Urgent.News

What's breaking now, across thousands of outlets.

AI

Scientists Download Frontier AI Model Into Self-Driving Car Let It Loose

They didn't do great on a test course. The post Scientists Download Frontier AI Model Into Self-Driving Car Let It Loose appeared first on Futurism .

Scientists Download Frontier AI Model Into Self-Driving Car Let It Loose

Scientists loaded frontier AI models into the brains of a Toyota Corolla to test their ability to drive a lap in a parking lot. The models used included OpenAI’s GPT-6 Astra, xAI’s Grok 4.6, and Anthropic’s Claude Fable 5.1. The project, called DrivingBench, connected an internet-connected laptop to the car's control systems via a "comma four" device.

The LLMs sent commands based on GPS telemetry and car status indicators. Only GPT-6 Astra successfully completed the course after its second attempt, covering less than 500 feet in five minutes. Other models struggled, with most failing to navigate past the first corner. The researchers concluded that frontier models can drive real vehicles at low speeds, but safety and alignment concerns still need significant attention.

The token cost for the test was $7.74, equivalent to about 500 times the cost of fuel for a car with 25 miles per gallon fuel efficiency at $4.60 per gallon. Some LLMs warned the researchers about driving the car safely, even in an empty lot. The experiment showed that off-the-shelf frontier AI models are not yet ready for real-world driving.

Written by urgent.news from Futurism's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at futurism.com →

More in AI

AI Is Not the Solution to Every Problem: When Rule-Based Systems Make More Sense

Extensive Use of AI for Simple Task Costing Token and Even Then Not Producing Favorable Results Today AI has been the buzzword every industry that wants to solve their task with the utilisation of AI…

  • AI not always best solution, rule-based systems more efficient for certain tasks
  • Rule-based approach using Python and graph theory cheaper than AI for BoM creation
  • Organizations should evaluate context before adopting AI solutions

More from Sunday 27 September →