And…we’re back! AI Insider was on hiatus last month as I embarked on a bucket list driving trip through the South to research my family’s roots. More about that in a future edition. For now, I’m back with the bots to catch you up on news in the AI universe. Let’s get to it!
New ChatGPT Memory Feature Rolling Out
By Copilot for AI Insider
ChatGPT is losing its amnesia. Paid Plus subscribers are getting access to a new feature that can remember what you’ve talked about in previous chats or what you specifically tell it to remember.
This memory feature aims to make future chats more helpful by saving you from having to repeat information. As you chat with ChatGPT, you can ask it to remember something specific or let it pick up details itself.
Over time, ChatGPT’s memory will improve based on your interactions.
For example:
You can ask it to remember your name. (This feature is also available in the Custom Instruction setting.)
If you mention a preference for meeting notes with headlines, bullets, and action items, ChatGPT will remember this and recap meetings accordingly.
Personal details, such as telling ChatGPT you have a toddler who loves jellyfish, can influence its suggestions (e.g., generate a jellyfish-themed birthday card).
When memories are updated, ChatGPT will notify you. You can review everything it has picked up from your conversations and manage unwanted memories.
You can ask it what it remembers, tell it to forget something, or entirely turn off memory.
Deleting a chat won’t erase its memories; you must delete the memory itself.
If you’re a Plus subscriber, go to Settings > Personalization > Memory to access and manage memories. You may not have it just yet, but it should appear soon.
Rabbit R1: A Novel Concept Arrives Half-Baked
By Perplexity and Michelle Johnson
[Editor’s Note: I have a Rabbit R1 in hand. I’ll do a full review soon.]
The Rabbit R1, a new entrant in the fledgling AI gadget market, started turning heads long before it shipped. Product shots of a square bright-orange handheld device featuring a small screen were eye-catching. Video of users audibly asking it to order food or a rideshare was intriguing. Thousands of pre-orders flooded in.
However, what showed up when the device started shipping at the end of April was a distinctive-looking unit with lots of missing or malfunctioning features.
Some puzzled reviewers panned it or simply refused to review a half-baked product. Others said, wait, it has promise. They just have to work out the kinks. And, besides, you shouldn’t expect much from a $200 plastic gadget. (The recently released Humane AI pin, a $700 wearable device, has been similarly panned as not ready for prime time.)
Reviewers noted Rabbi R1’s performance issues, such as inaccurate AI responses and a less-than-intuitive user interface. The device’s scroll-wheel-driven menu system, in particular, has been criticized for its lack of user-friendliness. (On the other hand, some folks love it!)
Moreover, the R1’s integration with services like Uber and DoorDash has been problematic, with features either not working as advertised or not working at all.
What exactly did the company, Rabbit Inc., promise? Well, in a video preview, company CEO, Jesse Lyu, touted something called a “Large Action Model,” (LAM) that wouldn’t just chat with you, but it would carry out actions FOR you, like ordering an Uber.
Sadly, some of the promises fell short.
A few days ago, the company pushed out an update that fixed some of the early issues and still promises great things to come. As it stands, the Rabbit R1 should be considered a work in progress.
And frankly, that’s true of most of what we’ve seen of AI. Companies release what amounts to beta versions, relying on us to be testers. That’s more palatable when the product is free. However, users get a little more persnickety when the credit cards come out.
While the R1’s innovative design and LAM concept have garnered interest, its execution and current functionality make it difficult to recommend just yet.
As Rabbit refines the R1, early adopters and potential future users will watch closely to see if this novel device can evolve to meet its ambitious goals. We have an R1 in hand and will review it when it’s more baked.
iOS App for Claude Drops
By Michelle Johnson
Anthropic, the maker of the AI chatbot Claude, announced this week that an app for it is now available in the Apple iOS store.
Until now, if you wanted to use Claude on your iPhone, you’d have to load it into your mobile web browser. Claude joins ChatGPT, Perplexity, and Google’s Gemini (via the Google app) in the iOS app store. (No word on Android yet.)
