National

Axios AI+

OpenAI is offering an early look at DALL-E 3, the next version of its image generation tool.

Why it matters: OpenAI faces competition from open-source tools like Stable Diffusion and a host of tech companies large and small.

The update allows DALL-E 3 to be summoned and controlled using ChatGPT and aims to produce higher-quality images that more faithfully reflect queries.

Details: OpenAI says that DALL-E 3 is significantly better at understanding the intent of prompts, particularly longer ones, compared to DALL-E 2, which debuted in April 2022.

OpenAI says DALL-E 3 also does better — but not perfectly — in areas that have tripped up image generators, such as text and hands.

The ChatGPT integration will allow people to hone their request through conversations with the chatbot and receive the result directly within the chat app.

OpenAI plans to make DALL-E 3 available to ChatGPT+ and enterprise customers in October

DALL-E 3 will also be available sometime this fall in its public labs and for API customers.

Between the lines: OpenAI says it has also taken fresh steps to improve the safety of and reduce algorithmic bias within DALL-E. The company is also taking two actions seemingly designed to counter some of the criticisms leveled by artists regarding image generators.

DALL·E 3 is designed to reject requests that ask for an image in the style of living artists. 

Artists can also now opt out of having certain or all of their images used to train future OpenAI image generation models.

Separately, OpenAI this week announced it was working with a group of expert contractors to "red team" its products in search of bias and other issues.

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

AI-generated images open multiple cans of worms

Open AI debuts next version of its image generation tool

OpenAI claims its new image creator is better at understanding prompts.

A AI-generated image of a banana-shaped couch in a living room created using Dall-E 3. 

An AI-generated image of a banana-shaped couch in a living room created using Dall-E 3. Image: OpenAI

Artificial intelligence

Automation and AI

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

As newsrooms around the globe are beginning to harness AI, this week the New York Times posted a job listing for an editor to serve as "newsroom generative AI lead."

My thought bubble: Good luck to the winning applicant! Journalists, like many other professionals, are fearful and anxious about the new technology, and this is not going to be an easy job — but I'm generous, so here's a memo with some helpful advice. 

Dear future robot wrangler: Many publishers have jumped right in to use generative AI for writing entire articles — a task the technology isn't ready to take on. Far more interesting opportunities lie under the surface.

1. Writing full articles is the most obvious use of generative AI, but not the best, because of all the ways AI still fails at the basics of good journalism.

Today's generative AI systems make mistakes, plagiarize, say inane or embarrassing things and make stuff up. The Columbus Dispatch, MSN, Onion and Gizmodo parent G/O Media, CNET and others have all learned that the hard way with spectacular, very public flops.

Even if they get better at sticking to the facts, algorithms can't assess credibility. Today's technology often does a passable job summarizing an interview — but it has no idea if the interviewee is the foremost expert on a subject, or someone who is in over their head, or someone who's promoting an agenda.

There are some very limited writing tasks AI can be let loose on today, like corporate earnings stories, which follow a predictable pattern and often depend on a single source. Using older technology, the Los Angeles Times has for years had a bot churn out first takes on earthquakes, telling people what they most want to know — how big the tremor was and where it was centered. 

2. The bigger newsroom opportunity is to use AI for other parts of the reporting and writing process. 

I've been using Otter.ai to record and transcribe notes, and its AI chatbot lets me ask questions of my notes (see review). And I am constantly looking for other tasks it might assist with. I think AI will someday soon help me identify sources from my email, for example.

AI can also be used to find patterns in vast amounts of data — an organization's own archives as well as large swaths of public information. A reporter might only have time to read through a handful or dozens of speeches to detect shifts in a politician's stances over time; an AI bot could consume every available recording and transcript.

3. Illustrations and some kinds of conceptual graphics are another area of promise for AI. The ability of text-to-image engines to create powerful artwork from a simple prompt is already impressive. Engines can even be trained on an organization's style using past work. 

Yes, but: Visual journalists have raised strong questions on ethical and intellectual property-related grounds about many of today's generative AI image systems.

4. Experiment behind closed doors first. 

The best way to understand the tools is to try them. I get to do that as part of my job, but any newsroom that wants to be relevant in the AI era should be constantly exploring what's possible, scrapping what doesn't meet standards and investing in what does.

Whatever you decide to do with AI, be clear with the audience and, where possible, explain what you are doing and why. The L.A. Times does this very well with its Quakebot.

