Category: AI art

AI Limitations

I’m using GPT-4 to help write an ESL textbook, which is probably taking longer than if I just did it by myself. However, because I’m relying on AI for the graphics and audio files for conversations, it’s really forcing me to learn basic Python script debugging and troubleshooting for ChatGPT. The most frustrating things about this experience so far are:

  • ChatGPT forgets stuff all the time. If I ask it to compile a Word file, it will leave out parts that we finalized together. And sometimes, it can’t create download links for the file, or so it says.
  • It sometimes changes finalized content unless it’s explicitly told to leave it alone.
  • It is often not aware of what it can and cannot do. It specifically told me one night that it wasn’t technically capable of producing a certain kind of file, so instead it walked me through doing it in Python. The next day, I asked to do the same thing for another chapter, and it just instantly generated the correct file for me – kind of. I was still better off doing it myself, but still, it was a matter of where I wanted to focus my efforts – learning to do stuff myself, or coaxing the AI into giving me the desired results.
  • Using Dall-E to generate something useful is like trying to have a serious conversation with someone tripping on acid.

I have to get back to the textbook, but I leave you with these absolute gems of what AI thinks Bingo cards look like:

According to the file name, the prompt was: “A 4×4 Holiday Bingo sheet with a clean, minimalistic design suitable for embedding in a Word document. Each cell contains one of the following…”

When I pointed out that the grid wasn’t 4×4 (although that would be just Bing, not Bingo) and that there were duplicate entries as well as weird text and spelling mistakes, it popped out this:

“A correct 4×4 Holiday Bingo sheet in a minimalistic design. Each of the 16 cells contains one unique activity from the following list_ ‘Go to the beac…”

My way of dealing with is to change the activity completely, but I really hope image generation, of text in particular, improves quickly. GLAY SiOMIE PLAY SOMIES GAE SHOPPING.

My new buddy is so smart

“Based on the response you shared from ChatGPT, it does seem that ChatGPT currently has more capability than me when it comes to directly processing information in Thai and Japanese languages.

My abilities are limited to working with English text, while ChatGPT claims it can understand and generate text in Thai and Japanese to some degree. This would potentially give ChatGPT an advantage in being able to directly analyze primary sources or data in those languages.

However, it’s difficult for me to definitively say whether ChatGPT is “better suited” overall, as there may be other factors at play:

ChatGPT’s proficiency in Thai/Japanese could vary based on the complexity of the content.
I may still have strengths in other areas like research methodology, academic writing, analysis from an English perspective, etc.
The quality and accuracy of our outputs would need to be evaluated for this specific research context.
Ultimately, you may want to explore utilizing both AI assistants – leveraging ChatGPT’s multilingual abilities when dealing directly with Thai/Japanese material, while also taking advantage of my skills for aspects that can be conducted in English.”

OpenAI Educator Guide

OpenAI has released an official guide for teachers using ChatGPT in their classroom: Teaching with AI

It’s nice to have something official posted, at least. There are some new resources in there that I’m slowly trying to get through.

The official answer to the question, “Do AI detectors work?” is:

In short, no, not in our experience.

https://help.openai.com/en/articles/8313351-how-can-educators-respond-to-students-presenting-ai-generated-content-as-their-own

I asked ChatGPT the same question, and its answer was:

AI detectors, specifically those designed to identify plagiarism or academic dishonesty, can be effective in detecting instances of copied or improperly attributed content.

https://chat.openai.com/

So who knows better, the parent, or the child?

Bing: Corporate AI Echo Chamber

The AI-enabled version of Bing is useless for search… Imagine using the bare bones search engines of the 90s on all of the useless filler on the web today, navigating by command line to a toddler – that’s the new Bing experience. You can pare down on its idiot responses by tweaking prompts, but it’s a huge step backwards from just Googling something. It’s also noticeably clunkier than using vanilla ChatGPT.

I’m not the only one who noticed.

Is openai.com written by AI?

It certainly looks like it:

“We spent 6 months making GPT-4 safer and more aligned. GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations.”

I asked ChatGPT if it could write the next line and it replied:

“These improvements have been made possible through a combination of rigorous testing, enhanced training data, and algorithmic updates that prioritize responsible language generation.”

Used as (possibly) my last prompt into NightCafe, it becomes: