News
Level 2 systems like Tesla Autopilot can improve drivers’ attentiveness: IIHS study
The Insurance Institute for Highway Safety (IIHS) recently published the results of a rather unique real-world driver attentiveness test. Using a 2019 Mercedes-Benz C300 equipped with a Level 2 Advanced Driver Assistance System (ADAS) and a giant pink teddy bear dressed in a yellow high-visibility jacket, the agency was able to roughly determine if systems like GM’s Super Cruise and Tesla’s Autopilot make drivers less attentive on the road.
The participants of the study included both drivers who were familiar with Level 2 driver-assist systems and drivers who had little to no experience with ADAS. The IIHS tasked those familiar with Level 2 systems to drive with the Mercedes-Benz C300’s ADAS activated. A group of drivers unfamiliar with ADAS was also tasked to drive with the system engaged. Lastly, a group of drivers who were unfamiliar with the vehicle’s Level 2 system was tasked to drive without the advanced driver-assist feature activated.
To test the drivers’ situational awareness, the IIHS had an SUV with a giant pink teddy bear strapped to its back pass the Mercedes-Benz C300 three times as the participants drove over a stretch of Interstate 70 in Maryland for about an hour. Each time, the SUV with the massive stuffed bear stayed in front of the drivers for about 30 seconds. Researchers then measured the participants’ reactions after their driving session, while asking if they saw anything odd during their hour-long drive.
Interestingly enough, nearly all of the drivers who were experienced with Level 2 systems noticed the giant pink bear. The same group also identified the number of times the bear overtook the C300 during the hour-long test. Drivers who were inexperienced with Level 2 systems didn’t perform as well, with a good number of inexperienced drivers who used the C300’s ADAS failing to remember the giant pink teddy bear at all.
“Our data suggest that Level 2 driving automation has the potential to improve a driver’s situational awareness (SA) once he or she is familiar with the technology, although it does not guarantee it. Unfamiliar drivers, however, appear to have even more difficulty maintaining SA when using the system than when driving without it. On average, participants who were familiar with Level 2 systems showed the highest degree of SA about the bear when using the system, unfamiliar participants who drove with the system off had moderate SA, and unfamiliar participants who drove with the system on demonstrated the lowest SA,” the IIHS wrote.
Videos from inside the C300 showed that the drivers who correctly identified the giant stuffed bear actually spent more time scanning the road ahead of them. These drivers, particularly those familiar with Level 2 systems, even tended to look out of the car’s side windows. On the other hand, those who missed the bear spent a considerable amount of time just focusing on the road straight ahead. Drivers who failed to spot the bear even once spent considerable time looking at various aspects of the C300’s dash.
Considering the results of the IIHS’ study, it appears that some experience with Level 2 systems would be best for drivers before they are allowed access to more advanced driver-assist systems such as Tesla’s Full Self-Driving Beta, which is poised to be rolled out to a greater number of electric car owners in a couple of weeks. Situational awareness, after all, is critical when driving, and having drivers nervously fiddling around their vehicles’ features while operating a Level 2 system may present some risks. That being said, the IIHS’ results do go in line with one of Elon Musk’s more notable points–systems like Autopilot could actually perform as a formidable safety feature, provided that they are used responsibly and properly, of course.
The IIHS’ situational awareness study could be accessed below.
IIHS Level 2 Autonomy Report by Simon Alvarez on Scribd
Don’t hesitate to contact us for news tips. Just send a message to tips@teslarati.com to give us a heads up.
Elon Musk
Elon Musk’s xAI bets $20B on Mississippi with 2GW AI data center project
The project is expected to create hundreds of permanent jobs, dramatically expand xAI’s computing capacity, and further cement the Mid-South as a growing hub for AI infrastructure.
Elon Musk’s xAI plans to pour more than $20 billion into a massive new data center campus in Southaven, Mississippi, marking the largest single economic development project in the state’s history.
The project is expected to create hundreds of permanent jobs, dramatically expand xAI’s computing capacity, and further cement the Mid-South as a growing hub for AI infrastructure.
xAI goes MACROHARDRR in Mississippi
xAI has acquired and is retrofitting an existing facility in Southaven to serve as a new data center, which will be known as “MACROHARDRR.” The site sits near a recently acquired power plant and close to one of xAI’s existing data centers in Tennessee, creating a regional cluster designed to support large-scale AI training and inference.
Once completed, the Southaven facility is expected to push the company’s total computing capacity to nearly 2 GW, placing it among the most powerful AI compute installations globally. The data center is scheduled to begin operations in February 2026.
Gov. Tate Reeves shared his optimism about the project in a press release. “This record-shattering $20 billion investment is an amazing start to what is sure to be another incredible year for economic development in Mississippi. Today, Elon Musk is bringing xAI to DeSoto County, a project that will transform the region and bring amazing opportunities to its residents for generations. This is the largest economic development project in Mississippi’s history,” he said.
xAI’s broader AI ambitions
To secure the investment, the Mississippi Development Authority approved xAI for its Data Center Incentive program, which provides sales and use tax exemptions on eligible computing hardware and software. The City of Southaven and DeSoto County are also supporting the project through fee-in-lieu agreements aimed at accelerating development timelines and reducing upfront costs.
