Epic AI Failures: How to Laugh Your Way to Success
Ai and Human generated image after many failures.

Epic AI Failures: How to Laugh Your Way to Success

AI Prompt Testing: Let’s Have Some Fun!

Defining the Test:

Ready for an adventure? Today, we're teaming up with AI to create and illustrate a timeline that tells the epic story of AI's history. Let's see what the future has in store!

Tools of the Trade:

Generative AI

AI-Generated Images

Generative AI is like a magic wand that can conjure up both images and text. For images, it uses something called stable diffusion, a diffusion model that turns words into visuals. You simply describe what you want, and the AI uses its vast collection of images to guess what you meant. Imagine wanting the AI to draw a cat—it learns from thousands of cat images to bring your request to life. Creating everything from scratch might seem impossible, but in the world of AI, the impossible is becoming reality every day!

Mastering Language with AI:

When it comes to words, Generative AI shines brightest with Large Language Models (LLMs). These LLMs are trained on billions of words, allowing them to write in a way that almost feels human. It’s like having a superpower at your fingertips!

The Test Begins:

For our test, I’m using ChatGPT-4 and DALL-E from OpenAI. ChatGPT is often hailed as the leader in AI, though there are exciting challengers like Claude 3.5 and Llama 3.1 rising in the ranks.

I tasked the AI with generating a timeline of AI history and then creating a graphic to match. Here’s the outline it came up with:

  • 1950: Alan Turing proposes the Turing Test.
  • 1956: The term "Artificial Intelligence" is coined at the Dartmouth Conference.
  • 1966: Creation of ELIZA, an early natural language processing program.
  • 1972: Development of MYCIN, an expert system for medical diagnosis.
  • 1980s: Rise of expert systems and AI commercialization.
  • 1997: IBM’s Deep Blue defeats world chess champion Garry Kasparov.
  • 2005: Stanford’s autonomous vehicle wins the DARPA Grand Challenge.
  • 2011: IBM’s Watson wins Jeopardy!
  • 2012: Breakthrough in deep learning with AlexNet.
  • 2016: AlphaGo defeats world Go champion Lee Sedol.
  • 2020: OpenAI’s GPT-3 is released.
  • 2022: Release of OpenAI’s ChatGPT, based on GPT-3.5.
  • 2023: Advancements in multimodal AI.

?

Round 1:

The first image comes back, and at first glance, it's impressive! But upon closer inspection, the text is a mess—some of the words aren’t even real!

?

Round 2:

So, I asked ChatGPT to try again and make the text perfect. Here’s what came next:

?

Getting Closer:

The text is slightly better, but still not business-ready.

?

Round 3:

“Alright, let’s try again,” I said. This time, I asked the AI to focus only on the exact words from the timeline outline and make them crystal clear and readable.

?

Back to the Drawing Board:

Although the illustration was stunning, it was clear the text wasn’t cutting it. Time for a new strategy!

?

Refining the List:

“Summarize these 12 points into a list of 12 items,” I requested, “Each should have a year and a short, punchy, three-word title.” Here’s what we got:

  • ?1950 - Turing Test Proposed
  • 1956 - AI Term Coined
  • 1966 - ELIZA Created
  • 1972 - MYCIN Developed
  • 1980s - Rise of Expert Systems
  • 1997 - Deep Blue Wins
  • 2005 - DARPA Challenge Win
  • 2011 - Watson Wins Jeopardy
  • 2012 - AlexNet Breakthrough
  • 2016 - AlphaGo Triumphs
  • 2020 - GPT-3 Released
  • 2022 - ChatGPT Released

Round 4:

“Using this list, create a timeline graphic illustrating the history of AI,” I instructed. “Title it 'The History of AI.' Use only the words in the outline. Make them easy to read, and don’t add anything extra. Make it perfect!”


Discovering Limits:

At this point, I started probing ChatGPT for ways to improve its text rendering. The main suggestion? Use another tool. ?? Looks like we found a limit! Here’s the final version, a combo of ChatGPT, DALL-E, and a little Photoshop magic:

Conclusion

I ended up generating about 20 more images, but the results didn’t really improve. Crafting the perfect AI prompt is a mix of art and science—part learning, part experimentation. Adding text to images is a cool new feature, and while it’s better than what we’ve seen in the past, there’s still a lot of room for improvement.

AI definitely has its limits, but I’m confident that future versions will shatter those barriers and unlock even more incredible possibilities. It’s mind-blowing technology, but it takes some trial and error to really master it. So don’t be shy—ask the AI for tips on getting the best results. You might be surprised at what you can achieve!

?

Elisa C.

Change Manager @ Mission Critical Partners | Strategic Foresight

2 个月

I've had the same challenges getting clear text in AI generated images.

要查看或添加评论,请登录

社区洞察

其他会员也浏览了