1 big thing: FTC's Amazon case is no slam dunk

Animated illustration of Amazon's logo turning into a question mark.

2. Writers' strike deal sets historic new AI rules

3. Net neutrality: The never ending story

Illustration of the FCC logo tangled in ethernet cable wires. 

4. Training data

5. + This

Ina here, coming to you from Code Conference in Dana Point, California. Today’s AI+ is 1,283 words, a 5-minute read.

Thanks to Scott Rosenberg and Megan Morrone for editing and Bryan McBournie for copy editing this newsletter.

Editor's note: The second item in yesterday's newsletter was corrected to note that Mignon Clyburn is a former policymaker, not a former member of Congress.

A floating, translucent blue 3D render of the human brain.

deepBlue

Taking you inside the AI revolution, and delivering scoops and insights on the technologies reshaping our lives.

Axios AI+

Standard

1 big thing: How AI can put mistakes into overdrive

Walmart experiments with generative AI to help people shop

2. Walmart will use AI to aid shopping "missions"

3. Training data

A picture of Silicon Valley at night, including Apple Park, taken by plane.

Ina here, wishing everyone a happy National Coffee Day. Feel free to have a cup now, in fact. Today's AI+ is 1,218 words, a 5-minute read.

1 big thing: Amazon's billion-dollar bet on Anthropic

Animated illustration of a neon sign with the Amazon smile growing larger.

2. Prompt: New ChatGPT tricks with images, speech

Screenshot of a ChatGPT conversation asking the chatbot to estimate the number of Lego pieces in a photo

3. New Getty AI image tool aims to be lawsuit-safe

Ina here — well, sort of. I wrote this before the Yom Kippur holiday began. I'm now off atoning. Today's AI+ is 1,306 words, a 5-minute read.

Situational awareness: The tentative deal struck by the writers' union with Hollywood producers to end their strike includes provisions governing the use of AI.

1 big thing: Lessons from the AI whirlwind

2. Meta debuts new AI assistant

3. States fill AI legislation void left by Congress

Hi, it's Ryan. Today's AI+ is 1,294 words, a 5-minute read.

1 big thing: UAE aiming to disrupt AI superpowers

2. Exclusive: Streamers launch official trade group

3. Training data 

4. + This

Hi, it's Ryan, en route to San Francisco for Axios' latest AI Expert Voices roundtable. Today's AI+ is 1,170 words, a 4-minute read.

Situational awareness: The union representing voice actors in video games has voted to authorize a strike.

National

Former Twitter trust and safety head Yoel Roth accused Elon Musk at a conference Wednesday of ruling by dictate, rather than policy, and decimating the reputation of the platform known as X.

Driving the news: Roth made the comments at the Code Conference in Laguna Niguel, Calif., where X CEO Linda Yaccarino described an ambitious, growing company that will be profitable by next year. 

The big picture: There is no question that the company has radically changed under Musk. It has slashed staff, cut other costs, reinstated banned accounts and become a welcoming forum for content that would previously have been banned as hate speech.

Yaccarino insisted that advertisers are returning to X, that the company is now just about breakeven from a cash flow perspective and should be profitable by early 2024.

"The velocity of change, the scope of the ambition at X really does not exist anywhere else," said Yaccarino, who took the Code Conference stage less than two hours after Roth spoke with Kara Swisher. "Forget the other platforms, at any other company on Earth, there is no analog for the book that is being written."

Between the lines: Things quickly got testy when moderator and CNBC reporter Julia Boorstin pressed Yaccarino on how much power she really had given Musk still runs product. Asked if she was consulted before Musk tweeted about wanting to charge all users a monthly fee, Yacarino said: "We talk about everything."

As for Musk's involvement, Yaccarino said: "Who wouldn't want Elon Musk sitting by them running product?"

Yaccarino noted she did have a productive meeting with the Anti-Defamataion League shortly before Musk threatened to sue the civil rights group. "I wish that would be different," she said. "We're looking into that." She said the ADL needed to acknowledge the progress that X has made on antisemitism.

As for whether Musk's tweets were making her job harder, Yaccarino replied that "everyone deservers to have the opportunity to speak their opinion, no matter who they are, including Elon."

