In this TLC episode, Tom Kurfess, “Cyber Kyle” Saleeby, both of Georgia Tech, and I (Stephen) talk about OpenAI’s ChatGPT – or, the hottest bot in town. The large language model (LLM) is both highly conversational and controversial. Let’s “chat” GPT, and break down what an LLM is, how to use it, and its utility to manufacturing.
First, to clear the air: no, ChatGPT won’t (and can’t) replace manufacturing jobs. But why?
Large language models leverage machine learning and words. They provide the most statistically relevant string that follows a given prompt. To do so, tools like ChatGPT use a massive dataset of text taken from the internet up to 2021, when it was released. It knows nothing of current events, and cannot predict the future.
In a sense, it’s more like “artificial interpolation.” Its responses are only as good as our human writing abilities.
These tools are excellent for summarizing long texts or generating code snippets but could be better for more complicated tasks. LLMs may not “listen” to a prompt, or flat-out lie, as they focus more on syntax and less on semantics. To be effective, users must be conversational, as the more input is given, the better the results will be.
Exercise caution when using LLMs. Despite safeguards, they can provide dangerous information. For example, ChatGPT refuses to answer illegal or hazardous questions, but clever rephrasing can work around its guardrails. ChatGPT, Google’s Gemini, and Microsoft’s Copilot often pick up on these tricks, but not always.
Even with safer models, users should avoid offering sensitive information. Models use new prompts to train further, which means all information is unprotected. As usual, use human intuition and skepticism when chatting with robots.
Do not ask ChatGPT for G-code – we’ve tried, and it's bad. Domain-specific jobs like CNC programmers are safe from robo-replacement. However, companies are quickly adding this technology into manufacturing tools. In the future, an AI may be able to say, “Try these feeds and speeds,” or, “Use this tool. And, by the way, you’re running low on material. Reorder some more?” It’s a promising future, where AI is not a replacement but a force multiplier.
Even if ChatGPT is a little dumb, it is the first entry into commercial AI. AMT sees artificial intelligence as a central part of manufacturing’s future, so staying relevant is vital. Be sure to visit IMTS 2024 on Sept. 9-14 in Chicago to see next-generation tools and exciting AI-powered equipment in development. Visit IMTS.com/Register to learn more and register.
To read the rest of the Automation & Robotics Issue of MT Magazine, click here.
