FDE engineers create dozens of nose designs
Colin Jarvis, who leads OpenAI's "FDE (forward deployed engineering)" team, which sends engineers to client companies to support AI adoption, described vehicle design research with Mazda on X on Nov. 4 (the evening of the same day Japan time). He said FDE engineers worked with Mazda's aerodynamics engineers and used the top-tier model "GPT-6 Astra" in the AI coding tool "Codex" to create dozens of variations on the shape of a car's nose and simulate airflow. Jarvis expressed hope that AI could be used for similar optimization in manufacturing and supply chains, and invited companies with the same challenges to get in touch.
One of the coolest things about our FDE team is that we get to work side by side with Research on frontier problems. Today we're announcing a new partnership with Mazda that shows what this looks like in automotive engineering. Working with Mazda's aerodynamics engineers, our FDEs used Astra in Codex to build and simulate dozens of variations on the nose of a car. We're excited about the potential applications for LLMs driving these kind of optimizations in manufacuturing, supply chain and more - if you've got these kind of problems, we'd love to hear from you! Read on here: https://openai.com/index/accelerating-engineering-for-the-physical-world/
On Nov. 3 (Nov. 4 Japan time), OpenAI published its efforts to use ChatGPT and Codex for mechanical design, electrical design and simulation, and disclosed the research with Mazda as part of that. The use by its internal hardware team and an attempt to have Astra design devices were covered in a previous article. Jarvis' post focused on the research with Mazda and explained the role of his own team.
Official research scope is aerodynamics and crash safety
According to OpenAI's announcement, the collaboration with Mazda is vehicle design research. OpenAI's research team and FDE team are working with Mazda's aerodynamics and crash safety engineers. They simulate airflow and the behavior of vehicle body structures to create and evaluate design proposals, also taking design and cost into account. In aerodynamics, Astra creates nose shape proposals, calculates airflow and narrows them down to ones likely to reduce drag.
The Mazda part of the announcement is limited to this design research. The announcement does not mention applications to manufacturing or supply chains. The parts of the same announcement that mention "manufacturing" concern preparing devices prototyped inside OpenAI for mass production, work on manufacturing drawings with U.S. racing team Chip Ganassi Racing, and a call for partners. Applications to manufacturing and supply chains are what Jarvis raised in his post as hopes for future uses.
Mazda has not issued its own announcement about the research. Impress Watch's article also does not include direct comment from Mazda.
Mazda engineer says it shows which features matter for aerodynamics
OpenAI's announcement includes remarks from Akihiro Nakata, a principal engineer at Mazda. Nakata said that exploring and simulating dozens of designs with Astra in Codex shows which features affect aerodynamics and where designers have room to freely explore shapes.
Real Sound reported that the work is at the research stage and does not indicate performance improvements or adoption in production vehicles. Figures such as how much drag was reduced have not yet been published. OpenAI said it will publish the progress and results of the research over the coming months.
In the same announcement, OpenAI disclosed that in preparing an internal prototype for mass production, Astra could not consistently take into account manufacturing constraints, tolerances (allowable ranges for dimensions) and assembly conditions. Between producing many design proposals and finishing them into a form that can be mass-produced, human checks are still needed.
Deployment team teams up with research
FDE is the team in which OpenAI engineers go into client companies' operations and build models into their data and work. At the "HumanX" conference in Amsterdam in September, Jarvis said about 80% of cases in which companies' AI adoption stumbles are deployment problems, not problems with model performance. In this post, he cited working on frontier problems side by side with the research team as a distinctive feature of FDE.
On May 11, OpenAI announced it would launch "OpenAI Deployment Company," a new company to support companies' AI adoption, with more than $4 billion in capital. U.S. investment firm TPG led the investment, and the company said it would acquire applied AI consulting firm Tomoro, adding about 150 engineers. As OpenAI expands its setup for sending engineers to client sites, the research with Mazda is an example of moving into automotive design.
In the announcement, OpenAI invited mechanical, electrical, manufacturing and other engineers, as well as design, simulation and manufacturing software companies, to join as partners.