Founded in 2023 by Elon Musk, xAI develops advanced artificial intelligence systems focused on large-scale reasoning and generative applications. Its flagship product, Grok, is integrated with the social media platform X, alongside a growing suite of APIs for image generation, voice, and autonomous agents, including offerings tailored for government use.
Elon Musk highlighted xAi’s growth and momentum in a comment about the matter. “xAI is scaling at an immeasurable pace — we are building our third massive data center in the greater Memphis area. MACROHARDRR pushes our Colossus training compute to ~2GW – by far the most powerful AI system on Earth. This is insane execution speed by xAI and the state of Mississippi. We are grateful to Governor Reeves for his support of building xAI at warp speed,” Musk said.
Tesla’s Head of AI, Ashok Elluswamy, says that something that was expected with version 14.3 of the company’s Full Self-Driving platform has already partially shipped with the current build of version 14.2.
Tesla and CEO Elon Musk have teased on several occasions that reasoning will be a big piece of future Full Self-Driving builds, helping bring forth the “sentient” narrative that the company has pushed for these more advanced FSD versions.
Back in October on the Q3 Earnings Call, Musk said:
“With reasoning, it’s literally going to think about which parking spot to pick. It’ll drop you off at the entrance of the store, then go find a parking spot. It’s going to spot empty spots much better than a human. It’s going to use reasoning to solve things.”
Musk said in the same month:
“By v14.3, your car will feel like it is sentient.”
Amazingly, Tesla Full Self-Driving v14.2.2.2, which is the most recent iteration released, is very close to this sentient feeling. However, there are more things that need to be improved, and logic appears to be in the future plans to help with decision-making in general, alongside other refinements and features.
On Thursday evening, Elluswamy revealed that some of the reasoning features have already been rolled out, confirming that it has been added to navigation route changes during construction, as well as with parking options.
He added that “more and more reasoning will ship in Q1.”
🚨 Tesla’s Ashok Elluswamy reveals Nav decisions when encountering construction and parking options contain “some elements of reasoning”
More uses of reasoning will be shipped later this quarter, a big tidbit of info as we wait v14.3 https://t.co/jty8llgsKM
— TESLARATI (@Teslarati) January 9, 2026
Interestingly, parking improvements were hinted at being added in the initial rollout of v14.2 several months ago. These had not rolled out to vehicles quite yet, as they were listed under the future improvements portion of the release notes, but it appears things have already started to make their way to cars in a limited fashion.
Tesla Full Self-Driving v14.2 – Full Review, the Good and the Bad
As reasoning is more involved in more of the Full Self-Driving suite, it is likely we will see cars make better decisions in terms of routing and navigation, which is a big complaint of many owners (including me).
Additionally, the operation as a whole should be smoother and more comfortable to owners, which is hard to believe considering how good it is already. Nevertheless, there are absolutely improvements that need to be made before Tesla can introduce completely unsupervised FSD.
Elon Musk
Tesla’s Elon Musk: 10 billion miles needed for safe Unsupervised FSD
As per the CEO, roughly 10 billion miles of training data are required due to reality’s “super long tail of complexity.”
Tesla CEO Elon Musk has provided an updated estimate for the training data needed to achieve truly safe unsupervised Full Self-Driving (FSD).
As per the CEO, roughly 10 billion miles of training data are required due to reality’s “super long tail of complexity.”
10 billion miles of training data
Musk comment came as a reply to Apple and Rivian alum Paul Beisel, who posted an analysis on X about the gap between tech demonstrations and real-world products. In his post, Beisel highlighted Tesla’s data-driven lead in autonomy, and he also argued that it would not be easy for rivals to become a legitimate competitor to FSD quickly.
“The notion that someone can ‘catch up’ to this problem primarily through simulation and limited on-road exposure strikes me as deeply naive. This is not a demo problem. It is a scale, data, and iteration problem— and Tesla is already far, far down that road while others are just getting started,” Beisel wrote.
Musk responded to Beisel’s post, stating that “Roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving. Reality has a super long tail of complexity.” This is quite interesting considering that in his Master Plan Part Deux, Elon Musk estimated that worldwide regulatory approval for autonomous driving would require around 6 billion miles.
FSD’s total training miles
As 2025 came to a close, Tesla community members observed that FSD was already nearing 7 billion miles driven, with over 2.5 billion miles being from inner city roads. The 7-billion-mile mark was passed just a few days later. This suggests that Tesla is likely the company today with the most training data for its autonomous driving program.
The difficulties of achieving autonomy were referenced by Elon Musk recently, when he commented on Nvidia’s Alpamayo program. As per Musk, “they will find that it’s easy to get to 99% and then super hard to solve the long tail of the distribution.” These sentiments were echoed by Tesla VP for AI software Ashok Elluswamy, who also noted on X that “the long tail is sooo long, that most people can’t grasp it.”
Elon Musk’s xAI bets $20B on Mississippi with 2GW AI data center project
Tesla AI Head says future FSD feature has already partially shipped