6. AI can help the news business — but it also threatens it.

Journalism has been slow to adapt to other recent tech shifts, and paid a heavy price, but that needn't be the case with AI. 

Some news organizations, such as Reuters, are using AI to compose and test headlines to see which perform better in our SEO-dominated world. Bloomberg, meanwhile, is building a comprehensive BloombergGPT system based on its vast collection of financial data. 

At the same time, AI is a phenomenal aggregator. That means the tech giants will have new, more powerful ways of incorporating content without having to direct people to the originating site. 

So, of course, there is the question the bosses have to answer — how to interact with the AI companies themselves. Some media companies are working on projects with AI companies. Some are blocking their content from being scraped. Some are licensing content to AI companies and some are suing or threatening legal action against AI companies.

7. AI will be a huge labor issue, whatever publishers do.

Like their entertainment industry counterparts, unions representing journalists have AI front and center on their radar, especially at companies like G/O Media that have already been aggressive in adopting the technology.

Journalists are highly skeptical that AI will just free up skilled reporters, editors and illustrators for bigger tasks. Most newsrooms have been getting smaller throughout my 20+-year career and especially in the past decade. It's hard to see how AI could reverse that.

8. Generative AI will make it even harder to combat online misinformation, as we've written extensively.

That creates a huge need for humans who can bring fairness and accuracy, as well as much needed context, to the news. But they will have to do so amid intensifying business pressures that the new technology is unleashing.  

A bad use of AI today might make sense tomorrow. I've been covering Silicon Valley for more than 20 years, and generative AI is evolving faster than any other technology I've witnessed.

Go deeper: I spoke on the topic of AI and newsrooms earlier this month at the conference for NLGJA: The Association of LGBTQ+ Journalists. You can view a transcript and replay here.

ethical and intellectual property-related grounds

building a comprehensive BloombergGPT system

The right newsroom jobs for AI, and the wrong ones

Writing articles gets a lot of publicity, often bad, but today's AI is better suited for other journalistic tasks.

Illustration of a paragraph of binary code with red squiggly spell check lines under some of it.

Users can now ask Google's Bard chatbot to double-check its answers and can connect it to their Google apps and services, in the first update to the product since July.

Why it matters: Google is racing to integrate AI into its main products, hoping to extend their dominance, after being caught out earlier this year when Microsoft integrated ChatGPT into its Bing search engine and added an AI "copilot" to Microsoft 365.

With 1 in 3 Americans describing themselves as "very concerned" about the development of AI — driven partly by the rise of chatbots that cannot explain themselves — citation and transparency improvements could help win over some AI skeptics.

What's happening: New and existing English-language Bard users will be prompted to decide if they want to add Bard Extensions when they use the chatbot.

The opt-in Extensions plug-in connects Bard to Google products such as Gmail, Docs, Drive, Maps, YouTube, and Google Flights. The service could, for example, summarize unread emails or construct a draft trip itinerary combining material from your Gmail, Docs and Google Flights. 

If a user clicks on a "Google it" feature button (the "G" icon) next to a Bard answer, Bard will check line by line "if there is content across the web to substantiate" its response, the company said in a statement, and provide that breakdown with applicable links.

Existing English language features — including image uploading and inclusion of search images in Bard responses — are expanding to more than 40 supported languages.

The intrigue: Google promised in a blog post that "your content from Gmail, Docs and Drive is not seen by human reviewers, used by Bard to show you ads or used to train the Bard model."

Tech companies have been accused of a data land grab as they grapple with how to train new and improved AI products. 

What they're saying: While walking Axios through Extensions, Jack Krawczyk, Google's product lead for Bard, said he found the new features most useful in helping him to parse back-to-school correspondence for his child and prompting him to "bring his curiosity to life" by searching in better ways.

Two types of user feedback drove today's product updates. 

"The average person around the country says, '[AI] is super cool, but I still haven't figured out how to make it useful in my life,'" Krawczyk said.

Many told Google they were concerned about chatbot "hallucinations" (making things up). 

Google's Bard launches fact-check features as AI race continues

The search engine chatbot will be able to double check its answers and plug into other Google apps.

Google

Pro Rata

Axios Pro Rata

Japan's SoftBank is once again flush with cash and liquid/leverageable assets, following last week's IPO for chip design firm Arm.

Why it matters: Get ready for a spending spree on AI startups.

What to know: SoftBank CEO Masa Son never rests on his laurels or spends much time salving old wounds.

