Tesla has begun rolling out “reasoning” features with the latest Full Self-Driving (FSD) V14.2 series.
The V14.2 series was released in late November, about a month after the launch of FSD 14.
From October onward, Tesla has been releasing weekly updates to the software, with bug fixes and new features. The latest update — v14.2.2.2 — started rolling out last week.
In an X reply to Bloomberg‘s Ed Ludlow, who questioned if “FSD v14 with “reasoning” isn’t out yet – right?,” the company’s VP of Software AI Ashok Elluswamy said that some features have already “shipped in 14.2.”
These include “navigation route changes during construction” and “parking options.”
The lead engineer added that “more and more reasoning will ship in Q1.”
Reasoning in V14
Even before Version 14 was launched, CEO Elon Musk had already affirmed that the most recent software would eventually be “sentient.”
He first stated that Tesla vehicles equipped with FSD would feel “almost sentient” by V14.2, in late September, an idea that he has reiterated several times since then.
In October, the chief executive teased on X that “by v14.3, your car will feel like it is sentient.”
He later wrote that “Tesla AI will feel increasingly sentient,” reacting to a video shared by shareholder Sawyer Merritt, experiencing FSD V14.1 in his Model Y.
Weeks later, during the third-quarter earnings call, Musk said he was “confident that we can solve unsupervised full self-driving at a safety level much greater than human,” after highlighting an uptick in human behavior-related accidents.
That was when the CEO stated that Tesla would “be adding reasoning to the car.”
By then, he anticipated reasoning to be included “in 14.3, maybe 14.4, something like that,” to which Elluswamy then added, “or by the end of this year, for sure.”
“With reasoning, it [the car]’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,” Musk exemplified, adding that “it’s going to spot empty spots much better than a human.”
Training Data
Despite the ambitious target of achieving fully autonomous driving, Elon Musk admitted on Thursday that “roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving,” saying that “reality has a super long tail of complexity.”
As of the end of the third quarter, and according to data shared in Tesla‘s Shareholder Deck, Tesla’s fleet had driven just below 6 billion miles with FSD (Supervised), with nearly 5 billion of those miles having been on V12 and beyond.
However, the latest FSD 14 was only released in early October, meaning that data from the 14.1 and 14.2 series is not yet contemplated in those figures.
Tesla will report the updated numbers on January 28.
AI Team Reinforced
In the same call, Musk stated that Tesla‘s “world simulator for reinforcement learning is pretty incredible.”
“Our Tesla reality simulator, when you see it, the video that’s generated by the Tesla reality simulator and the actual video looks exactly the same,” he said, noting that it “allows us to have a very powerful reinforcement learning loop to further improve the Tesla AI.”
In November, Not a Tesla App noticed that the company was recruiting an “AI Engineer, Reinforcement Learning and Distillation,” a role explicitly tasked with developing smarter, more contained models.
In the job description, Tesla stated that these models “continue to struggle in real-world physical reasoning, often struggling to tell left from right.”
The Tesla AI team aims to develop “Olympiad-level physical intelligence that will enable highly capable robots, both wheeled and legged,” the company stated, hinting at both the Optimus humanoid and the FSD software in cars.
“These models should be able to anticipate and reason about future movements of any object or scene at the level of a race car driver or professional athlete,” it further said.
