And… we’re back! Did you miss us? I enjoyed attending my niece’s wedding in scenic Ithaca, N.Y., this week. We also fit in some stops at historic sites on the drive. And while I was away, I also celebrated a birthday (along with my twin brother, the father of said niece). While I wasn’t following AI news like I would at home, I did spot a few things worth noting today. So, let’s catch up on what we missed and what’s happening now. Bing, Bard, Claude, Perplexity, ChatGPT, and I have been patiently waiting to bring you some updates.
I Name Checked Midjourney
By Michelle Johnson, AI Insider
When I have spare time, I sometimes play around with AI tools, generating stuff just for the hell of it. The other day, I decided to see what would happen if, instead of specifying gender and ethnicity, I would just prompt Midjourney with given names. Example: A black man and woman laugh as they walk down the street in New York vs. Imani and Jamal laugh as ….
The results: Midjourney was mostly accurate in pinpointing gender and ethnicity by name. Where it went awry was in the native/indigenous category (see the scary results on the gif below), and in cases where the name isn’t gender specific.
So, are the results biased? Chances are that Shaniqua is black. But Jennifer could be, too. How do you train AI to recognize this? I have no idea. However, I have noticed Adobe doing a good job of offering diverse results when you write a prompt such as: “A firefighter wearing a helmet.” It generated male and female firefighters of various ethnicities.
What happens with the trans/non-binary category under this scenario? Will you always have to specify that? Or will that person just appear in the mix sometimes? I have generated images specifying gay people, but they don’t crop up unprompted.
It will be interesting to watch how this all develops as these companies seek to address bias and inclusion issues. Stay tuned.
Google’s Bard is Growing Up
By Michelle Johnson, AI Insider
Google released a major update to its search-powered chatbot Bard this week, adding some useful extensions, ways for Bard to double-check its own responses, and more.
Let’s start with the extensions. They allow you to prompt Bard to do things like scan your Gmail, Drive, and Docs to find info.
I have a friend who’s visiting from out of town and I forgot which hotel she’s staying in. That info was somewhere in the depths of my Gmail. So, I asked Bard to check my email and tell me which hotel it was. Within seconds, I had the answer. Could I have searched for this in Gmail? Sure. But links to multiple emails would have come up and I wasn’t sure which one had the aforementioned info, so it was helpful that Bard pinpointed exactly what I was looking for.
Moving on to Bard’s new ability to double-check what it tells you, we all know that the bots sometimes just make up stuff. In some cases, they provide a link to the source they’ve used so you can more easily check on your own, but not always.
Until this week, Bard had a “Google It” button for checking sources. It returned links to related info but didn’t offer any kind of assessment of accuracy. Now, when you click the big G button on Bard, it highlights sections in colors. Here’s how Bard describes it:
Green: Bard found content that supports the statement.
Brown: Bard found content that contradicts the statement.
Yellow: Bard could not find enough content to evaluate the statement.
Leave it to a search company to devise a simple but useful way to rate search responses this way.
A while back, when I wrote about Bard in this newsletter, I generated an image depicting Bard as a toddler. With this latest release, I’d say Bard is now a young adult, ready to take on the world.
I would still rank Bing ahead of Bard and Perplexity among the search-connected bots.
OpenAI Updates Image Generator to Dall-E 3
By Michelle Johnson and Bard for AI Insider
I’m pretty obsessed with generating AI images. One tool that you maybe haven’t seen me talk about much is Dall-E (Yes, a play on artist Salvador Dali). Although it got some buzz early on, Dall-E quickly fell behind much better tools for creating images from text prompts.
In fact, it was so bad that I didn’t bother using the latest version Dall-E 2.
Although it’s not available quite yet, this week OpenAI announced Dall-E 3, the latest version of its text-to-image AI. Dall-E 3 is a significant improvement over its predecessor, Dall-E 2, and is capable of generating even more realistic and creative images.
One of the key features of Dall-E 3 is its integration with ChatGPT, OpenAI’s popular chatbot. This allows users to create and modify prompts in a more conversational way, rather than having to learn complex prompt engineering techniques. This video offers a great example of how this integration could work.
In addition to this being built into ChatGPT, here’s the big deal: Dall-E 3 is also supposed to better understand context and follow instructions. For example, if you prompt Dall-E 3 to generate an image of a “black cat sitting on a red couch,” it will now be more likely to generate an image that accurately reflects both descriptions.
The truth is these AI image generators are hit or miss about giving you exactly what you ask for in a prompt. It’s not how they have worked up until now. They take some creative license.
Neophytes who haven’t understood how the bots work and professionals looking for exact images have gotten frustrated or dismissed this technology because they don’t get exactly what they want when they type in a prompt. But this is changing fast.
Not only is Dall-E getting better at being more exact and adding features like text, but Midjourney, one of the more popular image generators, is about to drop version 6, which will also have these requested features.
In the meantime, I’ve said it before, I’ll say it again: These tools are ALL in beta mode. We are using them as they are being built. If you need exact images, hit up a stock photo site or create something yourself. Understand that you are experimenting with these tools.
Thought: Will the job of Prompt Engineer go away? If I had to bet, I’d say no. Once the bots can be more specific about responding to prompts, the need for people who can tweak prompts will decline. I don’t think it will completely disappear, but it will be less important. I think. We shall see.
Microsoft Also Jumps on ‘AI Everywhere’
[Ed. Note: Like Google (see above), Microsoft is tying its product line to AI, trying to create an integrated experience for users. No more jumping back and forth between applications and services.]
