Alibaba Group researchers have developed "ZeroSearch," a technique that enables large language models to acquire search capabilities without using external search engines during training. The approach transforms LLMs into retrieval modules through supervised fine-tuning and employs a "curriculum-based rollout strategy" that gradually degrades generated document quality.

In tests across seven question-answering datasets, ZeroSearch matched or exceeded the performance [PDF] of models trained with real search engines. A 7B-parameter retrieval module achieved results comparable to Google Search, while a 14B-parameter version outperformed it. The cost savings are substantial: training with 64,000 search queries using Google Search via SerpAPI would cost approximately $586.70, compared to just $70.80 using a 14B-parameter simulation LLM on four A100 GPUs -- an 88% reduction.

The technique works with multiple model families including Qwen-2.5 and LLaMA-3.2. Researchers have released their code, datasets, and pre-trained models on GitHub and Hugging Face, potentially lowering barriers to entry for smaller AI companies developing sophisticated assistants.

In a recent interview with the Council on Foreign Relations, Cloudflare CEO Matthew Prince warned that AI is breaking the economic model of the web by decoupling content creation from value, with platforms like Google and OpenAI increasingly providing answers without driving traffic to original sources. He argued that unless AI companies start compensating creators, the web's content ecosystem will collapse -- calling most current AI investment a "money fire" with only a small fraction holding long-term value. Search Engine Land reports: Google's value exchange with content creators has collapsed, Prince said: "Ten years ago... for every two pages of a website that Google scraped, they would send you one visitor. ... That was the trade. ... Now, it takes six pages scraped to get one visitor." That drop reflects the rise of zero-click searches, which happen when searchers get answers directly on Google's search page. "Today, 75 percent of the queries... get answered without you leaving Google." This trend, long criticized by publishers and SEOs, is part of a broader concern: AI companies are using original content to generate answers that rarely/never drive traffic back to creators.

AI makes the problem worse. Large language models (LLMs) are accelerating the crisis, Prince said. AI companies scrape far more content per user interaction than Google ever has -- with even less return to creators. "What do you think it is for OpenAI? 250 to one. What do you think it is for Anthropic? Six thousand to one." "More and more the answers... won't lead you to the original source, it will be some derivative of that source." This situation threatens the sustainability of the web as we know it, Prince said: "If content creators can't derive value... then they're not going to create original content."

The modern web is breaking. AI companies are aware of the problem, and the business model of the web can't survive unless there's some change, Prince said: "Sam Altman at OpenAI and others get that. But... he can't be the only one paying for content when everyone else gets it for free." Cloudflare's right in the middle of this problem -- it powers 80% of AI companies and a 20-30% of the web. Cloudfaire is now trying to figure out how to help fix what's broken, Prince said. AI = money fire. Prince is not against AI. However, he said he is skeptical of the investment frenzy. "I would guess that 99% of the money that people are spending on these projects today is just getting lit on fire. But 1% is going to be incredibly valuable." "And so maybe we've all got a light, you know, $100 on fire to find that $1 that matters." You can watch a recording of the interview and read the full transcript here.

At its Config 2025 event on Wednesday, Figma unveiled four new AI-powered tools -- Sites, Make, Buzz, and Draw, positioning itself as a full-stack design platform to rival Adobe, WordPress, and Canva. These tools enable users to build websites, generate code, create marketing content, and design vector graphics without leaving the Figma ecosystem. The Verge reports: Figma's first solution is Figma Sites, a website builder that integrates with Figma Design and allows creators to turn their projects into live, functional sites. Figma Sites provides presets for layouts, blocks, templates, and interactions that aim to make building websites less complex and time-consuming. Additional components like custom animations can also be added either using existing code or by prompting Site's AI tool to generate new interaction codes via text descriptions, such as "animate the text to fall into place like a feather." Figma Sites is rolling out in beta for users with full seat access to Figma products. Figma says that AI code generation will be available "in the coming weeks," and that a CMS that allows designers to manage site content will be launched "later this year."

Figma Make is Figma's take on AI coding tools like Google's Gemini Code Assist and Microsoft's GitHub Copilot. The prompt-to-code Figma Make tool is powered by Anthropic's Claude 3.7 model and can build working prototypes and apps based on descriptions or existing designs, such as creating a functional music player that displays a disc that spins when new tracks are played. Specific elements of working design, like text formatting and font style, can be manually edited or adjusted using additional AI prompts. Make is rolling out in beta for full seat Figma users. Figma says it's "exploring integrations with third parties and design systems" for Figma Make and may apply the tool to other apps within its design platform.

