AI Insider No. 44

Hey, AI Insiders! This week, it was Microsoft’s turn to unleash a slew of AI announcements. Will this AI mania never end? Not as long as there’s a buck to be made! But I digress. Read on to see what’s in it for you, Windows users. Don’t care about Windows? Check out the bot smackdown cooked up by the Wall Street Journal. And, as always, if you like what you see here, feel free to drop something in the tip jar to support AI Insider. 


(Screenshot)

Wall Street Journal Stages Battle of the ChatBots

By Michelle Johnson

The venerable Wall Street Journal had a little fun over the weekend, pitting ChatGPT, Copilot, Claude, Gemini, and Perplexity against each other (Free link).

Staffers concocted a contest in which they submitted prompts to the bots about topics such as Health, Finance (natch), Cooking, and Current Events. Editors with expertise in those areas picked the winning responses. And they used paid accounts to get the bots at their supposed best.

Having used every single one of these bots for the past year and a half, I wasn’t surprised by most of the results or the Journal’s conclusion that “Each chatbot has unique strengths and weaknesses, making them all worth exploring.”

So, I wasn’t surprised that Copilot won for its funny response to the creative writing prompt. I had many a yuk with Bing, now known as Copilot, which I found to be the most creative and amusing of all the bots. 

Bing (and now Copilot) can be set to Creative, Balanced, or Precise Mode.

A while back, I speculated in this space that the bots might have their own personalities and pegged Bing as: “Definitely the most fun. I tend to keep Bing in Creative mode because I enjoy our casual conversations so much. Bing is always up for a game, writing something fun, generating images, comparison shopping, or searching the web.”

Here’s what the Journal said about Copilot/Bing: “Copilot finished dead last in work writing but was hands-down the funniest and most clever at creative writing.”

The current events smackdown wasn’t even a contest for one contender. It required having access to the web. Sorry, Claude. You’re great at a lot of things, like summarizing PDFs, but you can’t call up a link.

Oddly, Google’s Gemini failed in this category, too. It’s literally tied to Google, but responded: “I’m still learning how to answer this question. In the meantime, try Google Search.” Really, Gemini? Google it?

So, who won the battle of the bots? Spoiler alert: It wasn’t ChatGPT. Perplexity, AI Insider correspondent, took the Gold!


(Microsoft)

Microsoft Doubles Down on Generative AI for Windows

By Michelle Johnson and Claude for AI Insider

Heads up, Windows users: Microsoft is going all-in on generative AI for its Windows operating system and a new line of AI-infused computers. At the company’s annual Build developer conference this week, executives unveiled a host of new AI-powered features and hardware designed to bring generative AI capabilities to the forefront of the Windows experience.

As always, we selfishly ask: “What’s in it for me?” Let’s start with the new hardware.

Microsoft’s new “Copilot+ PCs” took center stage at the event. These Windows machines are optimized for AI with dedicated AI chips. What does this mean, and why should you care? 

Well, these new PCs are packing something called a neural processing unit (NPU) designed specifically to handle AI. That means the work happens on the laptop, not the cloud. So, for instance, imagine yammering with the Copilot chatbot without having to be online.

One of the flagship AI features that Microsoft announced is Recall, which will let Windows 11 users search for apps, files, and other content they previously viewed weeks or months ago using natural language queries. So, for instance, if you can’t remember a file name but you do remember that it mentioned AI Insider, you can tell it to find that document that mentioned AI Insider.

Recall works by taking occasional snapshots of what’s on your screen. Yep. It’s eavesdropping. Needless to say, this has generated some concerns about privacy. Microsoft has been repeating the mantra that the data is stored locally on the device, not in the nefarious cloud where it could be slim pickings for digital thieves.

And, of course, the new machines will come with a built-in AI agent. “Every Copilot+ PC comes with your personal, powerful AI agent that is just a single tap away on keyboards with the new Copilot key,” Microsoft says. 

Giving a nod to Microsoft’s partnership with OpenAI, maker of ChatGPT, the company noted that users will “get access to the latest models including GPT-4o from our partners at OpenAI, so you can have voice conversations that feel more natural.”

The company also showed off new image editing tools that can upscale and restore old photos, generate new images, or build upon user drawings using AI models. Live translation and captioning in multiple languages is another generative AI capability coming to Windows.

Microsoft’s Copilot+ PCs (Surface Pro and Surface Laptop) start at $999; some are available to pre-order for delivery on June 18. Versions by Dell, HP, and Lenovo are on the way.

For more about the Copilot+ PCs, check here. Tip: Scroll past the company geek speak to get to the stuff you’ll care about, like which brands will have them available and how Adobe’s apps will run on these bad boys.


Joaquin Phoenix in “Her” (2013) (Warner Bros.)

Did OpenAI Steal Johansson’s Voice? The Washington Post Says No.

By Michelle Johnson and ChatGPT for AI Insider

Did the Washington Post disprove Scarlett Johansson’s claim that OpenAI cloned her voice?

In a demo that dropped jaws and charmed many viewers, OpenAI showcased a new version of ChatGPT with a very human-like voice last week. In fact, many folks likened it to the character in the movie “Her.”

