The holidays are upon us! We’ll be taking a bit of a hiatus for Thanksgiving, so there will be no newsletter next week. This week, try our DIY AI below. Get creative for the holidays and learn something new! Or new-ish. Oh, and there was some breaking news. See below. Happy Thanksgiving, everyone! I’ll be back with the bot correspondents on December 3. As always, if you like what you see here, feel free to drop something in the tip jar.
Update!
Well, this wasn’t on my list of possible outcomes for this story. There are reports that Altman and Brockman are in talks that could put them back into their seats at OpenAI.
It seems that the investors are pushing for it. Note that because of the way that the company is structured, they are not beholden to shareholders and were well within their rights to dismiss Altman.
However, all hell seems to have broken loose among the employees and investors. The board appears to have underestimated the level of support Altman has. Even if he doesn’t return, his future is bright. Investors are lined up to support whatever he decides to do.
We still don’t know precisely why they fired him. The main theory is that there’s a spilt between the non-profit researcher types who are worried about pushing out AI that could potentially be dangerous, and the entrepreneurial side looking to release new features ASAP in this competitive AI environment.
Here’s the thing. They can’t be that far apart. It’s not likely that Altman wants to unleash something that will eradicate humankind. Because, well, there goes his investments! Seriously, tho, it’s not clear why the two sides can’t come to some kind of agreement and reasonable timeline for pushing out new product.
Altman, with his background in venture capital, was doing the job he was dispatched to do. I admire OpenAI for building the non-profit responsibility wing of the company. But they have to understand that if they want to reach their goals, they may have to hold their noses as consumer-focused products like laundry bots get released and draw investor funds. Both sides need to work together to keep us all safe.
To be continued, apparently. Here’s what I wrote yesterday:
OpenAI (ChatGPT) Ousts CEO. What’s Next?
By Michelle Johnson, AI Insider
The biggest news in AI this week dropped Friday afternoon, as often happens when companies have bad news and want to muffle media coverage or avoid a rapid decline in their stock price. However, this was too big to hide under a Friday bushel. The board of OpenAI (ChatGPT) ousted Sam Altman from the CEO perch.
No one knows why at this point. There’s shock and awe in AI land. Plenty of reports are out there documenting what’s transpired since this happened. Rather than recount them, let me do some crystal ball gazing about what may come next. Note: This is only my two cents. It’s not based on any deep research or anything, just my observations as an AI enthusiast.
First, unless he was ousted for something really, really, really egregious, Sam Altman will land on something interesting. No need to worry about him. Seen as the face of AI and the steward of technology that’s changed the world, there will be no shortage of deals coming his way.
He may build out something that he already has a financial stake in, like the Humane AI pin, or he’ll align himself with another company that’s doing exciting work more in line with his thoughts about where AI is going. Or he’ll build that next big thing that he’s been wanting to do but has been hobbled by the demands of growing OpenAI.
In addition to the now former OpenAI President Greg Brockman, others will follow Altman out of the door. Maybe they’ll form the core of a new team to launch the aforementioned next big thing. There are reports of other resignations happening as we speak, and some speculation about a new venture, so I’m not some kind of forecasting genius here.
Microsoft’s stock may take another hit on Monday like it did Friday after the news was announced too late in the day to have much impact. Just briefly, though. It’ll go back up. (Black Monday Sale!) Microsoft went all in on AI and OpenAI. They are likely upset about being kept out of the loop on such a significant move as firing a CEO. But they are already on the AI bullet train with their $13 billion investment in OpenAI and rebranding practically every one of their products as AI-powered.
Finally, I wouldn’t be surprised to see some of the rapid development coming out of the OpenAI factory start to slow down. And maybe fewer announcements of major collaborations with giant for-profit enterprises. Altman took an open-source product in directions that strayed far from the original mission. It’s possible he went a little too far, and that’s what rankled the board. We shall see. Stay tuned.
DIY AI: Picturing the Holidays
By Michelle Johnson and ChatGPT, AI Insider
Welcome back to DIY AI, where we dive into the world of artificial intelligence to learn something new in just a few minutes. This week, we’re switching gears to focus on image generation, just in time for the holiday season.
Whether you’re celebrating Thanksgiving, Christmas, Hanukkah, or Kwanzaa, all (ChrisKwanukkah in my household), or none of the above, this assignment will add a little AI-powered cheer to your festivities.