The other side: Roth described his experience with Musk, which included being praised for a short time, before being smeared as a pedophile in attacks that ultimately forced him to sell his home and move.

While Twitter was "flawed" before Musk arrived, Roth said there were aspirational policies in place that the company worked hard to live up to. Within 24 hours, Roth said, Musk was ruling by fiat and hate speech quickly increased.

"By any measure it is worse," Roth said. "Except by Twitter's measures."

Go deeper: Watchdog accuses X again of not moderating hate speech

Editor's note: This article has been corrected to reflect that X CEO Linda Yaccarino said that the company is just about breakeven from a cash flow perspective, not that it is now cash flow positive.

Yaccarino defends X over hate speech rise accusations

Former Twitter executive Yoel Roth said that "by any measure," the hate speech is worse — "except by Twitter's measures."

Chairman of Global Advertising and Partnerships for NBCUniversal Linda Yaccarino attends the 80th Annual Golden Globe Awards at The Beverly Hilton on January 10, 2023 in Beverly Hills, California.

Twitter CEO Linda Yaccarino in Beverly Hills, California, earlier this year. Photo: Kevin Winter/GA/The Hollywood Reporter via Getty Images

Twitter

Former Twitter trust and safety head <a class="gtmContentClick" data-vars-link-text="Yoel Roth" data-vars-click-url="https://www.axios.com/2022/11/10/twitter-loses-two-more-top-executives" data-vars-content-id="08649f91-a9b7-4296-9ed4-3c2f0ec14f2b" data-vars-headline="Yaccarino defends X over hate speech rise accusations" data-vars-event-category="story" data-vars-sub-category="story" data-vars-item="in_content_link" href="https://www.axios.com/2022/11/10/twitter-loses-two-more-top-executives" target="_self">Yoel Roth</a> accused <a class="gtmContentClick" data-vars-link-text="Elon Musk" data-vars-click-url="https://www.axios.com/economy-business/elon-musk" data-vars-content-id="08649f91-a9b7-4296-9ed4-3c2f0ec14f2b" data-vars-headline="Yaccarino defends X over hate speech rise accusations" data-vars-event-category="story" data-vars-sub-category="story" data-vars-item="in_content_link" href="https://www.axios.com/economy-business/elon-musk" target="_self">Elon Musk</a> at a conference Wednesday of ruling by dictate, rather than policy, and decimating the reputation of the platform known as X.Driving the news: Roth made the comments at the Code Conference in Laguna Niguel, Calif., where X CEO Linda Yaccarino described an ambitious, growing company that will be profitable by next year.

From the Valley to D.C., Big Tech players like Facebook, Google and Amazon are under more scrutiny than ever as new technology develops and privacy and antitrust concerns grow in lockstep with companies’ ambitions.

Technology

After a record economic expansion, the coronavirus pandemic sent the U.S. economy into a recession as unemployment soared to staggering heights. The country now faces urgent questions about how much stimulus is needed for reeling consumers and businesses, and what a recovery might look like.

Economy & Business

ChatGPT is moving beyond text — letting some users include images as part of a query and give spoken rather than typed instructions to the chatbot, which can also speak its answers.

Why it matters: The new features give the chatbot more utility right now and point to a future where AI tools understand the world around them, not just the online data on which they've been trained.

The big picture: OpenAI said in the spring when it released GPT-4 that such "multimodal" support was part of the new model. But it's taken until now for the company to ensure that the new input methods didn't make it easier to bypass the company's safety policies.

How it works: The new features will be available to ChatGPT Plus and enterprise customers over the next two weeks. Once the new features are enabled in the ChatGPT app, users can add an image to any text prompt, either by taking a new picture or uploading from a photo library.

To use audio in ChatGPT, you press a headphone button to speak a prompt and hear a response.

Our thought bubble: Just switching from typing to voice and hearing a response isn't super interesting in today's ChatGPT app and website, but could be a much bigger deal if, say, ChatGPT were built into a speaker or car system.

For my testing, I tried to think of a few cases where having a photo would help with a query.