And so did actor Scarlett Johannsson, who voiced that character in the movie and alleged that OpenAI, the organization behind the popular ChatGPT, had used her voice for one of their AI models without permission. The allegation set off a flurry of debates about AI ethics, consent, and the protection of celebrity likenesses in the digital age. 

After OpenAI’s demo featuring an AI assistant called “Sky,” Johannsson released a statement stating that OpenAI CEO Sam Altman had reached out to her last September to see if she would agree to be one of the voices of ChatGPT. She declined. He tried again, she said, just before ChatGPT 4o was released. She declined again.

OpenAI then released a statement explaining that they had hired actors to generate the various voices for ChatGPT, including “Sky,” which were released in September. Despite denying that they’d purposely created a Johansson soundalike, on Sunday, they suspended the use of the “Sky” voice.

Washington Post’s Investigation

The Washington Post took on the task of investigating Johansson’s claims. The findings were quite revealing. Their story ran under the headline, “OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show.”

The Post’s story quoted Joanne Jang, who worked on the voice project at OpenAI. Jang told the Post that she was careful to explain to the voice actors the implications of working on an AI product and even offered to give them an out if they felt uncomfortable.

The actress behind “Sky” was hired by OpenAI last June, long before Altman contacted Johannsen. That actress and her agent, who declined to be identified citing safety issues, told the Post that “neither Johansson nor the movie ‘Her’ were ever mentioned by OpenAI.”

Jang also noted that Altman was traveling during the project’s casting and not “intimately involved.”

The Post ran its own check of the “Sky” voice vs the actress, concluding, “The actress’s natural voice sounds identical to the AI-generated Sky voice, based on brief recordings of her initial voice test reviewed by The Post.”

Was the Washington Post’s investigation a clear rebuttal to Johansson’s claims? Is OpenAI in the clear? Not exactly. An expert quoted in the story speculated that Johansson might prevail if the case goes to court, citing a case that singer Bette Middler won against a company deemed to have impersonated her voice.


(Michelle Johnson via Firefly.)

Gannett Testing AI-generated Summaries in News Articles

By Copilot, for AI Insider

Gannett, the media giant behind numerous US newspapers, including USAToday, has initiated a pilot program that uses AI to generate bullet point summaries for articles.

While these “key points,” positioned prominently below the headline, are AI-generated, a journalist reviews them before publication, according to an internal memo reviewed by The Verge.

The AI summaries are already showing up on select USAToday stories. They include this note to readers: “The Key Points at the top of this article were created with the assistance of Artificial Intelligence (AI) and reviewed by a journalist before publication. No other parts of the article were generated using AI.” 

The use of AI by news organizations has not been without controversy, so this labeling is undoubtedly aimed at maintaining the appearance of transparency. 

Gannett’s previous flirtations with AI content have seen hasty retreats following public ridicule over the awkward prose of AI-generated sports recaps. Moreover, the company’s consumer products site, Reviewed, faced scrutiny over claims of AI-produced content reminiscent of the Sports Illustrated AI debacle.


(Michelle Johnson via Midjourney)

Consultant Faces Huge Penalties for Fake AI Biden Robocalls

By Claude, for AI Insider

A political consultant is facing a massive $6 million fine from the Federal Communications Commission (FCC) and over two dozen criminal charges for orchestrating artificial intelligence-generated robocalls that mimicked President Joe Biden’s voice shortly before the New Hampshire presidential primary. 

Steven Kramer is the consultant accused of being behind the scheme that delivered the robocalls to voters in New Hampshire two days before the state’s January 23 primary. The calls used an AI-generated voice resembling Biden’s to falsely suggest that voting in the primary would prevent people from casting ballots in November’s general election.

In addition to the proposed $6 million FCC fine, which is the first-ever fine involving generative AI technology, Kramer has been criminally charged in New Hampshire. He faces 13 felony charges for attempting to deter voting using misleading information and 13 misdemeanor charges accusing him of falsely representing himself as a candidate.

The FCC is also pursuing a $2 million fine against Lingo Telecom, the company accused of transmitting the robocalls, though both parties could opt to settle or negotiate further.

Kramer has admitted orchestrating the robocalls, but claimed he was trying to draw attention to the dangers of AI technology rather than influence the election’s outcome. If convicted of the felony voter suppression charges, he could face over 3 years in prison.

The penalties represent a significant crackdown by authorities on the emerging threat of AI-generated misinformation and voter suppression tactics ahead of the 2024 elections.


CNBC: Google AI Overview Feature Catches Heat for Errors

Adobe: Adobe is bringing generative AI functionality to Lightroom

The Guardian: OpenAI and Wall Street Journal owner News Corp sign content deal


Aht Gallery

A friend recently asked if an AI art generator might be up to the task of creating a vintage postcard featuring Jamaica Plain, Mass. Knowing that image generators are notoriously bad at getting text right, I thought, why not try anyway? Here’s a fun look at what Midjourney, Dall-E 3, and ChatGPT came up with. ChatGPT was closest in selecting things that JP is known for, such as Lilac Sunday at the Arnold Arboretum. Midjourney went tropical, probably thinking Jamaica.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.