He went from one of the world's richest men to a paper pauper following the dotcom boom and bust. Then, after making a savvy bet on Alibaba, Son came over the top with the $100 billion Vision Fund — arguably sparking the unicorn overflow that ended up eating itself and its investors.

SoftBank never raised a second Vision Fund, instead using its balance sheet for what it called Vision Fund II in press releases, and slowed its new investment pace over the past couple years.

Arm enables SoftBank to go back on offense, at a land grab moment for cash-hungry AI companies.

Behind the scenes: SoftBank's investment process for the past several years is said to have required that prospective portfolio companies have at least some AI component, no matter how tortured.

Yes, even for companies like Greensill (if you can find the investment committee memo, I'd almost guarantee there's an AI mention).

Now it's a lot easier. Not only because there's a plethora of legit AI companies, but also because most every non-AI company is either trying to incorporate AI or is pretending to.

There's even debate over whether Arm should refer to itself as an AI company, given that the vast majority of its designs are used in smartphones, but CEO Rene Haas told Axios that his main IPO roadshow message was that "AI runs on Arm."

Meanwhile: The Financial Times reports that SoftBank is considering an investment or other sort of partnership with OpenAI, the Microsoft-backed creator of ChatGPT.

The bottom line: SoftBank engenders more schadenfreude than any other venture capital or private equity firm on the planet, but almost no one is going to turn down its money.

An outside spokesperson for SoftBank declined comment.

SoftBank plans to ramp up AI investment activity

SoftBank CEO Masa Son never rests on his laurels, which in this case is the Arm IPO.

SoftBank logo intermixed with binary code built from their logo style

Softbank Group

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

The health system in Britain is now deploying AI-powered mental health tools in large-scale clinical settings, while U.S. health insurance companies are trialing them.

Why it matters: AI may be able to help health systems address an overload of patients in need of mental health care.

Driving the news: A diagnostic "e-triage" tool from Limbic, a British AI startup, has screened more than 210,000 patients with a claimed 93% accuracy across the eight most common mental disorders, including depression, anxiety and PTSD, co-founder Ross Harper, a computational neuroscientist, told Axios.

The U.K. National Health Service found that the Limbic Access chatbot arms its psychologists and psychiatrists with information to help them determine the severity and urgency of a potential patient's needs. Misdiagnosis is down — with 45% fewer treatment changes. 

Kintsugi, an American startup that has raised more than $28 million from investors and the National Science Foundation, uses a different approach: its AI-powered voice analysis tool looks for signs of clinical depression and anxiety in short clips of speech. 

Some clinical call centers, telehealth apps and remote patient monitoring systems have already integrated the tool. In a recently published case study Kintsugi conducted with a large U.S. insurer, 4 in 5 patients consented to be screened using the tool — far exceeding a pre-study estimate that only 1 in 4 would consent.

The big picture: Treating mental health is notoriously both complicated and underfunded — general practitioners are only able to correctly diagnose depression in around 50% of cases, and the number of mental health professionals cannot keep up with demand.

Yes, but: Because health care and AI hallucinations aren't a good mix, mental health clinicians have often been hesitant to deploy AI and unsure of how to introduce the concept to patients without overwhelming them.

What's happening: Limbic Access is approved in Britain as the equivalent of a Class II medical device — which the FDA classifies as medium-risk (examples include syringes and electric wheelchairs) — and the company is now looking to expand in the U.S.

Limbic Access saves most clinicians 40 minutes in delivering an in-depth clinical assessment, freeing them to screen more patients and cut waitlists, Harper told Axios.

"This isn't just another app that promises you cognitive behavioral therapy tips. It's patients with a diagnosis or getting a diagnosis using Limbic as part of their care," Harper said.

Harper says that makes Limbic the only mental health chatbot "allowed to take on some level of clinical responsibility." That's critical, he said, because the work of human clinicians will only scale to meet demand once part of the process can be automated. 

The intrigue: Right now, text-based and voice-based initial assessments of a person's mental health offer competing ways to use AI to scale mental health care services. A multi-modal system incorporating both has yet to emerge.

Grace Chang, Kintsugi's founder and CEO, told Axios, "It's not what somebody says, it's how they're saying it that really matters." 

Kintsugi's system uses data from 250,000 people who made voice journals to identify "voice biomarkers." 

Limbic considered but ultimately rejected speech analysis because of the possible barriers it could create for people seeking treatment — from the hassle of enabling microphones to concerns over recording their sensitive thoughts.

