By skyforbes 
It even wrote a provisional patent
I had an idea. I can now disclose the idea in broad strokes because I have a provisional patent, thanks largely to ChatGPT: I’m combining control theory and AI to aid pharmaceutical manufacturing by, for instance, optimizing batch yield. I have open source code working for open-loop control, and closed-loop control is coming.
Does that explain the idea well? Probably not. But that’s not the point.
The point is: Did I get ChatGPT to help me turn an idea for an invention into something amazing, or did it mostly waste my time?
I’ll let you decide.
The idea
I came up with this idea after hearing my husband describe his job. He is a data engineer supporting pharmaceutical manufacturing. Supposedly, they’re using data science to support manufacturing. That’s what the big dogs say. Actually, they make Dashboards. Their most complicated model is linear regression to figure out how manufacturing control knobs relate to output. They do really complicated anomaly detection algorithms to make sure their process doesn’t go out of spec, but that’s it. The new-fangled ideas about batch yield optimization haven’t hit them yet.
So I thought, hey husband, why don’t they combine control theory with complicated AI prediction algorithms that use all the data you collect?
He hates working more than he has to, according to him, and basically spent several conversations coming up with very smart reasons that none of this would ever work.
Then my friend, who was formerly a CEO, said my idea was so obvious that it was probably already done.
I tried doing as much market research as I could with ChatGPT and found out that, as best as I could understand, my idea had not been done yet. There were companies who were getting close, but none had actually done the idea. No search for patents turned up what I was looking for.
So I started on a quest, which is still ongoing, to actually implement my ideas in my husband’s workplace, much to his chagrin. Throughout, I used ChatGPT to advise me, almost like it was advisor to someone who wanted to start a company. Here’s how it went.
Starting a business, filing a patent — what should I do, ChatGPT?
My idea, initially, was that I would get a patent, easily convince someone at my husband’s company that they should pay for the idea, and sell the patent. At this point, I had not talked to ChatGPT.
The colleges hemmed and hawed until I revealed that I had stupidly published half the idea for a 100% discount in a journal, at which point they declined interest in helping me pursue the patent, maybe because half of the idea was prior art and couldn’t be patented. I thought, no problem. It’s better for me because then I don’t have the colleges insisting that they get half-credit for the idea, right? And I applied for faculty research funding to help pay for the costs of a patent for the other half of the idea, which run from 10K-20K.
The faculty research funding group asked for market research explaining that this patent idea would actually get market traction. I didn’t have any market research. And this is where I first used ChatGPT: It filled out the market research application for me. All I had to do was seed it with the information that I had, and it told me what my market was and how big it was.
The faculty research group said, nope. They needed third party eyes to verify the market potential. AI was not going to cut it. So I asked for money to get funds to do market research, and my request was denied because it wouldn’t help my research program. I asked ChatGPT what to do, almost like it was advising a nascent company.
ChatGPT put together a sheet that I could send to angel investors explaining that they got a cut of the patent profits if they just gave me money for a patent. I had explained to ChatGPT that I didn’t actually want to start a company, so ChatGPT structured terms that made it clear that there was no company to invest in.
I sent this offer sheet, which looked pretty good to me and was literally entirely written by ChatGPT, to my former CEO friend, hoping that I could use his large, large network of angel investors. He said, absolutely not, because angel investors want to invest in a company. They’re not interested in pennies on the dollar, he said.
Welp, a dead end.
ChatGPT, unhelpfully sycophantic, explained when I typed my friend’s email into the chat that it had basically sent me on a wild goose chase for about a week.
Filing a provisional — written by ChatGPT!
At this point, I had no idea what to do. I wasn’t about to spend my own money on patent fees. I was tempted, but that seemed like a bad idea, and my husband was not on board.
This is where expert advice came in. I went to a Women in Biology meeting for entrepreneurs and heard them literally say that you should not sink your savings into your idea at all, that it leads to calamity sometimes. I asked them what I should do. The answer I got from some very nice venture capitalists was to juice up the buy or something like that, basically meaning that I should get my husband’s company to want to buy a patent, and that if there was apparent market interest, I might have more leverage for something.
This was not completely unreasonable. ChatGPT was able, after much prompting, to find some companies that were nearly doing what I was thinking of but not, and had seen great results in batch yield optimization from earlier instantiations. With that, my husband had, groaning, sent his buddy at his company an email asking for a meet. The email was written in a way that ChatGPT told me would work: emphasize that the original prior art was technically award-winning with an Editor’s Choice, explain that there were companies out there using lame versions of this technology to great effect, and then offer to consult. There was some immediate interest from my husband’s buddy. If the prior art works at batch yield optimization, I suspect that there will be a bite that… well, I wouldn’t know what to do, and ChatGPT doesn’t really seem to know either, but it’s a first step.
But in order to even get consulting to work, I needed to file a provisional patent to stake a legal claim out of something that was half-prior art now.
Provisional patents are not like patents. You don’t even need a claims section. As one lawyer that I talked to stated, a provisional patent is basically my article with a cover sheet.
So I got ChatGPT to write my provisional patent. I inputted my idea, my drawings (which I did myself), and the article, and told ChatGPT to write the entire thing. It did so. I made minor modifications, checked that I could use my remaining start-up to pay the $65 fee for the provisional patent, and submitted the provisional patent.
God willing, that will be enough to have conversations with my husband’s work friend. If not, I’m sure I’ll plug the patent lawyer’s response into ChatGPT and hear ChatGPT sycophantically and unhelpfully say, “So sorry, I forgot this entire section of the provisional patent, of course.”
So how is ChatGPT at getting a new idea off the ground?
Meh.
I understand it’s early days for ChatGPT and other Large Language Models, that they’re likely to improve. But this is a use case in which it’s not even that ChatGPT had agency and was doing everything, like they made Claude do for a business (unsuccessfully) where you procure products for consumers. I was spot-checking things, inputting the basic ideas instead of asking ChatGPT for novel ideas, deciding on what was best to do based on what ChatGPT was saying. It’s the preferred use case of someone using a Large Language Model to help them, which is supposed to be much better than using a Large Language Model with no brain.
And yet, ChatGPT made error after error in its planned business attack. It should have just told me that angel investors wouldn’t be interested if I didn’t start a business. It didn’t. I needed to talk to venture capitalists and a former CEO to find that out and come up with any strategy at all for what to do.
Now, I do trust ChatGPT to write a provisional patent, but I’m not sure I would ever trust ChatGPT to write a patent, even though it’s technically passed the bar. God, that would save me 10–20K, if I could just do that. But probably, it’d cost me more money to do it that way, since every rejection of your patent application comes with a fee for resubmission.
If this provisional patent goes through, I’m going to roughly look at ChatGPT as a very fast administrative assistant. Unfortunately, it just can’t advise in a way that mimics human experts.
That’s sort of weird, because it has all of human knowledge programmed in. But maybe it never will be able to advise well, because it has sycophancy programmed in as well. There were points during the conversation where ChatGPT should have shot me down and told me a particular business attack wouldn’t work, and it just didn’t do that.
In sum, right now, I would not use an AI employee if the AI employee is ChatGPT-5 for anything too complicated. Human experience beats. ChatGPT-5 is not matching them, and it’s weird, because it actually did pass the bar and supposedly has amazing business acumen — right?