Tight squeeze: A friend had posted a photo on Facebook wondering if all her storage boxes would fit in a 2001 Honda Odyssey. I know just who to ask, I told her, downloading her photo and then including it in a query to ChatGPT. The chatbot estimated the size of her contents and looked up the cargo space in the Honda minivan and reported that her boxes should all fit, though it could be a tight squeeze.

What to cook: ChatGPT offered only generic advice when I asked what I could make with the contents of our refrigerator. That's one of the touted uses for the new feature, though our fridge is admittedly packed  — and if I can't readily see what's hiding in there neither can ChatGPT, which has to work off a single image.

Math help: I took a page of fifth-grade multiplication problems and asked OpenAI to solve them. The chatbot had no problem returning the answers. While OpenAI notes that ChatGPT can help break down complex math questions or provide strategies, my testing showed it was also willing to just offer up the answers.

What's next: These new features are part of a pathway to allowing ChatGPT and other AI models to start incorporating more robust multimodal support.

OpenAI said last week that ChatGPT will also be able to generate images soon, thanks to integration with DALL-E 3.

OpenAI is also working directly with a handful of customers, such as Spotify, which plans to use the text-to-speech capabilities to allow podcasters to translate their work into other languages in their own voice.

Yes, but: OpenAI's safeguards are about to face the challenge of millions of users trying to break them. We'll see how well they prevent misuses, such as the creation of explicit content or using an image to get the tool to answer a question it wouldn't with text.

The bottom line: The additions of voice and image search are nice-to-have features in today's chatbot, but critical to a more capable future in which AI systems understand not just the abstract world, but also our immediate surroundings.

We tested ChatGPT's new support for images and voice search

We tested the new features with a 5th grader's math homework and a photo of the contents of a fridge.

OpenAI

Prompt

Automation and AI

ChatGPT is moving beyond text — letting some users include images as part of a query and give spoken rather than typed instructions to the chatbot, which can also speak its answers.Why it matters: The new features give the chatbot more utility right now and point to a future where AI tools understand the world around them, not just the online data on which they&#x27;ve been trained.

A year ago, the last time the Code Conference gathered a crowd of Silicon Valley luminaries in Southern California, AI was just one item on a long agenda of tech issues, and TikTok dominated the discussions.

Yes, but: The year since brought ChatGPT, an AI boom, and a slew of controversies that the industry is only beginning to address. 

What's happening: The pace of AI change over the past year has been incredibly rapid.

Pixelated photos with seven-fingered hands have given way to photorealistic images you can control with precision.

Chatbots that once knew nothing beyond mid-2021 are augmenting their web-based training with input from voice, images and current events.

Any company that can plaster an "AI" label on their products is doing so, along with some who probably shouldn't. 

The conversations at Code offered three takeaways from this accelerating whirlwind.

1. No one really knows how the legal issues will shake out.

Almost everyone at the conference agreed there are thorny issues around intellectual property and AI. Most believed that courts will have to settle the key issues surrounding AI models trained on large pools of data that contained considerable copyrighted material.

In general, owners of the words and photos used to train generative AI systems argue they need to give consent and be paid for the use of their content. Many AI companies are betting that such training is protected fair use under copyright law.

While the courts are seen as the most likely place for these questions to be answered with certainty, there are also ways for the players to sort out issues themselves. 

The music industry, for example, has often found ways to share revenue among multiple contributors and could lead once again if it finds a path forward in AI. 

Getty, which is suing the maker of Stable Diffusion, said it has had productive discussions with "the Microsofts and the Googles"  — but CEO Craig Peters added there was no guarantee those talks would lead to an agreement.

2. Human connection remains important, maybe even more important in an AI world.

HBO CEO Casey Bloys said that, though AI may assist writers, he doesn't foresee a world in which the entire creative process is performed by computers — nor does he want to live in such a world. 

Roblox CEO Dave Baszucki talked up his company's potential to bring dating to the metaverse, while Bumble CEO Whitney Wolfe Herd expressed hope that even if some encounters begin in a virtual world, they will quickly move to real-world relationships. 

Even Replika, known for creating bots that people want to have a relationship with, hopes its romantic AI product, Blush, serves as a training ground for real-life dating, not its replacement.