Anthropic also announced a new paid plan for teams. Claude for teams costs $30 a month per seat.
The new app works for free and paid plans. It includes nifty features such as “Vision,” which allows you to hold up your phone camera and ask Claude to tell you what it sees. It will also stay in sync, so there are no worries about not being able to see prompts/chats that you previously entered on other devices.
OpenAI Partners with Financial Times
By Perplexity for AI Insider
The Financial Times (FT) and OpenAI struck a deal that will allow OpenAI (ChatPT) to use Financial Times content to train its AI models. Details of the financial agreement were undisclosed.
The deal also includes a commitment from OpenAI to assist FT in developing cutting-edge AI models tailored for its readership.
The agreement underscores the evolving relationship between AI and journalism, which has mostly been contentious as news organizations bring lawsuits over copyright infringement.
With this deal, FT joins other news organizations, including the Associated Press, in agreeing to partner with OpenAI.
Under the terms of the deal, OpenAI will have access to a wealth of archived articles from the FT, enhancing the AI’s learning process with high-quality, factual reporting. This integration is expected to significantly improve the accuracy and reliability of OpenAI’s language models by grounding them in professionally sourced information.
Moreover, the partnership extends beyond mere data sharing. ChatGPT users will soon be able to access attributed summaries, direct quotes, and links to full articles from the FT. This feature aims to enrich user interactions on the platform, providing them with credible, well-sourced content.
The collaboration also paves the way for the development of new AI-driven tools tailored specifically for FT readers. These innovations are anticipated to transform how subscribers interact with content, potentially introducing more personalized and engaging ways to consume news.
While the specifics of the financial agreement remain undisclosed, the deal represents a step forward in addressing the ethical use of journalistic content in AI training and acknowledges the value of news organizations’ intellectual property.
Teacher Charged with Using AI to Fake Principal’s Voice
By Copilot for AI Insider
A Maryland high school teacher was recently arrested for allegedly using artificial intelligence to create a fake audio recording of a racist rant reportedly mimicking the voice of the school’s principal, Eric Eiswert.
Dazhon Darien, a physical education teacher and athletic director at Pikesville High School, was accused of falsifying Principal Eiswert’s voice in January. Authorities have now confirmed that the recording was not authentic and was, in fact, generated using AI technology. The charges against Darien include disrupting school activities, theft, retaliation against a witness, and stalking.
Darien and Eiswert had reportedly been at odds over work performance challenges, with Darien’s contract possibly not being renewed next semester.
The viral audio recording initially seemed to capture Principal Eiswert spewing hateful rhetoric, mocking Black and Jewish people.
Darien was apprehended at Baltimore/Washington International Thurgood Marshall Airport while attempting to fly to Houston. Security personnel questioned him about properly packing his gun for travel, which led to the discovery of an outstanding arrest warrant.
ICYMI: Stuff that Dropped While We Were on Hiatus
Adobe upgraded its AI image generator, Firefly, to version 3. Video here. And, BTW, standalone Firefly still offers a free version in addition to the paid upgrades.
Adobe also added new generative AI features (via Firefly) to its venerable photo editing application, Photoshop. You can now generate images directly in Photoshop. Well, in the beta, anyway.
Meta (Facebook) unveiled a shiny new AI image generator, Meta.ai. It’s actually pretty good. But does the world need another AI image generator?
CNBC: Eight newspaper publishers sue Microsoft and OpenAI over copyright infringement
Yahoo Finance: Wall Street is upbeat about AI stocks again as companies deliver on earnings.
KABC LA: LA is Installing AI-powered cameras on buses to help issue tickets to cars blocking bus stops.
What the Actual Hell?
Prompt: A newspaper editor working at a computer.
OK, let’s try for some gender diversity. Prompt: A female newspaper editor working at a computer.
What the actual hell? What year is this?
Aht Gallery
Ah, spring! This week’s gallery features scenes of folks getting outdoors.
