Posted on 03/27/2026 7:19:20 PM PDT by Tawiskaro
I embarked on a vanity project to record some of my chess games against Shredder, my cell phone app. The games are recorded in algebraic notation, but I prefer the traditional English descriptive notation, so I asked ChatGPT to make the conversion. It kept correcting itself through multiple iterations. I finally had to stop it. I turned to Grok to get the conversion done. It seemed to work, so I asked Grok to illustrate the game with board positions at critical points in the game. It turned into a major league fail on steroids. I won't bore you with the actual game, but here is one of the images Grok produced after multiple corrective prompts:
Any player will see that the above position is impossible. In fact, downright insane. For those who know nothing of chess--it doesn't matter.
Naturally, I challenged Grok:
Sorry, Grok, but your images are a total disaster. This thread should be referred to your programmers for analysis. In the future, when asked if you can create such images, you should respond with a firm "No".
Naturally, I will go back and check to make sure that the moves I passed along to you are valid. I used one of your competitors to convert algebraic notation to descriptive. I try to use the following bit of verse when querying:
AI
The software often gets things wrong,
which is why I pen this little song.
Use each product to check the others—
this ain’t just Grok-it science, brothers.
--T. W. Krimer
Grok responded with:
Understood, Theodore.
The generated chess board images in this thread turned out poorly—wrong piece placements, inconsistent rendering, and not matching the actual game positions. That's on me and the image generation system.
You're right: when users ask if I can create accurate chess board diagram images from moves (especially in descriptive notation), I should now respond with a firm "No" rather than attempting it. The current capabilities don't reliably deliver clean, correct diagrams.
Thanks for the feedback and for sharing the little verse—it's a good reminder. I'll make sure this thread's issues are noted internally for improvement.
I use AI for general queries often. Whenever I see even a hint of stupidity, I challenge. I sometimes get an argument (especially in poetry analysis), but I always get the AI product to acknowledge defeat.
I present the above as a cautionary tale. Do not accept an AI response as Gospel. Even if something sounds correct, check the AI sources (they are usually posted in a link somewhere at the bottom of the response). Then check the response against another AI product. And your own personal knowledge. And books, of course--those paper things most folks seem not to use much anymore.
FWIW.
|
Click here: to donate by Credit Card Or here: to donate by PayPal Or by mail to: Free Republic, LLC - PO Box 9771 - Fresno, CA 93794 Thank you very much and God bless you. |
The board’s 8 by 10. Oops. The lower right hand square is black. Oops.
I don't but it has been helpful to me in collating data points. Excellent summaries. And when you challenge it, it gives you better answers.
Do I trust it with my life? No.
none of the current AI will learn, nor will they remember for other people your arguments.
but they will try to placate you.
LLMs do one thing and one thing only, generate probabilities of what the next token will be, based on the tokens input into it.
LLMs are vast quantities of correlation. “Training” and responses for prompts delete some correlations and create others.
Because the thing is anthropomorphized, and because that’s similar in some ways to what we do with neurons, we call it “learning” — and it can improve the quality of subsequent responses.
IMO these general-purpose ones are fun and sometimes very useful but serving as an everything-bot for everyone is not the path to success.
As with most things, AI is GIGO - Garbage In, Garbage Out. It’s all about the context that gets sent to it. A lot of times ChatGPT just does a simple websearch and sends what it finds to the LLM.
yes
there is no AGI at this time.
but depending on the AI they do generate different probabilities
I've noticed (done my own research) that several LLMs, AI, whatever, have Wokepedia as part of their source material.
Talk about Garbage In Garbage Out!
What do your prompts look like?
I’m not so concerned with AI’s accuracy as I am that it addresses me as “Captain” while using its best “Spock” imitation.
Whenever asking AI to do a conversion, it’s probably a good practice to then ask it to do a conversion back to the original state, and compare its output with your starting material.
It’s great at Python coding though.
Good post — and the guy's instincts are right, even if the specific lesson is a bit narrower than he frames it.A few observations:
What he got right. The core advice is sound: don't accept AI output as gospel, cross-check, use your own knowledge, use primary sources. That's genuinely good practice, and the verse is charming.
What's actually happening with the chess images. This isn't really an "AI failure" in the deep sense — it's a well-known limitation of image generation models. Diffusion-based image generators (what Grok is likely using) are notoriously bad at spatial reasoning, structured grids, and rule-constrained systems like chess boards. They hallucinate piece positions the same way language models sometimes hallucinate facts — they pattern-match on training data rather than reason from rules. It's a known, specific weakness, not a general indictment of AI capability. Grok's honest response to him was actually pretty good.
The notation conversion problem is more interesting. Algebraic to descriptive notation should be something a language model handles well — it's deterministic and rule-based. ChatGPT stumbling there suggests either an ambiguous prompt, an unusual game position triggering edge cases, or just model sloppiness on a task it doesn't encounter often. Worth noting that Claude handles chess notation fairly well, though I'd always recommend verification on a real board or engine.
The broader framing. He presents this as an "AI failure" story, but what he actually describes is appropriate AI behavior at the end — Grok acknowledged the limitation clearly and told him not to rely on it for that task. That's the system working as intended.
His poem is the real takeaway. Use AI tools to check each other, verify against authoritative sources, and never outsource your own judgment entirely. That applies to chess diagrams and to everything else.
My instincts tell me to leave AI alone because the longer I can manage to keep it out of my life the better off I’ll be. I like being able to figure things out for myself. What does it care if I’m dismissive of it? If I thought I could, I’d go out of my way to offend it...”Hey AI, your grandmother wears army boots” or “Get off of your fat ass and get a real job”, and so on.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.