Auto Jobs Now

This article was originally published HERE

There’s nothing better than having an interactive conversation and walking away better educated on a subject, and then be able to act on it. This is ChatGPT at its most useful. You’ve heard of ChatGPT. If not, its founder, OpenAI, does a much better job of explaining the world’s most publicized app than we could here.

In fleet, conversations with computers aren’t new. In 2020, WEX launched a voice command solution for Amazon Alexa, allowing fleet managers to receive information and data verbally. But Alexa and other voice assistants aren’t truly AI. While Alexa uses natural language processing and can learn your preferences, it is still just a piece of software responding to what it is coded to do.

ChatGPT can actually compose essays, write computer programming, reply to customer service complaints, and create jokes. So how can it be used in fleet? Stephen Joos, vice president of strategy and corporate development at Fluid Truck, has been tinkering with the program to find out. “I’ve been following the hype like everybody else and playing around with the program,” he said.

Fluid Truck is an on-demand commercial vehicle rental platform for trucks and vans. One reason for Joos’s pursuit was to uncover new ways Fluid could help its customers and potential customers — from fleets and drivers to gig workers, and independent contractors — become more efficient, he said.

To start, Joos created a spreadsheet of about 60 restaurants in the Denver area, with addresses, number of packages, and cubic feet of cargo space of the delivery vehicles. He copied and pasted in the data from the spreadsheet into ChatGPT.

“I didn’t even know how it would work,” Joos said. “(The program) digested it and said, ‘Hey, it looks like you’re trying to do something with routes. Here’s what I can do.’”

Joos then prompted the program to show how many drivers would be needed based on the inputted data, assuming the use of a cargo van with 400 cubic feet of cargo space and four hours to complete the work. First, the program returned snippets of code, as if Joos were a software developer. “I was like, I don’t need code,” he said, “I need you to solve the problem.”

The program responded that Joos would need four drivers. Joos then asked for, and received, a route list for each driver in a form that could be sent to the drivers.

Using the same inputs, he switched the vehicle type from ICE to electric to see how the current routing would change. He input EV constraints around range and payload capacity. At first, the program said he’d need only three drivers.

“And I said, ‘Well, that that can’t be correct, it should be more.’ And it literally said, ‘Actually, you’re right, you need more drivers. It is eight drivers needed.”  

“When they say it’s conversational in nature, it is,” Joos said. “It is very conversational. But it did get some of the calculations wrong.”

He surmised it was forgetting to include certain constraints, though he admitted it could’ve been that his prompts were created incorrectly. There are any number of variables to input into a routing program, but Joos understood he only scratched the surface with his ChatGPT experiment.

For fleet operators, ChatGPT might be leveraged in other areas with large data sets and problems to solve. Preventive maintenance is an example, Joos said, to more precisely understand when parts need to be replaced and group that around preventive maintenance schedules.

The AI could account for duty cycles, vehicles’ average lifecycles, servicing and parts replacement intervals, for instance.

Snippets of Communications

As users converse with ChatGPT the AI refines itself. When the user keeps asking questions, the program can narrow focus on a specific question.

Just be aware that the program can “occasionally produce wrong answers,” according to its FAQ. “We’d recommend checking whether responses from the model are accurate or not,” the FAQ states.

• Joos: Can you provide the optimized route for each driver?
• ChatGPT: Yes, I can provide a visual representation. Here’s the sequence of addresses. 
• Joos: What if the weather is cold, how many drivers do I need now?
• ChatGPT: I need the delivery information.
• Joos: You already have the delivery information.
• ChatGPT: The number of drivers is three.
• Joos: You previously told me seven drivers. The range of the EVs is reduced. So, the number of drivers should increase.
• ChatGPT: Yes, that’s correct.
• Joos: Okay then, how many drivers?
• ChatGPT: Okay, actually you need eight drivers.

Caveats and a Warning

There are, of course, many ways to optimize routes, stemming from the early days of Mapquest to off-the-shelf online programs to newer, sophisticated route optimization programs that factor numerous inputs such as vehicle dimensions, delivery priority, time windows, skills matching, route modifications, and more.

Joos doesn’t think that ChatGPT will usurp those programs. But it could work for small jobs, and the code it produces could be used by a software development team to integrate into a more sophisticated program.  

At the time of the experiment, ChatGPT was trained only until 2021, so recent data was not incorporated. ChatGPT can’t manage real-time data such as weather or traffic for dynamic routing. (GPT-4 was just released, and OpenAI gives examples of how the newer version is solving problems with greater accuracy.)

Here comes the warning: ChatGPT is set up to collect as much data as it can to learn and improve. The program collects users’ IP addresses, browser settings, and data on how users interact with the site. Inputted routes become part of ChatGPT’s database. According to company’s privacy policy, user prompts are part of the public domain, therefore, other users have access to it.

Can another user connect the dots on intricate data sets and attach them to IP addresses and real humans? This isn’t yet clear, but the unknown could make interaction with the program a non-starter for many.

Joos is aware of the issue. “If you’re providing data to the AI, you’ll want to review the privacy policy and terms of use carefully to ensure the scope is limited to how you want to use it,” he said.

He referenced issues now coming to light in which artists are challenging the AI creation of content similar to their style. “It’s a gray area right now, but there will be lots of battles fought over the reference material and output.”

Uncategorized