I’m very anti-AI. I have yet to see any results that make me go “Wow, I need this!” But - I’m not a total Luddite. If AI could do something useful for me, I would consider using it.
To that end, I have some old schematics which were done in a currently unsupported program. The file format is binary, and not publicly documented. Are any of the AI tools able to take these files and convert them into a newer, well-documented format, or is that asking too much? I’m willing to spend some small amount of money to do this, but not very much, and I don’t have a huge number of documents to train AI on. Is this currently achievable?
I also have some old PCB layouts where the file format is ASCII, but is also not publicly documented. I would like to convert those, too.
AI only works if you’re good enough to spot when it’s made a mistake. Otherwise it could slip in lots of errors without your realizing it. It’s a fantastic time-saver, money-saver, energy-saver in 90-99% of situations, but it’s those small issues you must watch out for. I wouldn’t use AI for something that was absolutely important or must be error-free. For me, even when doing a lot of document revision, I still manually proofread everything it’s done.
What’s being called Artificial Intelligence (AI) nowdays is really just an expanded version of predictive input. Way back years ago we had applications that watched what you were typing and tried to ‘help’ you by predicting what you would type next. Mostly they were so inaccurate & intrusive that most people considered them a dang nuisance* and turned them off (if possible).
The new AI still basically tries to predict what word(s) should come next, based on your input, or your directions to it. Technically it’s called an Large Language Model (LLM) because it now has billions of online language samples to analyze, and so can do a much better job producing what you want to say (based on copying what many, many others have said). But it has no ability to recognize the accuracy of what it copies – so if many online sources give the same wrong info, it just accepts that. And it knows the rules of grammar & linguistics, so it can parrot this in a reasonable sounding manner – which is why in can sometimes produce completely wrong answers that seem so correct.
For the OP’s problem, I don’t think AI would help much. Only if there were a lot of online examples of people converting schematics into your newer format would AI be able to reproduce that for you.
I have a similar hesitation towards using AI, but your envisaged use case seems pretty harmless to me, especially if you take into account Velocity’s recommendation to carefully check for errors.
I just asked what some AI say themselves. Kimi (a Chinese AI) says it can’t do it itself but recommends specific programs/tools. ChatGPT boasts about being able to do the job itself, but asks which ASCII format you have. My suggestion is to give ChatGPT a try with the files you have, unless they are sensitive (national security/privacy wise). I always presume any AI will keep your data and use it for its own purposes.
I’m a tech writer, being told to use the shiny new AI tools, and AI auto complete for text goes from amazing to nonsense at the drop of a virtual hat. It will correctly predict the need for “complete the following:“ before a numbered procedure, for about ten edits then decide to add tonnes of fluff text around it on the eleventh.
I’ve found it much more useful as a powerful find and replace tool. Instead of me having to build up a complex find and replace instruction using regex etc. I can just type one sentence with the instruction and the tool will do it.
Or it can find parts of a document that match very specific but weird requirements. i.e. a bullet list that follows a third level heading but only if there isn’t text between the heading and the list but there is a blank line and the content is not in a task. This is a real example I used the other day and it saved me about an hour of work.
Testing it for content generation though and it’s mostly useless beyond creating a very rough and ready draft document.
For the ASCII formatted one, try going to Claude and uploading the file. Then, ask it if it can convert it or whatever you’re trying to do. Maybe it can convert it to Mermaid or HTML or something.
For the weird binary format, I don’t know.
The question in the title doesn’t really match the OP, but to answer the title question, you can talk to it in natural language and ask what it can do. If the results aren’t what you expect, explain why and tell it to try again. I use AI all the time now for little programming tasks, and if there are errors, I just tell it what the errors are and it fixes them for me.
It’s true, of course, that in technical terms, what LLMs do is predict the expected next word in a text based on correlations observed in the vast data used to train the model. But I think it’s wrong and smug to conclude from this that LLMs are essentially just a glorified autocomplete. The things that modern LLMs (especially the pay-to-use tiers) are capable of accomplishing are astonishing, and we haven’t yet seen a fraction of the impact they will have on job markets. And that’s just LLMs, without speaking of autonomous agentic AIs. Ultimately, as long as AIs get a complex job done fast (and they do), it doesn’t matter whether, underneath the hood, they’re “just predicting tokens”.
Man, do they ever. Nine months ago, we were barely using them and now they’re an essential part of our workflow. They can look at a document with hundreds of pages and come up with a really insightful summary in seconds.
I’m starting to think that all intelligence is just predicting tokens at this point, given how good the AIs have gotten so quickly.
To set a mental model of what a modern LLM (such as ChatGPT) can and can’t do, it helps to think of it as : a person who has read all the relevant parts of the Internet, including many books it shouldn’t legally have access to. It remembers this content (not exactly all of it, but certainly more than you or I could), and can write and interact on any of these subjects while imitating the style of humans.
1 Use AI as a search engine. Ask if the AI can tell you how to convert this, or if someone else has already done the work and published his knowledge / attempts / converter on the internet.
2 You don’t say whether you still have that old program or can still run it. AI might tell you how to run the program, by using compatibility options, or walk you through setting up a virtual machine in which the program will run .
3 If you, in any way, have a picture of the finished schematics, AI might (I really don’t know) be able to turn that picture into a description or file format that a modern program can use.
4 You could try using a coding agent (Github Copilot, Claude etc.) and instruct it to write a program that will automatically try to figure out the file format, maybe given some clues as to the components found in the schematics. Definitely an involved process.
5 You could ask AI to guide you through deciphering the file format yourself. Start with asking what tools you need and how to understand the file header.