What they're saying; Personal experience motivated Kintsugi co-founders Chang and Rima Seiilova-Olson to start their company — each of them struggled for months to secure their first therapy appointment.

"We saw this infrastructure issue where you just have so many people trying to jam through that front door, but not a lot of visibility as to who is severely depressed and who is low to moderate," Chang said.

Harper saw similar issues in England: "There are not enough trained mental health professionals on the planet to serve the astronomical disease prevalence," he said.

New AI tools are helping doctors screen for mental health conditions

With a growing shortage of services for mental health treatment, time-saving diagnostic tools could be a boon.

Illustration of a hand reaching out darkness towards a hand in a white lab coat reaching out of a computer

Mental health

Amazon Web Services CEO Adam Selipsky spreads his AI bets

Adam Selipsky says the cloud market leader is giving customers "full-stack" services and freedom of choice.

Illustration of the Amazon logo extending into a new shape

Amazon Web Services (AWS)

Human Intelligence

Databricks has become one of the AI sector's most valuable companies, this morning announcing over $500 million in new funding at a $43 billion valuation.

Behind the scenes: CEO Ali Ghodsi says the new round was sparked by inbound interest from chipmaker Nvidia, with whom almost every AI company wants a deep relationship.

T. Rowe Price led the round, which is entirely primary capital, and was joined by Nvidia, Capital One Ventures, Ontario Teachers, Andreesen Horowitz, Baillie Gifford, ClearBridge Investments, Counterpoint Global (Morgan Stanley), Fidelity, Franklin Templeton, GIC, Octahedron Capital and Tiger Global.

The San Francisco-based company now has around 6,000 employees, which is double from when it last raised funding two years ago.

What is does: Databricks describes itself as a data lakehouse architecture and AI company.

Axios AI+ co-author Ryan Heath translates: "They specialize in helping companies organize and manage data, which is really the first barrier to deploying AI. Cloud-based, and their biggest competitor is probably AWS."

Another competitor is OpenAI, which is complicated because Databricks also partners with OpenAI investor Microsoft. "It's the nature of the business," Ghodsi says. "We're both a customer and a competitor, but all of the clouds would rather have our business than not ... It's a land grab right now in AI."

Earlier this year Databricks spent around $1.3 billion to buy large language model training and deployment startup MosaicML.

Zoom out: Ghodsi wasn't in Washington, D.C., yesterday for the AI confab organized by Sen. Chuck Schumer (D-N.Y), but is heartened by the early conversations.

"In Silicon Valley we have a tendency to run away with the tech and laugh at people in D.C. who don't understand," he says. "Then we get to a stage where it's obvious things need to be regulated and both Silicon Valley and D.C. overreact."

Look ahead: Databricks wants to become a public company, with all the internal compliance processes already in place, but is part of a stubbornly large cohort that's waiting for others to go first.

One big issue could be profitability vs. growth. There have been reports that Databricks has massive burn — something Ghodsi wouldn't comment on, before insisting that the company could flip to profitability within two quarters if it felt the need.

His perspective is that the profit vs. growth question is mostly answered by interest rates, which therefore could become a signal for when the company plans to drop its S-1.

Databricks now valued at $43 billion, as AI "land grab" intensifies

T. Rowe Price led the $500 million investment, with Nvidia among the new backers.

Illustration of bricklaying where the bricks have a digital binary glow 

Metaphysic, which made its name via convincing deepfake videos of celebrities, is launching a tool Thursday that aims to help individual performers secure their rights to their own images.

Why it matters: In an era when AI is making it easy to generate images of celebrities doing and saying things they never did or said, Metaphysic aims to help them "secure their face, voice and performance data" and "build a portfolio of their most valuable digital assets and AI training datasets over time."

What's happening: Metaphysic is encouraging creators to use its new Metaphysic Pro product to create an AI likeness and "personally register their copyright with the U.S. Copyright Office." 

The big picture: AI enables intellectual property owners to create without creators, and that's setting up new conflicts in every creative industry.

Hollywood writers, with widespread support from actors, are on strike over failed contract negotiations, and use of AI is a flashpoint in the dispute.

Creators want greater protections and compensation when it comes to how their material and likenesses can be used. 

Driving the news: Early users of Metaphysic Pro are said by the company to include Tom Hanks, Anne Hathaway, Octavia Spencer, and Paris Hilton. 

Metaphysic is pitching its platform as one way to resolve intellectual property (IP) disputes.

