidence synthesis webinar series, and sector-focused webinar formats such as the ARM Institute Summer AI webinar series 2025 that emphasize workforce-ready, demo-driven learning.

The payoff is tangible: rather than scrolling through a hundred chat messages, hosts get an evidence-backed one-minute brief that flags who raised which claim and why it matters - a small change that can turn a chaotic Q&A into actionable follow-ups and faster, verifiable decisions.

4) UCR rolls out Google NotebookLM for teaching and learning (KTLA spotlight) - Campus instructors now have access to NotebookLM's suite of teaching tools that turn uploaded course materials into study-ready outputs: AI-generated Audio Overviews (even two ‘podcaster' voices for a conversational breakdown), new Video Overviews that produce narrated slides, interactive Mind Maps and shareable featured notebooks packed with expert-sourced material; administrators can consult Google's rollout notes for admin controls and availability across Workspace editions Google Workspace update on NotebookLM availability and admin controls, while product posts detail the Video Overviews and Studio upgrades that let instructors build multiple tailored outputs from one notebook Google Labs article on NotebookLM Video Overviews and Studio upgrades.

Expect classroom wins on the practical side - faster briefings, citation-grounded answers, and soon-to-appear AI flashcards and tighter Drive search integration to pull lecture slides into a single study hub - so a student can walk out of a lecture with a curated audio summary and a mind map instead of a stack of unorganized PDFs CNET review of NotebookLM and practical classroom implications.

“This is the first AI-focused partnership The Economist has signed and we are excited for NotebookLM users to explore a selection of articles from The Economist's annual special issue, The World Ahead 2025.” - Luke Bradley-Jones, The Economist

5) NSF-funded study: wearable sensors + AI track infant motor development - UC Riverside's Perception, Action, & Development Lab won a three-year, nearly $424,000 NSF award to mail wearable sensors to families and use AI to turn a full week of at-home movement into measurable, predictive signals; the project will collect continuous data from seven- and nine-month-olds to study how daily routines and caregiver behavior shape opportunities for physical activity and to predict sitting and standing proficiency by 11 months, with the aim of producing a large, shareable longitudinal dataset that clinicians (especially pediatric physical therapists) can use for earlier, evidence-backed interventions - read the UCR announcement for grant details: UCR announcement: Psychologist will use AI to study infant motor development and the PADLab project overview on methods and national-scale data collection plans: PADLab project overview - wearable sensing NSF grant.

“Early motor skills are the building blocks for later development in areas like communication and spatial awareness... observe infants in daily life... cutting-edge tools that are practical to use in the home.” - John Franchak

6) UCR advances privacy with source-free certified unlearning (ICML) - Machine unlearning tackles the “right to be forgotten,” and a new arXiv paper, “Certified Unlearning for Neural Networks,” proposes a source-free, provable approach that uses noisy fine-tuning on the retained data - leveraging privacy amplification by stochastic post-processing - to formally remove specific training examples without assuming any particular loss function; the authors analyze the efficiency vs.

accuracy trade-offs and show empirical gains over prior baselines, turning what used to be a blunt retrain-or-forget choice into something more like a surgical eraser that preserves model utility while excising sensitive influence.

Read the technical summary on the arXiv abstract page Certified Unlearning for Neural Networks - arXiv abstract and technical summary and explore the authors' implementation on GitHub in the Certified Unlearning for Neural Networks GitHub repository for reproductions and practical details.

7) Environmental costs of AI: water and energy concerns hit home - As generative models multiply, the unseen toll is becoming stark: data centers already account for a growing slice of national power (roughly 4.4% of U.S. electricity today) and analysts warn AI-specific demand could balloon in the years ahead, reshaping grid planning and carbon exposure - see the MIT Technology Review analysis of AI's energy footprint MIT Technology Review analysis of AI's energy footprint.

Training and nonstop inference also drive heavy water use for cooling - some facilities use millions of gallons a day - and a regional study projects roughly 7 billion gallons of water demand for Western data centers by 2035, a strain flagged in the IEEE at Penn State coverage on AI's energy and water pressures IEEE at Penn State: why AI uses so much energy and what we can do and KUNC reporting on AI data centers stressing Western water and grid capacity KUNC: AI data centers could stress Western water and grid capacity.

For a clear primer on why model size, inference volume, cooling choices and supply chains all matter, read the MIT News explainer on generative AI's environmental impact MIT News explainer on generative AI's environmental impact - the vivid takeaway: a single “tiny” query stacks into real grid strain and water demand when multiplied by millions of users.

“a generative AI training cluster might consume seven or eight times more energy than a typical computing workload.”