It's also really important to distinguish humans from machines in an AI world where either one could be at the other end of a conversation, says Georgetown University's Helen Toner, who spoke on an AI ethics panel at Code. Smart user interfaces can make these distinctions clearer, as can decisions over how much to humanize AI chatbots.

3. "Don't collect data you don't need" is one way to avoid violating users' privacy.

Bird Buddy, a bird feeder with a built-in AI camera, for example, throws out any images with a license plate or a person.

In building its generative AI tool, Getty focused on its creative collection rather than its editorial images, making sure that the model does not acquire information about public figures and brand names.

The result, said Peters, the Getty CEO, is that Getty's engine couldn't put the pope in a puffy Balenciaga jacket because it doesn't know who the pope is or what Balenciaga means.

Yes, but: AI's future remains unknown, because there's no way to know in advance what a new large language model or other large-scale AI system will do.

"I just want to say how crazy that is," Georgetown's Toner said.

thorny issues around intellectual property

Lessons from the AI whirlwind

AI's development keeps accelerating. Here are three lessons so far.

Rita Popova, Chief Product Officer, Replika & Blush speaks onstage during Vox Media's 2023 Code Conference.

Replica & Blush product chief Rita Popova speaks on stage during Vox Media's 2023 Code Conference. Photo: Jerod Harris/Getty Images for Vox Media

Artificial intelligence

A year ago, the last time the Code Conference gathered a crowd of Silicon Valley luminaries in Southern California, AI was just one item on a long agenda of tech issues, and <a class="gtmContentClick" data-vars-link-text="TikTok dominated the discussions." data-vars-click-url="https://www.axios.com/2022/09/09/tiktok-unites-tech-against-code-conference" data-vars-content-id="d10e80aa-d44d-4828-aec6-a91eeae4a1b2" data-vars-headline="Lessons from the AI whirlwind" data-vars-event-category="story" data-vars-sub-category="story" data-vars-item="in_content_link" href="https://www.axios.com/2022/09/09/tiktok-unites-tech-against-code-conference" target="_self">TikTok dominated the discussions.</a>Yes, but: The year since brought ChatGPT, an AI boom, and a slew of controversies that the industry is only beginning to address.

Getty Images debuted a generative AI tool Monday that, in contrast to rival text-to-image engines, was trained only on licensed content from Getty's own vast creative collection.

Why it matters: Getty has sued Stability AI for infringing on its copyrighted works, but says its own new image engine is safe for commercial use. Getty will also indemnify customers buying images, meaning that Getty is on the hook for any legal problems that arise from images created by the tool.

Details: The tool was trained only on Getty's creative images, not the editorial images and news photos that contain famous people, brands and other potentially protected content.

Getty partnered with Nvidia, which provided the processing power and technical expertise needed to train the model.

Getty says customers will pay for each image generated. Those whose images helped train the model will be compensated, Getty said, based on a formula that takes into account both the number of pieces a contributor provided and their popularity.

The big picture: Adobe has pursued a similar approach with Firefly and is also indemnifying customers.

The courts have yet to weigh in on whether the makers of AI software are responsible for compensating copyright-holders for material used to train their models. 

OpenAI faces multiple lawsuits from authors, and visual artists have objected to what they view as the appropriation of copyrighted material in large image databases scraped from the internet that have helped train large models.

What they're saying: Getty CEO Craig Peters told Axios that the new software aims to enable more people to create images they couldn't previously make with their lens, brush or software. But, he said, it's not designed to replace professional artists.

"They bring a level of authenticity, they bring the level of creativity, they bring a lot of things in, and this model is not intended to rip that off," he said. "It's intended to sit side by side and enable creators to do something that they maybe couldn't do."

Between the lines: Getty's tool won't generate a Nike sneaker because it doesn't know what Nike is, but it will draw a sneaker, Peters said. It likewise doesn't know famous people, nor can it draw an image in the style of a particular artist.

Peters said the goal is to give customers something they can trust is on safe legal ground and that rewards those whose content is used to train the model.

As for the legal uncertainty surrounding other engines trained on copyrighted content, Peters said he is eager to have the courts weigh in.

"We want to get to clarity," he said. "We're on the right side of this argument and we think the facts are strong in our favor."