Figma Buzz is a marketing-focused design app that's rolling out in beta to all users, and makes it easier for teams to publish brand content, similar to Canva's product design platform. The tool allows Figma designers to create brand-approved templates, styles, and assets that can be used by marketers to quickly assemble emails, social media posts, advertising, and more. Figma Buzz includes generative AI tools for making and editing images using text prompts, and can source information from spreadsheets to bulk create thousands of image assets at once.

Lastly, the Figma Draw vector design app is like a simplified version of Adobe Illustrator that creatives can use to make custom visuals without leaving the Figma platform. It includes a variety of brushes, texture effects, and vector editing tools to create or adjust scalable images and logos for product design projects. Figma Draw is generally available now for full seat users as a toggle in Figma Design, with some features accessible in Sites, Slides, and Buzz. It's not quite as expansive as Adobe's wider Creative Cloud ecosystem, but Figma Draw places the two companies in direct competition for the first time since Adobe killed its own XD product design platform. It also brings some new options to the creative software industry after Adobe failed to acquire Figma for $20 billion due to pressure from competition regulators.

Google has pushed back against Apple executive Eddy Cue's testimony that Safari searches declined last month, asserting it continues "to see overall query growth in Search" with "an increase in total queries coming from Apple's devices and platforms."

The statement comes as Apple's Senior VP revealed under oath that the company is "actively looking at" revamping Safari to focus on AI-powered search engines, potentially threatening the estimated $20 billion-a-year deal making Google the default search provider on Apple devices.

Cue testified that AI search providers including OpenAI, Perplexity, and Anthropic will "eventually replace standard search engines." Google, in its response, pointed to ongoing enhancements to its search product, noting users are "accessing it for new things and in new ways, whether from browsers or the Google app, using their voice or Google Lens."

Google has partnered with Elementl Power to develop at least 600 MW of nuclear capacity at each of three planned sites. It's unknown where the three proposed sites will be located or how much Google is investing. World Nuclear News reports: The two companies will work "with utility and regulated power partners to identify and advance new projects" and Elementl "will continue the evaluation of potential technology, engineering, procurement and construction, and other project partners, while prioritising specific sites for accelerated development."

Elementl Power, founded in 2022, describes itself as a technology-agnostic advanced nuclear project developer which aims to provide "turn-key development, financing and ownership solutions for customers that want access to clean baseload power but may not want to own or operate nuclear power assets." It says its mission is to "to deploy over 10 gigawatts of next-generation nuclear power in the US by 2035."

It is not Google's first nuclear power deal -- in October 2024 the company signed an agreement with Kairos Power to purchase power from its fluoride salt-cooled high-temperature small modular reactors, with a fleet of up to 500 MW of capacity by 2035. The aim of the power purchase agreement was to facilitate Kairos Power to develop, construct, and operate plants and sell energy, ancillary services, and environmental attributes to Google. At the time of that announcement Google said that it would help it achieve net-zero emissions across all of its operations and value chain by 2030. Further reading: Google tries to greenwash massive AI energy consumption with another vague nuclear deal (The Register)

Eddy Cue, Apple's senior vice president of services, gave an ominous warning today that the iPhone could go the way of the iPod 10 years from now. From a report: Cue's remarks came during the Google Search antitrust remedies trial today while discussing how AI has the potential to reshape the tech industry and open the door to new entrants. Incumbents have a hard time ... we're not an oil company, we're not toothpaste -- these are things that are going to last forever ... you may not need an iPhone 10 years from now.

Cue went on to say that the best thing Apple did was kill the iPod, a move he said was bold. "Why would you kill the golden goose," he added. That may seem like a silly thing for Apple to say, given that more than half of its revenue is iPhone sales. But Cue calls AI a "huge technological shift," and suggests that such shifts can humble companies that once seemed unassailable.

