How ChatGPT scanned 170k lines of code in seconds, saving me hours of work

Undecided how you can greatest apply synthetic intelligence (AI) to your distinctive, specialised wants? You have come to the appropriate place. We’ll go over how you need to use a software like ChatGPT to resolve complicated issues rapidly, as long as you’ve got the appropriate prompts and a touch of skepticism. 

First, a quick 3D printing rabbit gap

Our context for this lesson is 3D printing. A particular take a look at in 3D printing known as a 3DBenchy checks printer efficiency by serving to customers take a look at velocity and varied print-quality measures, and it takes most printers an hour or two to print out.

I just lately examined a brand new printer that is purported to be quicker than many others. On this printer, the Benchy took 42 minutes, whereas on different 3D printers within the Fab Lab, it took 60 to 70 minutes. However here is the factor: the take a look at model supplied by the corporate that makes the printer took 16 minutes. That is a heck of a distinction.

3D printers are managed with G-code, a program custom-generated by a software known as a slicer that controls how the printer strikes its print head and print platform, heats up, and feeds and retracts molten filament.

The pre-sliced G-code supplied by the manufacturing facility for the printer I used to be testing resulted in a 16-minute print. The G-code I generated utilizing the corporate’s slicer resulted in a 42-minute print. I wished to know why.

Sadly, nobody on the corporate’s help workforce might reply my query. Regardless of quite a few tries, I could not get a solution about what slicer settings to alter to get the G-code I produced utilizing their slicer to carry out in addition to the G-code generated utilizing their slicer.

After many internet searches and studying posts from annoyed Reddit posts, it was clear that different prospects had the identical downside. Here is a machine able to greater than double the efficiency, but none of us might reproduce that efficiency efficiently.

Can AI assist?

That is the place ChatGPT comes into the image. G-code consists of 1000’s of traces that appear like this:

G1 X122.473 Y140.422 E4.23368
G1 X122.222 Y140.6 E4.24443
G0 F14400 X122.703 Y140.773
G1 F3600 X122.117 Y141.251 E4.27083
G1 X121.767 Y141.66 E4.28963
G1 X121.415 Y142.139 E4.31039
G1 X121.105 Y142.641 E4.33099

Collectively, each Benchy G-code information had 170,000+ traces of code. I did not intend to spend a Saturday afternoon sifting by means of that stuff manually. However I believed, maybe, AI might assist.

I had the G-code I generated utilizing the slicer. I might additionally export and save the G-code supplied by the manufacturing facility. Utilizing ChatGPT Plus, I fed each information into the AI.

I began by confirming ChatGPT might learn the information. After I uploaded every file, I requested:

Are you able to learn this?

ChatGPT confirmed, stating, “I can learn the contents of the file. It seems to be a G-code file, usually used to regulate 3D printers.” That was a superb begin.

To make sure we have been clear on which file was which, I gave ChatGPT some labels for the information:

Let’s name the primary file uploaded “common print” and the second file uploaded “quick print”. Okay?

The AI agreed, so then I instructed it:

Describe how the information differ.

Apart from naming one of many information “quick print”, I gave ChatGPT no indication of what I used to be on the lookout for. Even so, the bot recognized that one print had increased print speeds, though the temperature settings have been the identical.

I then requested the AI:

Why is “quick print” a lot quicker?

At this level, ChatGPT began to bother me. As a substitute of giving me particulars from the code I supplied, it speculated. The AI used phrases containing “probably,” “might,” and “may” to explain why the print was quicker.

However I had given it G-code information that described precisely what the printer was doing, so I wished an actual reply about what the printer was doing. As is usually the case with ChatGPT, the dialog was loads like speaking to an excellent grad scholar who’s considerably cussed and uncooperative.

Why the appropriate immediate makes all of the distinction

I lastly landed on this immediate, which teased out workable solutions:

The G-code supplied in each information is the one factor that’s totally different for these prints. Utilizing solely the G-code supplied as comparisons, what slicer settings can be totally different? Do not speculate on what different settings could be. Base your evaluation solely on the code supplied.

ChatGPT recognized three key elements:

• The common print G-code reveals feed charges usually between 2400 mm/min and 3400 mm/min. The quick print G-code reveals a lot increased feed charges, with some instructions reaching as much as 14400 mm/min. This price is a major enhance in velocity, instantly contributing to the decreased print time.
• In comparison with a layer peak of 0.2 mm for the common print, the quick print set layer peak to 0.25 mm, lowering the variety of layers wanted to finish the print.
• The quick print G-code contains rather more aggressive acceleration settings, permitting the printer to succeed in increased speeds faster and preserve them, contributing to quicker printing.

