AI prevents 200,000+ accidents
June 16, 2025
Good morning. Google is reportedly ditching Scale AI after Meta dropped $14.3 billion for nearly half the company, with Microsoft and OpenAI also supposedly pulling away from the data labeling startup. Apparently nothing kills a business relationship quite like your vendor getting a little too cozy with your biggest rival.
— The Deep View Crew
In today’s newsletter:
🧠 AI for Good: The potential of brain-computer interfaces in medicine
🏗️ Taiwan tightens export controls on Huawei and SMIC
🚗 Samsara’s AI driver coaching software claims major safety wins, but adoption challenges remain
🧠 AI for Good: The potential of brain-computer interfaces in medicine
Source: Midjourney v7
What if paralyzed stroke survivors could control robotic arms with their thoughts, or autistic children could engage in therapy through mind-controlled games? Researchers worldwide are making these possibilities reality through AI-powered brain-computer interfaces.
What's happening: Scientists are developing systems that read electrical brain activity through scalp electrodes and use AI to translate those signals into commands for external devices. Recent studies show these non-invasive approaches can help stroke patients regain motor function and assist autistic children in social engagement activities.
At Holland Bloorview Kids Rehabilitation Hospital in Toronto, researchers successfully used brain-computer interfaces as recreational therapy for autistic children, allowing them to control remote-controlled cars through mental focus. The program helped improve attention and engagement while providing therapeutic benefits without the stress of traditional interventions.
How it works:
Electrodes on the scalp collect electrical brain activity
AI interprets the brain signals linked to movement or intention
The system provides real-time feedback based on mental focus
This creates a closed loop that helps the brain practice tasks
Progress continues even if the body cannot move yet
Meanwhile, a comprehensive review published in March 2025 analyzing 18 studies found that brain-computer interfaces show significant promise for stroke rehabilitation. The technology works by detecting brain signals linked to intended movements, even when patients cannot physically move, and providing real-time feedback that encourages neural recovery.
Why it matters: Traditional stroke rehabilitation requires some remaining motor function, leaving severely paralyzed patients with few options. Brain-computer interfaces offer hope for the 30-50% of stroke survivors with complete chronic paralysis by creating new pathways for the brain to practice and potentially rewire itself.
University of Melbourne researchers are pioneering an endovascular approach called the Stentrode, which deploys brain interfaces through blood vessels rather than invasive skull surgery. The device remains effectively invisible to the brain, reducing rejection risk while enabling direct neural control of external devices.
For autism applications, the technology's appeal lies in its engaging, game-like interface that can maintain children's attention while supporting therapeutic goals like social communication and focus training.
A New Era Of Coding
AI is fundamentally changing how developers write code and ship software.
This shift calls for a new interface — one built for working with agents across every stage of development.
On June 24, join Warp Founder and CEO Zach Lloyd during their live-streamed launch event reveal Warp 2.0 — the new workbench for coding with AI agents.
Get a front-row seat to see how developers will use agents to code, debug, deploy, and more.
Spots are limited — reserve yours now to be among the first to see what’s next. Click HERE to register.
🏗️ Taiwan tightens export controls on Huawei and SMIC
Source: Midjourney v7
Taiwan has blacklisted China's Huawei Technologies and Semiconductor Manufacturing International Corp (SMIC), requiring Taiwanese companies to obtain government permits before selling any goods to the Chinese tech giants.
The details: The Taiwan International Trade Administration updated its strategic high-tech commodities entity list on June 10, adding 601 entities from Russia, Pakistan, Iran, Myanmar and mainland China. Huawei and SMIC now join a restricted list that includes various sanctioned organizations and companies.
The decision follows revelations that TSMC manufactured more than 2 million Ascend 910B logic dies that ended up with Huawei via shell companies to circumvent existing US restrictions. A TechInsights teardown in late 2024 discovered TSMC-manufactured chips in Huawei's advanced AI processors, prompting TSMC to halt shipments and notify US authorities.
Why it matters: The timing is significant, coming weeks after the US warned that the use of Huawei Ascend AI chips "anywhere in the world" violates the government's export controls. Taiwan's action cuts off access to Taiwan's plant construction technologies, materials, and equipment, potentially setting back China's efforts to develop new AI semiconductors.
Indudstry analysts suggest the practical impact may be limited. Taiwan's move to blacklist Huawei and SMIC drew little reaction from local tech firms, as most Taiwanese suppliers had already pulled back from working with the companies following earlier US restrictions.
Ray Wang, an independent semiconductor and tech analyst, told CNBC the addition is likely aimed at "reinforcement of this policy and a tightening of existing loopholes" and could raise punishments for any potential future breaches.
The semiconductor restrictions reflect broader efforts to maintain Western technological advantages as China's most advanced AI chip designer and logic chip manufacturer, Huawei and SMIC, respectively, will most likely remain stuck at 7 nanometers (nm) or perhaps a flawed 5 nm technology node for many years without access to advanced manufacturing equipment.
ACI.dev: MCP server for VibeOps — Automate DevOps while Vibe Coding
ACI.dev’s Unified MCP Server can turn any agentic IDE you use into Lovable/Bolt/V0.