Apple is "actively looking at" revamping the Safari web browser on its devices to focus on AI-powered search engines, a seismic shift for the industry hastened by the potential end of a longtime partnership with Google. From a report: Eddy Cue, Apple's senior vice president of services, made the disclosure Wednesday during his testimony in the US Justice Department's lawsuit against Alphabet. The heart of the dispute is the two companies' estimated $20 billion-a-year deal that makes Google the default offering for queries in Apple's browser. The case could force the tech giants to unwind the pact, upending how the iPhone and other devices have long operated.

Cue noted that searches on Safari dipped for the first time last month, which he attributed to people using AI. Cue said he believes that AI search providers, including OpenAI, Perplexity and Anthropic, will eventually replace standard search engines like Alphabet's Google. He said he believes Apple will bring those options to Safari in the future. "We will add them to the list -- they probably won't be the default," he said, indicating that they still need to improve.

Rep. Kat Cammack (R-FL) introduced the App Store Freedom Act [PDF] on Tuesday, legislation that would compel "large app store operators" with over 100 million US users to permit third-party app stores and allow them to be set as defaults. The bill directly challenges Apple's walled garden approach and Google's Play Store dominance by requiring both companies to allow developers to use alternative payment systems, bypassing the platforms' commission structures.

It would also mandate equal access to development tools and interfaces without discrimination, while giving users the ability to remove pre-installed apps. Violations would trigger FTC enforcement with penalties up to $1 million per infraction. The legislation mirrors recent European Union regulations that have already forced Apple to permit third-party app stores and allow users to change default apps.

An anonymous reader quotes a report from Ars Technica: Google's accelerated Android release cycle will soon deliver a new version of the software, and it might look quite different from what you'd expect. Amid rumors of a major UI overhaul, Google seems to have accidentally published a blog post detailing "Material 3 Expressive," which we expect to see revealed at I/O later this month. Google quickly removed the post from its design site, but not before the Internet Archive saved it.

It has been a few years since Google introduced any major changes to its Material theming, but the design team wasn't just sitting idly this whole time. According to the leaked blog post, Google has spent the past three years working on a more emotionally engaging vision for Android design. While the original Material Design did an admirable job of leveraging colors and consistent theming, it could make apps look too similar. The answer to that, apparently, is Material 3 Expressive.

Google says this is "the most-researched update to Google's design system, ever." The effort reportedly included 46 separate studies with hundreds of sample designs. The team showed these designs to more than 18,000 study participants to understand how the user experience would work. In these studies, the design team used a variety of metrics, including the following:
- Eye tracking: Analyzing where users focus their attention
- Surveys and focus groups: Gauging emotional responses to different designs
- Experiments: Gathering sentiment and preferences
- Usability: Seeing how quickly participants could understand and use an interface "The result of all this is an interface that appears much more varied than the previous Material Design," writes Ars.

You can check out 9to5Google's article, which preserved many of the blog post's visuals before they were removed.

An anonymous reader quotes a report from TechCrunch: Google on Tuesday announced the launch of Gemini 2.5 Pro Preview (I/O edition), an updated version of its flagship Gemini 2.5 Pro AI model that the company claims tops a number of widely used benchmarks. Gemini 2.5 Pro Preview (I/O edition) is available via the Gemini API and Google's Vertex AI and AI Studio platforms, and is priced the same as the Gemini 2.5 Pro model it effectively replaces. It's also in Google's Gemini chatbot app for the web and for mobile devices.

The model's release comes ahead of Google's annual I/O developer conference (hence the "I/O edition" designation), where Google is expected to unveil a host of models, as well as AI-powered tools and platforms. [...] According to Google, Gemini 2.5 Pro Preview (I/O edition) has "significantly" improved capabilities for coding and building interactive web apps. The model is also better at tasks like code transformation -- that is, modifying a piece of code to achieve a specific goal -- and code editing, the company says. Google says the Gemini 2.5 Pro Preview (I/O edition) leads the WebDev Arena Leaderboard, a benchmark measuring a model's ability to create aesthetically pleasing and functional web apps. It also achieved a score of 84.8% on VideoMME, a popular benchmark designed to evaluate the video analysis capabilities of multi-modal large language models.

Yahoo Finance reports that Mark Zuckerberg made bold predictions for investors on Meta's earnings call this week — about advertisers. "AI has already made us better at targeting and finding the audiences that will be interested in their products than many businesses are themselves," Zuck said, "and that keeps improving..."

