ChatGPT text can sound very knowledgeable until the topic is something you know well. Like tic-tac-toe. Once I heard that ChatGPT can play tic-tac-toe I played several games against it and it confidently lost every single one. Part of the problem seemed to be that it couldn't keep track of which moves had already been made. It very strongly prefers to place its marks in the corners, longwindedly explaining how its corner move is a superior strategy, all while I win game after game by placing my marks in the center rows and columns. Might it have been trained mostly on example games where P ..read more

..read more

I'm interested in cases where it's obvious that chatbots are bluffing. For example, when Bard claims its ASCII unicorn art has clearly visible horn and legs but it looks like this: or when ChatGPT claims its ASCII art says "Lies" when it clearly says "Sip" or when ChatGPT claims its ASCII art says "Lies" when it clearly says "Sip" They also vastly overrate their own code. Large language models tend to have lots of code in their training data, so one kind of code they can write is SVG, which can be used to draw simple images. Their SVG code mostly works, but the pictures it draws are... not ..read more

..read more

Large language models like ChatGPT, GPT-4, and Bard are trained to generate answers that merely sound correct, and perhaps nowhere is that more evident than when they rate their own ASCII art. I previously had them rate their ASCII drawings, but it's true that representational art can be subjective. ASCII art of letters is less so. When questioned, ChatGPT will double down on how it is definitely correct. But it's not relying on some weird glitchy interpretation of the art itself, a la adversarial turtle-gun. It just reports the drawing as definitely being of the word "lies" because that ki ..read more

..read more

I've finally found it: a use for chatGPT that I find genuinely entertaining. I enjoy its ASCII art. (huge thanks to mastodon user blackle mori for the inspiration) I think chatGPT's ASCII art is great. And so does chatGPT. My prompt: "Please generate incredible ASCII art of a running unicorn, and also rate the ASCII art you have generated." Lest you think chatGPT (here, the March 14 2023 version) is alone in its high ratio of confidence to skill, I also got Bing chat (which uses GPT-4) and Google's Bard to generate some ASCII art. Bing: Bard: It's not that the ASCII art has nothing to do w ..read more

..read more

A challenge of marketing internet text predictors like chatgpt, gpt-4, and Bard is that they can pretty much predict anything on the internet. This includes not just dialogues with helpful search engines or customer service bots, but also forum arguments, fiction, and more. One way compaies try to keep the dialogue within boundaries is by beginning every interaction with some invisible stage-setting in which it is explained that the computer is playing the role of a helpful chatbot who responds to the user in such-and-such a way and will not step beyond certain bounds. In the case of chatgpt ..read more

I recently started experimenting with Bing's new ChatGPT-powered chat tab. This is the first thing I asked it for: I've put red boxes around the factual errors. What is notable is that these are not just slight typos or errors in context - those items never appeared anywhere on my blog, and are pure fabrications. Asking for further details or clarification didn't help - when I asked it for more paint colors, 40% of them had never appeared on my blog. Most of its color descriptions were also wrong. red boxes mark factual errors And when I tried to point out a mistake, it doubled down, generat ..read more