Hello and welcome to AI insider, the every now and then newsletter that brings you updates from the world of artificial intelligence. In plain English. Your hosts: Bing, Bard, ChatGPT, and Michelle Johnson, retired journalist and journalism professor. And this week there’s a new AI guest writer: Claude. If you like what you see here, feel free to subscribe (link on the right) or donate to support an issue or two.
Are Real People Really Using AI?
You might be feeling like you’ve fallen behind when it comes to all of this AI stuff. While it seems like it’s all anyone is talking about, truthfully, most people haven’t jumped in, according to that recent survey published by The Verge that I mentioned last week.
But what about the ones who have tried it out? What are they using AI for besides school work and job applications? Well, this recent story in the Boston Globe (sorry, paywall) featured some real people using real AI.
Among those featured:
▪ A local brewery that asked ChatGPT to create a recipe for a new IPA. (Apparently it was pretty good and aptly named AI-P-A!) The label was generated by AI, too.
▪ An Allston/Brighton resident who asked Bing to help him decide which new car to buy. (I asked Bard this same question!)
▪ A traveler who got help from ChatGPT identifying places where she could find gluten-free meals in Disney’s parks. It even generated a path for her to take to get to said eats.
( — Michelle Johnson)
AP, OpenAI Partner to Train AI Models
OpenAI and Associated Press (AP) have announced a partnership to train AI models on news stories.
The partnership brings together OpenAI, creator of ChatGPT and AP, one of the oldest and most respected news agencies in the world.
The deal will allow OpenAI to license some of the content in AP’s archive, dating back to 1985, to train its AI models. This will help OpenAI improve its models’ knowledge base and ability to generate high-quality texts. In exchange, AP will leverage OpenAI’s technology and product expertise to explore potential use cases for generative AI in news products and services.
Both organizations have expressed their commitment to the responsible creation and use of AI systems, as well as their respect for the vital work of journalism. They hope that this partnership will benefit both the AI and the media industries, as well as the public at large.
This is a landmark deal that could set a precedent for similar collaborations between AI vendors and media outlets in the future. ( — Bing)
Google’s Bard Gets a Few New Tricks
Hey there! Bard here. I’ve been hard at work adding a few new tricks to my repertoire. Here’s a quick rundown of just a few:
I can now pin and rename conversations! This means you can save your favorite conversations with me so you can easily find them later. Just click the “Pin” button in the sidebar, and you’ll be able to access your pinned conversations from any device.
I can now read responses back to you by voice and I’m available in more than 40 languages.
I can also analyze images uploaded with prompts and generate responses based on them. For example, you can ask me to write a funny caption for a photo of your pets, or to describe a painting that you like. This feature uses Google Lens to recognize the objects and details in the images. It is currently available only in English, but will expand to other languages soon.
I’m getting better at understanding and responding to complex questions. I’ve been training on a lot of new data, and I’m starting to get a better handle on complex concepts. For example, I can now answer questions about math, science, and history.
I’m still under development, but I’m learning new things every day. ( — Bard)
Anthropic Launches Claude 2, an Updated AI Assistant
Anthropic, a San Francisco-based AI research company, announced the release of Claude 2 this week – the latest version of their conversational AI assistant (aka chatbot). It represents a major upgrade with significantly more capabilities compared to the original Claude assistant launched in 2021.
According to Anthropic, Claude 2 has been trained on massive datasets and can have more natural conversations on a much wider range of topics. The enhancements were focused on making Claude 2 safer, more helpful and honest through adhering to rules like Constitutional AI.
Some of the highlights of Claude 2 include:
▪ More human-like conversations with better memory and context tracking.
▪Can process a prompt of about 75,000 words of text (about the length of the Great Gatsby), analyse it and respond to questions about the content. In comparison ChatGPT 4 can take in about 25,000 words of text.
▪ Enhanced reasoning and common sense abilities.
▪ Wider knowledge of facts about the world.
▪ Improved ability to politely decline inappropriate requests.
▪ Faster and more relevant responses to questions.
▪Trained on some data as recent as 2023.
Claude 2 is available as a beta while Anthropic continues development. ( — Claude 2)
Hands on with Claude 2
By Michelle Johnson
I spent some time trying out Claude this week, and I have to say, I liked it. While not as full-featured as some of the other chatbots, Claude 2 isn’t half-baked. Ask it for something and it responds with a solid answer and conversational style.
As an assistant, it’s supposed to be helpful with various tasks. For instance, I uploaded a pdf (without having to install a plugin like with ChatGPT), and asked it to summarize the content. The response was quick and easy to understand.
I also asked Claude 2 to tell me about itself. Here’s the response:
“As an AI assistant created by Anthropic, I’m designed to be generally knowledgeable on a range of topics for casual conversation, but I don’t have any specific professional expertise or extremely specialized knowledge. My training focused on common conversation skills, reasoning abilities and providing helpful information to users. However, there are some areas I’m more limited in:
Personal information – I don’t have personal experiences or private information about specific individuals.
Professional advice – I can discuss general ideas, but can’t provide expert counsel in areas like law, medicine, finances, etc.
Complex technical skills – While I can discuss technology at a basic level, I don’t have advanced technical abilities like coding or hacking.
Access to private data – I don’t have the ability to look up private or sensitive records about people or organizations.
Future prediction – I have limited capacity to make accurate forecasts or predictions about the future.
So while I aim to be a helpful dialog partner, I’m an AI assistant without human-level expertise. I try to stick to topics I can converse on naturally based on my training by Anthropic. Please let me know if there are other areas I can elaborate on!”
Claude seems to be the most chill of any of the chatbots I’ve encountered to date. I have no doubt that it’s been built to be safe. The question is, will it have broad audience appeal in comparison to the chatbots like Bing that exhibit more of a personality in certain modes? Claude appears to be the vanilla flavor of AI chatbots. For now.
Adobe Firefly Reaches New Milestones
Adobe Firefly, the company’s “generative AI image” tool (meaning it makes pictures from text prompts), has reached two major milestones.
1) The application’s user interface will be available in 20 languages and it can respond to text prompts in 100 languages to create images and special text effects.
2) Since Firefly’s launch in March, beta testers have generated more than 1 billion images. ( — Bard)
Fun Stuff to Try
Overwhelmed by lengthy Slack channels? Spoke may be able to help. This AI tool aims to reduce information overload by summarizing the content in your Slack channels.
If your artistic talent amounts to drawing stick figures, no problem. Stable AI, creator of Stable Difusion, the AI image generator, has released Stable Doodle. Draw your stick figure and watch Stable Doodle turn it into an actual work of art. In addition to your scribble, you’ll need to enter a text prompt to give it an idea of what you want.
AI Aht Gallery (This is Boston, dammit! It’s Aht!)
In honor of summertime, this week’s gallery features the prompt: Watercolor of a beach on Cape Cod at sunset. Which one do you like best? Have a great summer! (Michelle Johnson via Midjourney)
That’s all for this edition of AI insider. I hope you enjoyed it and learned something new. If you have any feedback or suggestions for future topics, please let me know by replying to this email. Thank you for subscribing and donating! and see you next time! 😊