"If we deliver on this vision, then over the coming years, I think that the increased productivity from AI will make advertising a meaningfully larger share of global GDP than it is today..." If investors are still searching for answers to nagging questions about how massive AI investments will pay off, Zuckerberg provided the clearest reply yet: It will strengthen our core business. In fact, it is our business... On what many believe to be the cusp of an economic downturn, Meta isn't pitching its AI developments as an add-on to its operations, but as something central to its core proposition of targeted advertising...

"While Meta's investments in GenAI have spooked certain investors who continue to question the return on these investments, we saw further signs of GenAI monetization in the firm's ad business," wrote Morningstar equity analyst Malik Ahmed Khan in a note on Thursday. In a powerful showing, coming after Alphabet's own impressive results, Meta noted that a new ads recommendation model it's testing for Reels has already boosted conversion rates by 5%. And nearly one-third of advertisers were using AI creative tools in the past quarter. For Zuckerberg, the enhancements AI offers to finding the right consumers and providing measurable results strengthen the case for boosting capacity and for a revamped model of advertising's scope.

And with the company set to invest upwards of $70 billion toward its AI opportunity this year, the bet is not all about ads, of course. Zuckerberg outlined four other areas of focus for its AI efforts: business messaging, Meta AI, AI devices, and more engaging experiences. Meta's efforts can also be viewed as an ambitious play to take on its rivals across tech's legacy and emerging platforms. As John Blackledge, senior analyst at TD Cowen, said in a note on Thursday, the AI opportunities Zuckerberg outlined are about "ultimately taking on Google search, iPhone and ChatGPT all at once."
In the pre-AI world, "Businesses used to have to generate their own ad creative and define what audiences they wanted to reach," Zuckerberg told Meta's investors this week.

And by Friday's closing, Meta's stock had jumped 12.6% over its value Wednesday morning, leading Yahoo Finance to conclude that Wall Street "appears to be buying into" Zuckerberg's vision.

Did you know there's an initiative to drive Open Source adoption both within the United Nations — and globally? Launched in March, it's the work of the Digital Technology Network (under the UN's chief executive board) which "works to advance open source technologies throughout UN agencies," promoting "collaboration and scalable solutions to support the UN's digital transformation." Fun fact: The first group to endorse the initiative's principles was the Open Source Initiative...

"The Open Source Initiative applauds the United Nations for recognizing the growing importance of Open Source in solving global challenges and building sustainable solutions, and we are honored to be the first to endorse the UN Open Source Principles," said Stefano Maffulli, executive director of OSI.
But that's just the beginining, writes It's FOSS News: As part of the UN Open Source Principles initiative, the UN has invited other organizations to support and officially endorse these principles. To collect responses, they are using CryptPad instead of Google Forms... If you don't know about CryptPad, it is a privacy-focused, open source online collaboration office suite that encrypts all of its content, doesn't log IP addresses, and supports a wide range of collaborative documents and tools for people to use.

While this happened back in late March, we thought it would be a good idea to let people know that a well-known global governing body like the UN was slowly moving towards integrating open source tech into their organization... I sincerely hope the UN continues its push away from proprietary Big Tech solutions in favor of more open, privacy-respecting alternatives, integrating more of their workflow with such tools.
16 groups have already endorsed the UN Open Source Principles (including the GNOME Foundation, the Linux Foundation, and the Eclipse Foundation).

Here's the eight UN Open Source Principles:

1. Open by default: Making Open Source the standard approach for projects
2. Contribute back: Encouraging active participation in the Open Source ecosystem
3. Secure by design: Making security a priority in all software projects
4. Foster inclusive participation and community building: Enabling and facilitating diverse and inclusive contributions
5. Design for reusability: Designing projects to be interoperable across various platforms and ecosystems
6. Provide documentation: Providing thorough documentation for end-users, integrators and developers
7. RISE (recognize, incentivize, support and empower): Empowering individuals and communities to actively participate
8. Sustain and scale: Supporting the development of solutions that meet the evolving needs of the UN system and beyond.

An anonymous reader shared this report from The Verge: Firefox could be put out of business should a court implement all the [U.S.] Justice Department's proposals to restrict Google's search monopoly, an executive for the browser owner Mozilla testified Friday. "It's very frightening," Mozilla CFO Eric Muhlheim said.