We let vibe coders go from idea → code → deploy → distribute, without needing to touch devOps at all.
It’s as simple as adding a single MCP server to your Cursor.
Our entire platform is open-source, so you can see how it works underneath.
See our blog where we run a side by side demo against Lovable.
Salt Lake is testing AI to pick up 911 calls
Ramp hits $16 billion valuation in new funding round
Toledo’s athletic department is going all in on AI
People are forming real emotional bonds with their AI assistants
Chipotle says new AI hiring tool slashed recruiting time by 75%
States with the worst road rage
Scale AI founder says he's waiting for Elon Musk's Neuralink before he has kids
🚗 Samsara’s AI driver coaching software claims major safety wins, but adoption challenges remain
Source: Midjourney v7
Fleet management companies are racing to prove that AI can prevent crashes and cut costs, with Samsara leading the charge by claiming its AI coaching technology has prevented over 200,000 accidents while delivering an 815% return on investment.
What's happening: The commercial fleet industry is embracing AI-powered dash cameras and real-time driver coaching as insurance costs soar and regulatory pressure mounts. Companies like Samsara, Geotab, Lytx and Motive are deploying computer vision systems that monitor driver behavior and provide immediate feedback through in-cab audio alerts.
The technology works by analyzing video feeds from dashboard cameras to detect risky behaviors like phone use, seatbelt violations and harsh braking. When the AI identifies a safety issue, it immediately coaches drivers through speakers rather than waiting for a post-incident review.
How it works:
AI models trained on millions of images analyze real-time video from dash cameras
Systems detect behaviors like distracted driving, following too closely and drowsiness
Immediate audio coaching provides "nudges" to correct risky behavior
Edge computing processes data locally to avoid connectivity delays
Fleet managers receive safety scores and trend analysis
An IDC study of 130 Samsara customers found organizations saved an average of $2.02 million annually, with DHL reporting a 26% reduction in accidents across 20 sites. Eurovia USA saw mobile phone usage events drop 80% within 30 days of deployment.
But the industry faces significant headwinds. AI accuracy varies widely between providers, and false positives remain a persistent problem that can frustrate drivers and undermine safety programs if not properly managed.
Cost remains a major barrier. Customer reviews frequently cite pricing complaints, with some systems requiring multi-year commitments and additional fees for advanced features. Smaller fleets often find the technology cost-prohibitive despite potential safety benefits.
Industry challenges include:
Privacy concerns over continuous driver monitoring
Technical limitations like data overload (vehicles generate up to 25GB per hour)
Cultural resistance requiring extensive change management
Integration complexity with existing fleet systems
The competitive landscape shows different strategic approaches. Geotab leads market share with 4.6+ million connected vehicles through an extensive partner ecosystem, while Samsara focuses on integrated AI capabilities. Traditional players like Verizon Connect compete on reliability and coverage.
Recent developments suggest the technology is maturing rapidly. Samsara's December launch of Samsara Intelligence includes generative AI tools for operational insights, while partnerships like Continental's tire monitoring integration expand beyond basic safety monitoring.
Driver acceptance varies significantly. "There were a few bad habits I didn't know I had, and Samsara was able to point them out to me so I could fix them," said Vanessa Veloz, a driver at Chalk Mountain Services. However, privacy advocates and some drivers view the technology as intrusive surveillance rather than helpful coaching.
The global fleet management market is projected to grow from $28.6 billion in 2024 to $55.6 billion by 2028, driven partly by AI adoption. With 94% of vehicle crashes attributed to human error, the technology addresses a critical safety challenge while promising measurable ROI.
As regulatory requirements tighten and insurance costs continue rising, fleet operators increasingly view AI safety technology as essential rather than optional. However, successful adoption requires addressing cost barriers, privacy concerns and technical limitations that currently prevent widespread deployment across all fleet sizes.
The AI driver coaching industry shows genuine promise in preventing accidents and reducing costs, but the technology remains uneven in quality and accessibility. While Samsara's impressive numbers demonstrate real potential, significant disparities in AI accuracy and high implementation costs suggest the industry is still in early stages. Success will depend on companies developing more accurate, affordable solutions that balance safety benefits with legitimate privacy concerns. The most impactful developments may come from making these technologies accessible to smaller fleets where the greatest number of commercial drivers operate.
Which image is real? |
🤔 Your thought process:
Selected Image 1 (Left):
“The other one has a strange perspective, and the little clouds in the sky made it a little too unreal.”
“Primarily this was a gut check. I grew up by the ocean and have seen a lot of beaches. Minor details though, the fake one looked just a little too perfect. Too blue, too nice a beach between the dunes and water. Whereas the real photo was imperfect in ways that felt more accurate. Steep dunes almost meeting the waterline. Grey-blue water, a bit of haze to the air.”
Selected Image 2 (Right):
“Now you're adding a gradient to the sky in the fake photo. Patently unfair.”
💭 Poll Results
Here’s your view on “Who wins the AI infrastructure battle?”…
AMD (15%), Nvidia (50%), Big Tech (13%) , Someone Else (22%)
Thanks for reading today’s edition of The Deep View!
We’ll see you in the next one.
P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!
If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.