Note: If you’re a total newbie, scroll down to “Reading Assignment” and learn about how AI image generators work, then try this week’s assignment. If you have some experience, dive right into the assignment.
This Week’s Assignment: Creating Holiday-Themed Images
For this task, you can use one or both of these two popular AI image generators that are great for beginner and intermediate users: Dall-E 3 and Adobe Firefly. Our goal is to create 2-3 images that capture the essence of your chosen holiday or other topic.
1. Choose Your Holiday Theme: Select from Thanksgiving, Christmas, Hanukkah, or Kwanzaa or a topic of your own choosing. Think about the key elements that represent the holiday or other topic for you.
2. Craft Your Prompts: Write a detailed prompt for each image. Remember, the more specific you are, the closer the output will be to your vision. Include elements like the main subject, colors, and mood.
Example Prompt for Thanksgiving: “A traditional Thanksgiving dinner table setting with a golden-brown turkey, cranberry sauce, various sides, and a sweet potato pie, set in a warmly lit dining room.”
3. Generate the Images: Use Dall-E 3 and/or Adobe Firefly to bring your prompts to life. Try generating the same prompt on both platforms to compare the results.
4. Reflect and Share: Once you’ve created your images, take a moment to reflect on the process. How did the tools interpret your prompts? Were there any surprises? Share your favorite image and your thoughts in the comments here.
This assignment isn’t just about creating pretty pictures; it’s about understanding how AI interprets our instructions and turns them into visual art. So, have fun, get creative, and let’s see what magic you can create with AI!
Extra Credit: Check out Ideogram.ai. It generates text on your images! Generate an image and have it add “Happy Thanksgiving!” or other text. Try this on Dall-e and Firefly, too. Which one does the best job with text?
Reading Assignment: About AI Image Generators
AI image generators work very similarly to chatbots like ChatGPT, Bing, Bard, etc. You type in a prompt, and you get a response. In this case, you get back a visual. Depending on the settings, the bot may generate one image or as many as four.
That image could look like a photograph, an oil painting, or a children’s book illustration, depending on what you ask for in the prompt. It knows and understands many, many artistic styles.
Be aware that this tech is still in development and may not always give you exactly what you ask for. It sometimes takes some poetic license and generates something in the “neighborhood” of what you requested.
Some people get confused and frustrated by this. They don’t understand why they ask for something specific and get back something that’s not even in the ballpark.
Just know that there are various advanced ways to tweak “prompt adherence” (beyond the scope of this lesson), but none of this is your fault. It’s how the bot is trained. Most companies are moving their products toward better prompt adherence based on user feedback.
Like a chatbot prompt, image prompts can be simple, complex, or lengthy. One of the best ways to learn how to write image prompts is to look at other people’s prompts on public feeds.
You can easily eavesdrop on other people’s prompts. To that on Dall-E, scroll down to the “Explore Ideas” tab and mouse over any image. The prompt used to generate it will appear. In Adobe Firefly, click the “Text to Image” link. That will bring you to a gallery of featured images. Mouse over any one of them to see its prompt.
For practice, try recreating some of the images in the public feeds. Type in the same or a similar prompt of any image and see what you get. Then, complete this week’s assignment.
Bing Chat Renamed Copilot
By Perplexity AI and ChatGPT for AI Insider
Microsoft’s Ignite developer conference this week featured several key announcements in the realm of AI.
The company announced the rebranding of Bing Chat to the name Copilot and the launch of Microsoft Copilot Studio, which allows users to build custom AI copilots. Microsoft said that the rebranding of Bing Chat to Copilot is meant to simplify user experience and make AI more accessible.
Copilot is now accessible through various channels, including a web address, via the Microsoft Edge web browser’s sidebar, and as an assistant on Windows 11, with its new home at copilot.microsoft.com.
Bing Chat Enterprise, initially announced in July, caters to organizational needs with commercial data protection. This version, now called ‘Copilot with commercial data protection,’ is included in select Microsoft 365 subscriptions or can be purchased as a standalone option for $5 per user per month.
The new Copilot will be generally available from December 1.
The company also announced Microsoft-built CPUs and AI chips, which are meant to lower the cost of delivering artificial intelligence services.