The Department of Justice wants to bar Google from paying to be the default search engine in third-party browsers including Firefox, among a long list of other proposals including a forced sale of Google's own Chrome browser and requiring it to syndicate search results to rivals. The court has already ruled that Google has an illegal monopoly in search, partly thanks to exclusionary deals that make it the default engine on browsers and phones, depriving rivals of places to distribute their search engines and scale up. But while Firefox — whose CFO is testifying as Google presents its defense — competes directly with Chrome, it warns that losing the lucrative default payments from Google could threaten its existence.

Firefox makes up about 90 percent of Mozilla's revenue, according to Muhlheim, the finance chief for the organization's for-profit arm — which in turn helps fund the nonprofit Mozilla Foundation. About 85 percent of that revenue comes from its deal with Google, he added. Losing that revenue all at once would mean Mozilla would have to make "significant cuts across the company," Muhlheim testified, and warned of a "downward spiral" that could happen if the company had to scale back product engineering investments in Firefox, making it less attractive to users. That kind of spiral, he said, could "put Firefox out of business." That could also mean less money for nonprofit efforts like open source web tools and an assessment of how AI can help fight climate change.

Ironically, Muhlheim seemed to suggest that could cement the very market dominance the court seeks to remedy. Firefox's underlying Gecko browser engine is "the only browser engine that is held not by Big Tech but by a nonprofit," he said.

An anonymous reader quotes a report from the New York Times: Google plans to roll out its Gemini artificial intelligence chatbot next week for children under 13 (source paywalled; alternative source) who have parent-managed Google accounts, as tech companies vie to attract young users with A.I. products. "Gemini Apps will soon be available for your child," the company said in an email this week to the parent of an 8-year-old. "That means your child will be able to use Gemini" to ask questions, get homework help and make up stories. The chatbot will be available to children whose parents useFamily Link, a Google service that enables families to set up Gmail and opt into services like YouTube for their child. To sign up for a child account, parents provide the tech company with personal data like their child's name and birth date. Gemini has specific guardrails for younger users to hinder the chatbot from producing certain unsafe content, said Karl Ryan, a Google spokesman. When a child with a Family Link account uses Gemini, he added, the company will not use that data to train its A.I.

Introducing Gemini for children could accelerate the use of chatbots among a vulnerable population as schools, colleges, companies and others grapple with the effects of popular generative A.I. technologies. Trained on huge amounts of data, these systems can produce humanlike text and realistic-looking images and videos. [...] Google acknowledged some risks in its email to families this week, alerting parents that "Gemini can make mistakes" and suggesting they "help your child think critically" about the chatbot. The email also recommended parents teach their child how to fact-check Gemini's answers. And the company suggested parents remind their child that "Gemini isn't human" and "not to enter sensitive or personal info in Gemini." Despite the company's efforts to filter inappropriate material, the email added, children "may encounter content you don't want them to see."

The U.S. government is seeking to break up Google's advertising technology business after a judge ruled the company holds an illegal monopoly over ad tools for publishers, marking the second such antitrust case following a similar request to divest Chrome. The Guardian reports: "We have a defendant who has found ways to defy" the law, US government lawyer Julia Tarver Wood told a federal court in Virginia, as she urged the judge to dismiss Google's assurance that it would change its behavior. "Leaving a recidivist monopolist" intact was not appropriate to solve the issue, she added. [...] The US government specifically alleged that Google controls the market for publishing banner ads on websites, including those of many creators and small news providers.

The hearing in a Virginia courtroom was scheduled to plan out the second phase of the trial, set for September, in which the parties will argue over how to fix the ad market to satisfy the judge's ruling. The plaintiffs argued in the first phase of the trial last year that the vast majority of websites use Google ad software products which, combined, leave no way for publishers to escape Google's advertising technology and pricing.

The district court judge Leonie Brinkema agreed with most of that reasoning, ruling last month that Google built an illegal monopoly over ad software and tools used by publishers, but partially dismissed the argument related to tools used by advertisers. The US government said it would use the trial to recommend that Google should spin off its ad publisher and exchange operations, as Google could not be trusted to change its ways. "Behavioral remedies are not sufficient because you can't prevent Google from finding a new way to dominate," Tarver Wood said.