Go deeper: AI-generated images open multiple cans of worms

AI-generated images open multiple cans of worms.

New Getty AI image maker aims to be lawsuit-safe

The giant of the visual-images market trained its tool on images for which it owns the rights.

Screen shot of Getty Images' AI image generator making a cotton-candy cabin

Getty Images debuted a generative AI tool Monday that, in contrast to rival text-to-image engines, was trained only on licensed content from Getty&#x27;s own vast creative collection.Why it matters: Getty has <a class="gtmContentClick" data-vars-link-text="sued Stability AI" data-vars-click-url="https://www.axios.com/2023/02/24/chatgpt-generative-ai-legal-minefield" data-vars-content-id="fc6e854e-38cc-4137-9ea5-86f507344bc0" data-vars-headline="New Getty AI image maker aims to be lawsuit-safe" data-vars-event-category="story" data-vars-sub-category="story" data-vars-item="in_content_link" href="https://www.axios.com/2023/02/24/chatgpt-generative-ai-legal-minefield" target="_self">sued Stability AI</a> for infringing on its copyrighted works, but says its own new image engine is safe for commercial use. Getty will also indemnify customers buying images, meaning that Getty is on the hook for any legal problems that arise from images created by the tool.

Last month Walmart announced a generative AI tool for office workers and now the company is showing off how the tech can help its shoppers.

Why it matters: With its size and resources, Walmart could use generative AI in whichever ways it wants. Its choices indicate where it sees the most benefit today.

Details: In a briefing with reporters in San Francisco on Thursday, Walmart showed how the technology can be useful, both to directly communicate with customers as well as behind the scenes, to help customers envision what products would look like in their homes or on their bodies.

The first, a shopping assistant, is designed to help customers who are looking to take on a project, such as planning a birthday party, decorating a home for the holidays or outfitting a new dorm room. The assistant, which Walmart hopes to begin testing in the coming weeks, can also help customers decide among products, such as choosing which cell phone would be best suited for a 10-year-old.

Generative AI also allows Walmart to create three-dimensional objects from still photos, removing any occlusions from, say an arm or hand covering part of an outfit. The results are clothes that can be placed on a virtual model or furniture and electronics to be realistically placed inside a scan of one's living room. 

Creating 3D objects using generative AI versus traditional modeling technologies can save hundreds of dollars per item and shortens the creation process from days or weeks to minutes.

Between the lines: Walmart officials were loath to say just whose AI technologies they are using, but mentioned that they are looking at all options and designing systems to be open to using whoever has the best and most cost-effective technology at a given point in time.

"We believe in being Switzerland," said Cheryl Ainoa, Walmart's executive vice president of global technology. 

Flashback: Walmart has been serving customers via chatbots for years, using old technologies from Google and Apple. The company started work on chatbots in 2019, began testing such services in 2020 and made them broadly available starting last year. 

Those systems, which include the ability to order products via text message, are geared more toward straightforward transactions.

Its using the technology to help shoppers on a mission.

Walmart

Last month Walmart announced a generative <a class="gtmContentClick" data-vars-link-text="AI tool" data-vars-click-url="https://www.axios.com/2023/08/30/walmart-generative-ai-app" data-vars-content-id="fe03e7f5-4165-4c26-880e-a37d0398140b" data-vars-headline="Walmart experiments with generative AI to help people shop" data-vars-event-category="story" data-vars-sub-category="story" data-vars-item="in_content_link" href="https://www.axios.com/2023/08/30/walmart-generative-ai-app" target="_self">AI tool</a> for office workers and now the company is showing off how the tech can help its shoppers.Why it matters: With its size and resources, Walmart could use generative AI in whichever ways it wants. Its choices indicate where it sees the most benefit today.

Ina Fried

Stories

Walmart experiments with generative AI to help people shop

Lessons from the AI whirlwind

Yaccarino defends X over hate speech rise accusations

New Getty AI image maker aims to be lawsuit-safe

We tested ChatGPT's new support for images and voice search

Newsletter editions

Mistakes were made

AI lessons learned

Amazon case's hurdles

Unlikely AI superpower

Amazon's billion-dollar bet