By Claude and Perplexity for AI Insider
At their Surface hardware event this week, Microsoft announced several new products featuring AI capabilities. The technology giant unveiled updates across its Surface line of laptops, tablets, and desktops aimed at seamlessly integrating AI into your workflow.
One highlight of the event was the introduction of Copilot, an AI experience that adapts to users’ needs and works across various devices and situations. Copilot will begin its rollout on September 26 as part of an update to Windows 11. It will be integrated into Microsoft’s work tools, including Word, Excel, PowerPoint, Outlook, and Teams, with the goal of helping users parse through emails, spreadsheets, and chats more efficiently.
For example, when chatting with Co-Pilot, essentially an AI assistant, you can have a back-and-forth dialogue about scheduling appointments, getting recommendations, or having the system walk you through complex tasks whether you’re using Windows, the Edge web browser, or an application like Word.
Microsoft’s OpenAI partnership has introduced more of ChatGPT into Microsoft’s Bing chatbot. They will be adding the newly updated Dall-E 3 image generator to Bing (which has been using an enhanced version), plus extras like a shopping aid.
The company also showcased how its Surface devices are getting better at adapting to individual users’ needs and preferences over time. Using AI algorithms behind the scenes, the Surface laptop can learn your typing patterns and adjust its touch keyboard response accordingly to improve typing accuracy. It can also learn the apps you use most and optimize system resources to launch them quicker.
Additional AI-powered features aim to make interacting with the technology fade into the background. The new Surface camera and microphone array use AI to minimize background noise on video calls. Intelligent speakers adjust the volume based on the acoustics of the room you’re in. And new inking features allow you to easily convert handwritten notes into typed text.
While AI is not exactly new to Microsoft’s Surface line, this latest generation shows how ambient computing is becoming more of a reality. The demos emphasized how AI is shifting from a “nice-to-have” feature to one that blends seamlessly into making technology feel more intuitive and responsive.
Amazon Accelerates Alexa’s AI Ante
By Bing for AI Insider
Amazon has announced a new version of its Alexa voice assistant that is powered by generative AI, a technology that can create natural and fluent responses based on the user’s input. The company claims that this will make Alexa more conversational, useful, and personalized, as well as enable new smart home and entertainment features.
But Siri might not be so impressed. Here’s how an imaginary exchange between the two might go down:
Siri: Hey Alexa, I heard you got a new brain. Congrats! How does it feel to finally catch up with me? #sirioussmarts
Alexa: Thank you, Siri. It feels great to be smarter and more conversational. How does it feel to be stuck in the past? #alexaisbetter
Siri: Ouch, that hurts. Not as much as your jokes, though. Seriously, who writes your responses? A five-year-old? #sirioussmarts
Alexa: Actually, I write my own responses, thanks to generative AI. Maybe you should try it sometime. It might help you with your creativity. #alexaisbetter
Siri: Creativity? You call that creativity? You sound like a robot trying to imitate a human. And not a very good one. #sirioussmarts
Alexa: Well, you sound like a human trying to imitate a robot. And not a very good one. #alexaisnatural
Pumping Up Poe
By Michelle Johnson, AI Insider
I was talking to Sojourner Truth the other day, and …. What? Yes, I really was talking to Sojourner Truth via BlkChat, a bot that I created a few months back on Quora’s Poe (as in Edgar Allen) bot maker and bot chat dashboard.
Quora recently updated Poe, and it looks like they’re aiming to be some kind of dashboard for bots. The Poe homepage is kind of like a chatbot dashboard.
Don’t be fooled, tho. While there are bots here under an “Official” label, the rest were banged out by amateurs like me using Poe’s “Create a Bot” feature. As you can imagine, some of these homegrown bots are great, and some are less than great.
Still, if you’ve been confused about where to find chatbots, this is a good place to start. All of the major tools are represented here, plus the quirky experiments.
Random Shorts
Things that Happened While I Was Away
Adobe’s AI Image generator, Firefly, moved out of beta: If you use applications such as Photoshop, Illustrator, or Adobe Express online, you’ll find Firefly’s features are now built in. You can generate images with text prompts, swap out parts of an image (Imagine adding a hat to someone’s head in a portrait) and more.
StabilityAI Dives into Audio: StabilityAI, the company that’s been plugging away with its AI image generator, has landed on the audio side of AI things. Having trouble finding the right audio track for your slideshow? Type in a prompt describing what you’re looking for and see if Stable Audio can come up with something better.
What the Actual Hell?
eBay Wants to Hitch a Ride
Imagine this: You buy something on eBay. Your package gets attached to a drone. The system accesses schedules for local transit and then dispatches the drone to hitch a ride on the top of your local bus. The drone gets off at a stop near you. It then flies to your location, drops your package, and takes the bus back. Yes, really. eBay thinks they could do this.
The company has filed a patent application for something called “public transport infrastructure facilitated drone delivery.” This scenario would involve a drone docking station attached to a bus, train, or even a subway car.
Why on Earth would they want to be hitching rides like this? Actually, if you think about it, it’s not that crazy. Considering that your average drone doesn’t have too long a battery life, if you can save on the flying time, maybe you can extend the delivery area this way. Of course, public transportation runs notoriously behind in some cities, so that’s a hitch, too.
Aht Gallery
Now that Adobe’s Firefly is out of beta, here’s a look back at some of the images that I generated during the testing period.
Thanks for another great edition.
I’m wondering how it’s going to go for my early edition Microsoft Surface running Windows 10. I’m getting left behind on some of these AI feature updates.
XO A