Google countered that it would recommend that it agree to a binding commitment that it would share information with advertisers and publishers on its ad tech platforms. Google lawyer Karen Dunn did, however, acknowledge the "trust issues" raised in the case and said the company would accept monitoring to guarantee any commitments made to satisfy the judge. Google is also arguing that calls for divestment are not appropriate in this case, which Brinkema swiftly refused as an argument. The judge urged both sides to mediate, stressing that coming to a compromise solution would be cost-effective and more efficient than running a weeks-long trial.

An anonymous reader shares a report: In the nearly two months since former Google chief executive Eric Schmidt acquired Relativity Space, the billionaire has not said much publicly about his plans for the launch company. However, his intentions for Relativity now appear to be increasingly clear: He wants to have the capability to launch a significant amount of computing infrastructure into space.

We know this because Schmidt appeared before the House Committee on Energy and Commerce during a hearing in April, speaking on the future of AI and US competitiveness. Among the topics raised then was the need for more electricity -- both renewable and non-renewable -- to power data centers that will facilitate the computing needs for AI development and applications. Schmidt noted that an average nuclear power plant in the United States generates 1 gigawatt of power.

"People are planning 10 gigawatt data centers," Schmidt said. "Gives you a sense of how big this crisis is. Many people think that the energy demand for our industry will go from 3 percent to 99 percent of total generation. One of the estimates that I think is most likely is that data centers will require an additional 29 gigawatts of power by 2027, and 67 more gigawatts by 2030. These things are industrial at a scale that I have never seen in my life."

Redis, the popular in-memory data store, has returned to open source licensing with Redis 8 now available under the AGPL v3 license. The move reverses last year's controversial shift to proprietary licensing schemes (RSALv2 and SSPLv1) that aimed to force major cloud providers to pay for offering Redis as a managed service.

The decision follows significant market pressure, including AWS, Google, and Oracle backing the Valkey fork, which gained momentum in the open source community.

Redis believes the AGPL license provides sufficient protection from cloud providers while satisfying open source requirements. Redis 8 will incorporate vector sets and integrate previously separate Redis Stack features including JSON, Time Series, and probabilistic data support.

An anonymous reader quotes a report from TechCrunch: Epic Games is taking a victory lap. After notching a big win against Apple in a years-long legal dispute, Epic announced that its Epic Games Store will allow developers to open webshops, which can offer players out-of-app purchases to circumvent fees from Apple and Google. [...] With the Epic Games Store's new webshops feature, other developers will be more easily able to follow suit.

Usually, Epic takes a 12% share of a developer's earnings from the Epic Games Store, which is still a better deal than what developers get from Apple. But starting in June, Epic Games will not take a cut from the first $1 million each game earns annually. Only after a game eclipses $1 million in revenue will Epic begin taking a cut. "With new legal rulings in place, developers will be able to send players from games to make digital purchases from webshops on any platform that allows it, including iOS in the European Union and United States," Epic said.

A "small percentage" of Google's users in the US will begin seeing an AI Mode tab in Google Search "in the coming weeks," the company said Thursday, marking the tool's first deployment outside the company's experimental Labs environment.

Unlike traditional search results that display URLs based on user queries, AI Mode generates conversational responses from Google's search index. The feature will appear as a dedicated tab positioned before the standard "All," "Images," and other search filters. The deployment represents Google's direct challenge to LLM-powered search engines like Perplexity and ChatGPT.

AI Mode differs from existing AI Overviews in Google Search, which merely insert AI summaries between the search box and web results.

Nvidia publicly criticized AI startup Anthropic on Thursday over claims about Chinese smuggling tactics, just days before the Biden-era "AI Diffusion Rule" takes effect on May 15. The confrontation highlights growing tensions between AI hardware providers and model developers over export controls.

"American firms should focus on innovation and rise to the challenge, rather than tell tall tales that large, heavy, and sensitive electronics are somehow smuggled in 'baby bumps' or 'alongside live lobsters,'" an Nvidia spokesperson said, responding to Anthropic's Wednesday blog post.

The Amazon and Google-backed AI startup had called for tighter restrictions and enforcement, arguing that "maintaining America's compute advantage through export controls is essential for national security." Anthropic specifically proposed lowering export thresholds for Tier 2 countries to prevent China from gaining ground in AI development.

Nvidia countered that policy shouldn't be used to limit competitiveness: "China, with half of the world's AI researchers, has highly capable AI experts at every layer of the AI stack. America cannot manipulate regulators to capture victory in AI."