8) Riverside.fm review: AI-first tools for creators and remote production - Riverside has shifted from a recording-only studio into an all-in-one, AI-powered production tool that can radically shorten post‑production while preserving broadcast-quality captures: local 4K recording with separate stems, a text-based editor that trims by editing the transcript, and AI helpers like Magic Clips, Magic Audio and AI Show Notes that turn long interviews into ready-to-publish clips and captions in minutes; the practical upside is obvious - clean, studio-grade audio and video (you can literally count the hair on a guest's beard in some comparisons) without a complex rig - explore feature details on Riverside's AI-powered product page Riverside AI-powered product page and features or join the company's full walkthrough to see AI B-roll, per-scene captions and Video Dub in action on the Riverside feature walkthrough page Riverside full feature walkthrough with AI demos.

For creators who want fast, reliable remote capture plus smart pre-editing, Riverside is a practical, time-saving choice.

Nava Labs pilot: AI referral tool for social and health services (Goodwill, First 5 Riverside) - Nava Labs is launching a pilot with Goodwill Central Texas and Goodwill Keystone Area to test an AI-powered resource referral tool that's integrated into case management workflows, aiming to help staff spot and connect clients with health and social services beyond what a single nonprofit can offer; read Nava Labs announcement on the AI pilot with Goodwill organizations Nava Labs announcement: AI referral pilot with Goodwill Central Texas and Goodwill Keystone Area.

Funded by a Gates Foundation grant and described in industry coverage as the second in a series of trials, the project builds on earlier work (including an LA chatbot) and ties into Nava's broader Generative AI work with California partners like First 5 Riverside under a Google.org cohort - coverage and context are available in ExecutiveBiz coverage of the Nava Labs referral tool pilot ExecutiveBiz: Nava Labs artificial intelligence referral tool pilot and Nava's Google.org grant announcement about the Generative AI Accelerator Nava Labs Google.org Generative AI Accelerator grant announcement.

The practical payoff is simple but meaningful: by cutting back on paperwork and lookup time, the tool aims to let caseworkers spend more of their shift helping clients navigate services rather than wrestling with forms.

“This AI-powered referral tool has the opportunity to reduce administrative burden for caseworkers, enabling more and better tailored referrals for clients.” - Genevieve Gaudet, Director of Nava Labs

10) AI in grassroots investigations: how consumer tools helped recover a stolen Lamborghini - When private cameras, phone clips and simple consumer tools are combined with modern AI, neighborhood sleuths and small reporter teams can turn scattered footage into a lead fast: private camera networks can push real‑time alerts with location and thumbnails to apps, a setup The Marshall Project detailed in New Orleans where residents' cameras fed notifications to officers via Project NOLA Marshall Project report on private AI camera alerts in New Orleans.

Layering in AI-powered evidence management - automated ingestion, frame‑by‑frame analysis, license‑plate OCR and smart tagging - speeds review of hours of video into searchable clips and timestamps that point investigators to a precise street and minute, the kind of tooling VIDIZMO and public‑safety analyses say is reshaping digital forensics VIDIZMO blog on AI public safety technology and digital forensics.

The payoff can be dramatic: a single app ping with a timestamped thumbnail and a match on a plate or gait can convert a cold lead into a tow‑yard recovery - but the same work highlights the tradeoffs between faster recoveries and the privacy and oversight questions that follow.

“When you make this a private entity, all those guardrails that are supposed to be in place for law enforcement and prosecution are no longer there, and we don't have the tools to do what we do, which is hold people accountable.” - Danny Engelberg

Conclusion: Balancing innovation, ethics and infrastructure in Riverside's AI future - Riverside's momentum now reads like an ecosystem: RAISE Institute at UC Riverside (RAISE Institute at UC Riverside), SoCal OASIS has already secured more than $65M to seed a research-and-innovation park, and a subscription agreement with Google Public Sector promises a 2–3x boost in compute and storage that makes large-scale, privacy-conscious experiments feasible (Google Public Sector and UC Riverside cloud subscription for research access).

Together with targeted NSF investments - like recent awards to improve AI-powered imaging - these moves show how hardware, funding and governance can converge; the test now is workforce readiness and practical guardrails, which is where short, applied programs such as Nucamp AI Essentials for Work 15‑Week Bootcamp (registration) fit in by teaching prompt craft, tool use, and everyday ethics so city halls, newsrooms and nonprofits can use AI responsibly.

The key takeaway: scale the infrastructure, fund the safety research, and train the people who will put both to accountable use.

“Through this new service structure we aim to empower faculty and students to focus on their research by removing administrative barriers and providing quick access to infrastructure and service.” - Matthew Gunkel