"Half a year feels like an eternity ago, an incomprehensible Japanese person."
The beta version of Tesla’s driver-assistance feature, Full Self-Driving (FSD), was released to the first wave of owners in October 2020. Over four years have passed since then. At the time of its release, its performance was limited to the level of ADAS (Advanced Driver Assistance Systems), failing to live up to its ambitious name.
Below is an overview of the progress in Tesla’s FSD performance over the past two years, measured by the distance driven without human intervention:
FSD Performance Evolution
Version:
① Distance without intervention
② Distance without intervention in urban areas (mile = 1.60934 km)
Version ① ②
A 2023.01 11.3x: 35 miles (56.3 km), 6.5 miles (10.4 km)
B 2023.05 11.4x: 57 miles (91.7 km), 9.2 miles (14.8 km)
C 2024.01 12.3x: 79 miles (127.1 km), 18.7 miles (30.1 km)
D 2024.07 12.5x: 85 miles (136.7 km), 17.2 miles (27.6 km)
E 2024.12 13.2x: 724 miles (1165.1 km), 179.8 miles (289.3 km) → Exceeding the emergence threshold
The most significant leap in performance occurred between versions 12.5x and 13.2x. In general driving patterns, the performance improved by 852%, and for urban areas, it marked a 961% increase compared to January 2024. In just about six months, FSD surpassed a critical threshold, showcasing the rapid advancements in AI training. Moving forward, this exponential pace of development in AI-FSD is likely to accelerate even further.
The latest Tesla vehicles (from September 2023) are equipped with HW4 (Hardware 4), an autonomous driving computer with five times the performance of HW3. The next-generation AI 5 hardware, boasting approximately ten times the inference capability of HW4, is expected to be introduced into production lines for customer vehicles between August and December 2025.
By the end of 2025, Tesla’s FSD version is projected to reach 14.2x or higher, coupled with HW5 (Hardware 5), and is likely to achieve an unprecedented level of autonomous driving (Level 4 or low-speed Level 5 autonomy).
Furthermore, the progression from AI to AGI to ASI will inevitably lead to transformative changes across all products involving human interaction, such as humanoid robots.
Japanese executives of old-economy conglomerates are convinced that "EVs will at most capture a 30% market share, with the remainder being hybrid vehicles and others," and are basking in confirmation bias as if this prediction is about to come true. However, this is a dangerously misleading perspective.
In Japan, however, much of the old-fashioned economy-focused mass media continues to ignore these factual data points, frequently labeling Tesla’s FSD as immature. Such misleading commentary risks pampering Japan’s automotive industry, potentially leading to irreparable consequences.
The ability to correctly and quickly acquire factual data is crucial in this AI era. Outdated data, even from two or three years ago, has little to no value today. It is troubling to see Japan’s government and agencies seemingly lagging behind, akin to using calculators for statistical work.
We’ve shifted from an era where three years marked significant change to one where even six months can feel like an eternity. The data above provides evidence of this rapid evolution.