That outcome was fascinating. Nevertheless, I wished to know whether or not the corporate hand-optimized the G-code or generated it instantly within the slicer. So, I requested ChatGPT:

Are you able to inform if quick print has been hand-coded or was generated by a slicer? Maybe search for inconsistent instructions or non-standard feedback.

The AI responded with three fascinating issues:

• The variations in G-code instructions like adjustments in feed charges, switching between totally different motion instructions, and alterations in different parameters are typical of slicer-generated code. These adjustments are designed to optimize the print course of and are in step with the conduct of a slicer.
• If the G-code have been hand-coded, we’d count on to see extra irregular and unpredictable patterns, however these transitions comply with a logical sequence, attribute of a slicer’s optimization course of.
• Due to this fact, the presence of those instructions doesn’t point out hand-coding; quite, it helps the conclusion that this G-code was generated by a slicer.

What these outcomes inform me is that it’s in all probability doable for customers to change their slicer settings to get related efficiency.

Some ideas from the feedback

We have had some very energetic feedback for this text. For essentially the most half, I’ve gone in and answered questions as they got here up. I encourage you to go to the feedback to take part and browse what different readers should say on this matter. 

Here is a fast record of some ideas that the feedback impressed:

• You do not have to make use of AI to resolve these issues. However it may be instructive to see how AI handles fixing many classes of various issues, so I encourage you to make use of chatbots, even when simply to see what they’ll do. Then, you will have a software for future use.
• I am typically utilizing ChatGPT Plus for these exams. The free model can do a few of the work, however as a result of it throttles utilization after an excessive amount of interplay, if I am to get an article completed on deadline, I have to have full entry to the chatbot.
• When attempting to wheedle a solution from a chatbot, I often consider it as a proficient scholar or worker. Typically, I will even use “please” and “thanks” to maintain the conversational tone going. However as anybody you speak with may get distracted or stubbornly refuse to see your level, maintain attempting, change up your questions, ask questions in several methods, and make clear, even whenever you assume what you are clarifying needs to be apparent.
• Utilizing AIs could be lazy, positive. However they’ll additionally save time. Virtually something you are able to do with energy instruments in a workshop, you too can do with hand instruments. Typically you wish to deal with the artwork. Typically, you simply wish to get completed so you’ve got time for sufficient sleep that night time. By no means hesitate to make use of a software if it can provide you a leg up. However do take into account that in case your work output particularly precludes the usage of these instruments (for instance, we by no means use AI to put in writing our articles right here on ZDNET), then do not use the software.
• Positively take care with what you add to the AI. Not every thing is corporate or in any other case confidential. However some issues are. Most of what I add is my very own information, or, as on this case, one thing that is already revealed within the open.

I do not at all times get the prospect to answer feedback, however I attempt. Typically, folks publish days, weeks, and even months after the articles go up and I’ve moved onto different articles. However I at all times welcome reader feedback. As a result of most ZDNET readers are execs, the feedback are sometimes wealthy with helpful (if sometimes painful to learn) info. I’ve discovered loads from ZDNET feedback, and I am positive you’ll, too.

What have we discovered?

We have discovered that ChatGPT understands G-code. That is unsurprising as a result of, in my earliest exams, we discovered that ChatGPT has a reasonably good command of even essentially the most obscure programming languages.

We additionally discovered that ChatGPT can sift by means of and evaluate 170,000+ traces of machine directions and attain actionable conclusions in seconds.

Lastly, we discovered we are able to use AIs like ChatGPT to discover complicated issues from a number of angles. Not solely did ChatGPT clarify the huge velocity distinction between the 2 information, but it surely was additionally capable of validate whether or not or not the factory-provided file had been hand-tweaked.

In conclusion, do not settle for what the AI tells you as absolute fact. Do not make essential choices primarily based on its solutions. And keep in mind that you typically have to barter with the AI earlier than it is prepared to offer you useful solutions.

This take a look at is yet one more case the place I have been capable of flip to the AI and discover a solution for a really me-specific query with out coding in minutes.

You probably have a query that requires plenty of textual content or numerical evaluation, think about operating it by ChatGPT or one of many different AIs. You may get a helpful reply in minutes.

Writing this text about the issue took me just a few hours. The precise evaluation course of, from begin to end, took me lower than 10 minutes. That is some severe productiveness proper there.

You possibly can comply with my day-to-day undertaking updates on social media. You’ll want to subscribe to my weekly replace publication, and comply with me on Twitter/X at @DavidGewirtz, on Fb at Fb.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.