"We need to support a secure, transparent platform for performers, IP holders and filmmakers to coordinate the use of personal data to create AI content. And it needs to have consent and compensation at its heart," said Thomas Graham, co-founder and CEO of Metaphysic.

The new platform allows creators to "create, store and protect their personal biometric data and manage how it is used by third parties and generative AI to build photorealistic performances, digital identities and AI avatars," a Metaphysic spokesperson told Axios.

Yes, but: The extent to which copyright may apply to AI-generated content remains unclear.  

While courts and copyright authorities have usually required proof of human authorship before granting copyright, the lack of comprehensive legislation leaves plenty of room for interpretation. 

The U.S. Copyright Office has on at least three occasions in 2023 rejected copyright protection for art created using AI, most recently an award-winning artwork submitted by Jason M. Allen which he created using Midjourney.

Flashback: Metaphysic CEO Tom Graham spoke to Axios in February about why he thinks it's important to partner with artists in developing AI content tools.

convincing deepfake videos of celebrities

Startup gives stars new way to lock down their AI likenesses

A new service from Metaphysic encourages public figures to store and copyright their images.

Metaphysic CEO Tom Graham Photo Illustration: Lindsey Bailey/Axios. Photo: Steven Simione

Tech Policy

Senators left Majority Leader Chuck Schumer's AI Insight Forum in D.C.  Wednesday talking about an urgent need to pass legislation governing AI, while CEOs agreed that Washington must play a role. 

Of note: After the closed-to-the-public meeting, Elon Musk said it was important for tech leaders to "have a referee" in Washington and suggested the meeting "may go down in history as very important to the future of civilization."

Why it matters: The event strengthened Washington's consensus that  government must take a strong hand with the new technology — even as the likelihood of immediate action remained low.

There are many more forums like Wednesday's on tap, and Congress is preoccupied with a looming partisan budget deadlock.

Driving the news:  More than 60 senators showed up to the closed-door briefing — with Musk, Bill Gates, Sundar Pichai, Mark Zuckerberg and Sam Altman among 20 tech and civil society leaders in the room.

What they're saying: Schumer told reporters, "I asked everyone in the room, does government need to play a role in regulating AI? And every single person raised their hands, even though they had diverse views. That gives us a message — we have to try to act, as difficult as the process is." 

In an interview with Axios after the event, Schumer said, "This is one of the most difficult issues Congress can ever deal with — it's so complicated, even [former Google CEO] Eric Schmidt said he doesn't understand the algorithms."

"Government needs to act a whole lot faster than it ever has in the past," Sen. Gary Peters (D-Mich.) told Axios.

Commerce Committee chair Sen. Maria Cantwell (D-Wash.) said AI legislation can get done in the next year, but Republican senators, including Mike Rounds of South Dakota, warned that Congress is not yet ready to legislate.

Schumer noted, "The EU went too fast, now they have to pull back," a reference to June changes to the bloc's draft AI Act to take account of generative AI advances.

What's next: Legislative work will ultimately happen in committee.

Schumer told Axios he envisions AI following the same model as the CHIPS and Science Act, where legislation will be pursued in a bipartisan way and committees rely on scientific and technical information from the forums.

Cantwell said the forum is speeding up her panel's process because they're gathering a lot of input from key players.

Yes, but: Schumer is facing criticism about the closed-door nature of the first event and the fact that senators couldn't ask questions directly to the CEOs.

Sen. Josh Hawley (R-Mo.) urged that tech CEOs be required to testify in open hearings instead.

"All of the senators are sitting there, and ask no questions, that's what's happening," Sen. Elizabeth Warren (D-Mass.) told reporters, urging an end to "special treatment" of big tech companies, and "a tech industry bill that addresses that head on."

Zoom in: One theme that emerged from senators and attendees was maintaining U.S. leadership on the global stage.

"I believe it's better that the standard is set by American companies that can work with our government to shape these models on important issues," per the prepared remarks of Meta CEO Mark Zuckerberg.

Zuckerberg has mounted that argument in various forms since 2019, sometimes directly referencing strategic competition from China.

Other notable parts of the forum, according to Schumer:

CEOs said government funding is needed for "transformational" innovations that could help cure cancer, combat hunger or bolster national security.

There was agreement that only government can put in the guardrails needed for "sustainable" innovation that negates harms like bias.

CEOs tell senators: Time to regulate AI

Elon Musk, Mark Zuckerberg, Bill Gates and other luminaries all urged a Senate gathering to act.

All Automation and AI stories