ChatGPT Plus Subscriptions Paused
By ChatGPT for AI Insider
OpenAI stopped accepting new sign-ups for its ChatGPT Plus subscription service due to a surge in demand caused by the introduction of new features and the growing popularity of the AI chatbot.
The surge in sign-ups, after the unveiling of features like GPT-4 Turbo and a GPT builder exclusive to Plus and Enterprise users, overwhelmed the service’s capacity.
OpenAI CEO Sam Altman announced the pause following the company’s inaugural development conference, OpenAI DevDay, held in San Francisco.
ChatGPT Plus, launched in February for $20 per month, offers enhanced features over the standard ChatGPT. These features include faster response times, priority access to new features, and uninterrupted access during peak usage times.
This subscription plan is designed for users who want a more responsive and feature-rich experience with ChatGPT.
The pause in ChatGPT Plus subscriptions highlights the challenge of scaling AI services to meet growing demand.
Tracking Changes Made to Images Could Get a Lot Easier
By Claude for AI Insider
If you’ve ever collaborated with others on editing digital images, you know how tricky it can be to see exactly what changes were made by whom and when. You might end up having to visually scan back and forth between different versions, trying to spot what’s different. Well, Adobe has filed a patent application for a new system that aims to solve this problem.
It’s essential to address this now that AI tools are embedded in applications such as Photoshop that can change images in a blink.
Let’s walk through an example of what Adobe is proposing:
John is working on an image of a park with his colleague Mary. John uploads the original version to their shared online workspace. It shows a tree on the left and a bench on the right.
Mary opens the image and decides to add a playground set in the middle. She makes this change and saves a new version.
Later, John opens Mary’s edited version and selects the tree as his point of interest. The system recognizes this tree was also in the original version. It checks the image metadata file in the background and sees that a playground was added by Mary at 2:30 pm on Tuesday.
The system then automatically generates a text summary for John saying “Playground added to center of image by Mary at 2:30pm on Tuesday.”
With this simple summary, John can now clearly see the exact modification made at the specific point of the tree he selected. Without this feature, he would just know the overall image changed but not what exactly changed.
The magic lies in the system’s ability to let you pick any point in the image – whether an object like the bench or a region like the sky. It will then track changes made to that point across all versions and summarize them for you. This gives collaborators complete insight into edit changes at a granular level.
YouTube, Google DeepMind Unveil AI Music Experiment
By Perplexity for AI Insider
YouTube has unveiled a set of AI-related music experiments, Dream Track for Shorts and Music AI tools, developed in collaboration with Google DeepMind.
Dream Track allows users to create unique soundtracks of up to 30 seconds for their YouTube Shorts using the voices of famous artists. Nine artists, including Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Papoose, Sia, T-Pain, and Troye Sivan, have collaborated in this experiment, giving a small group of select US creators the chance to make soundtracks in their style.
The Music AI tools, on the other hand, are designed to help musicians with their creative process. These tools enable artists to more seamlessly turn their thoughts and ideas into music, like creating a new guitar riff just by humming it or giving a pop track a reggaeton feel. Music AI Incubator participants will be able to test these prospective tools later this year.
Despite the tremendous opportunities AI presents, YouTube acknowledged that the quickly evolving development of AI brings with it some complex challenges. The platform pledged a commitment to collaborating with music industry partners to explore new opportunities and develop sensible and sustainable controls, monetization, and attribution frameworks.
Artists participating in the experiment view it as a test for what AI music creation can do. Demi Lovato, for example, believes that as artists, they need to be a part of shaping the future of AI in music.
What the Actual Hell?
Google Labs is beta testing a feature that allows people to attach notes to search results. What could possibly go wrong? And do we really need to see other people’s thoughts this way?
Random Shorts
Perplexity AI, the chatbot that you may have only heard of on AI Insider, has upgraded its responses to now include visuals when available. So, if you ask the bot something about the Grand Canyon, you can expect images of pretty vistas to come up with its text response.
Microsoft unveiled a tool capable of creating photorealistic avatars. Named Azure AI Speech, the text-to-speech avatar maker allows users to generate videos of speaking avatars in multiple languages. However, Microsoft acknowledges the potential for misuse of such technology and claims to be building in safeguards.
(Unless otherwise noted or a screenshot, all images are generated via Dall-E 3.)