Inside Google’s AI culture clash

Happy Thursday from Los Angeles. I’m headed up to the Bay Area next week for Google’s I/O conference on Wednesday. It feels like an unusually consequential I/O this year for reasons I get into below. If you’re reading this and will be there as well, I’d love to meet up. In case you missed it last Friday, I was on Slate’s “What’s Next” podcast with Stephen King and Jon Favreau to talk about the past year of Twitter drama. You can find the show at this link and listen in your podcast player of choice. --- [Image: https://cdn.vox-cdn.com/thumbor/sBQJCMLspwB5KR_IN-FsYHV-74s=/0x0:480x479/480x479/filters:focal(240x240:241x241)/cdn.vox-cdn.com/uploads/chorus_asset/file/24706014/Command_Line_DeepMind.jpeg] “The only company to ever be acquired by Google twice” Two weeks ago, Google announced what will likely go down as its most consequential reorg in years: DeepMind, the London-based AI research group acquired by Google for over $500 million in 2014, will merge with Google Brain, the search giant’s in-house research group headquartered in Mountain View. Google CEO Sundar Pichai said the combination of the two groups will “significantly accelerate our progress in AI.” I’m told the timing of the news shocked employees on both sides, including some senior folks who were hearing that a merger likely wouldn’t happen for at least several more months. Now that it’s official, the hard part starts. As has been well documented over the years, Google Brain and DeepMind have operated as independent fiefdoms with totally different cultures. Under the leadership of Jeff Dean, Brain has historically fostered a bottom-up work environment that gives wide latitude for researchers to pursue their interests and publish their work externally. DeepMind, by contrast, is known for a more top-down culture where CEO Demis Hassabis spins up teams to tackle specific challenges. Researchers there have complained that “they have to battle through layers of internal approval before they can even submit work to conferences and journals.” In my conversations with Google employees over the years, I’ve heard enough to know that integrating the teams won’t be easy for Hassabis, who is now leading the combined group as CEO of “Google DeepMind.” Some Google employees have privately grumbled that DeepMind is more focused on marketing and hype than delivering breakthroughs for the industry like Brain’s seminal Transformer research, which the “T” in ChatGPT stands for. DeepMind, meanwhile, lost its battle to gain even more independence from Google in 2021. “Neither side ‘won’ this merger,” according to Brian Kihoon Lee, a senior Google Brain researcher who was recently laid off. “I think both Brain and DeepMind lose. I expect to see many project cancellations, project mergers, and reallocations of headcount over the next few months, as well as attrition.” The coming reorg will be a test of where power lies. Is it with Google’s leadership team, which clearly feels a newfound sense of urgency to ship AI products and consolidate focus? Or is it with the company’s highly-paid, increasingly-vocal AI researchers who have a plethora of competing firms knocking down their doors with job offers? (“Brain drain” from Google has already been a problem for some time.) As a DeepMind researcher put it on Twitter when the merger was announced, “DeepMind may be the only company to ever be acquired by Google twice.” The first time, Google was swooping in to outbid Facebook and secure early dominance in the field of AI. This time, Google feels suddenly behind as the AI race is heating up like never before. It won’t be easy to integrate two radically different organizations while also trying to compete. --- This week in AI * The “godfather of AI,” Geoffrey Hinton, left Google to sound the alarm about how AI may end up destroying humanity. This story really broke into the mainstream zeitgeist. My favorite reaction is Snoop Dogg’s: “I’m like, is we in a fucking movie right now or what? The fuck, man?” * Two reports revealed how generative AI is already being abused at scale. One report from NewsGuard shows how websites are using AI to create SEO spam. And research from Meta found more than 1,000 websites distributing malware through ChatGPT-like scams. * Chegg’s stock price lost almost half its value after the company said that its business is being hurt by students using ChatGPT for their homework. * Inflection, the AI startup led by DeepMind co-founder Mustafa Suleyman and Reid Hoffman, debuted its AI chatbot called Pi. There are already too many chatbots out there, so I’ll be curious to see if Pi can manage to stand out from the crowd. * The White House gathered AI CEOs for a meeting with Vice President Kamala Harris today. Meta wasn’t invited because the administration thinks it’s not a “leader” on “the consumer facing product side” (someone should tell them how the News Feed works). The National Science Foundation will also spend $140 million on new AI research centers, which is roughly what a buzzy AI startup can raise for its Series A these days. * Microsoft announced that Bing’s AI chatbot is now open to everyone, and that it can show video and image results. Developers can also start building plug-ins that work similarly to OpenAI’s for ChatGPT. The relationship between Microsoft and OpenAI is looking increasingly conflicted, given how the two keep releasing competing features. * An allegedly leaked document from an anonymous Google AI researcher is making the rounds today. It argues that open-source models will ultimately win out, and that “paradoxically, the one clear winner in all of this is Meta.” Maybe staff at the White House will read it at some point. --- Spotted recently * Sergey Brin mingling and taking pictures with Google employees after an internal event at the company’s HQ. * Chris Riedy, Twitter’s VP of sales and marketing, addressing a room of advertisers at Twitter’s NYC office. * Sam Altman and Sundar Pichai walking into the White House together today. --- Closing the loop * After I reported last week on Meta talking to Microsoft and OpenAI about paying for a dedicated ChatGPT instance for its employees, The Information reported that Microsoft is gearing up for a formal unveil of the product: “Later this quarter Microsoft’s Azure cloud server unit plans to sell a version of ChatGPT that runs on dedicated cloud servers where the data will be kept separate from those of other customers, according to two people with knowledge of the upcoming announcement.” Relatedly, Samsung has banned the use of ChatGPT internally after employees uploaded internal code to it. * A couple of weeks ago, I speculated that Snap would try and hook its My AI chatbot into its ads business. This week, the company announced that it will start testing sponsored links in My AI and experiment with using its data to tailor how ads are targeted elsewhere in the app. --- People moves * Eric Han, TikTok’s head of US trust and safety, is leaving. (Not a good sign for Project Texas.) * Srinivas Narayanan, Meta’s former VP of AI applied research and engineering, has joined OpenAI as VP of engineering. * Mike Seavers, the ex-EVP of development for Epic Games, is joining Yuga Labs to be CTO. * Nola Weinstein, formerly of Twitter, is Apple’s new head of global events. * Patrick Harris, formerly an ads leader at Meta, has been named Snap’s SVP of partnerships. And David Sommer, the ex-chief commercial officer at Fetch, is now Snap’s head of verticals in the revenue division. * Brent Harris, Meta’s VP of governance who helped create the Oversight Board, has moved to its AR division. * Mike Acton, a Unity VP, is leaving after the company said it is laying off 8 percent of its workforce. * Abraham Shafi, the CEO of the Softbank-backed social media startup IRL, has been suspended by the board due to a “pattern of misconduct.” * Jason Yuan, a member of Apple’s tight-knit design team, is leaving to start an AI company. * Rob Stets is leaving Google after 20 years. --- Interesting links * An analysis of executive compensation at founder-led tech companies by Alimeter’s Thomas Reiner. * The 2023 Forbes Midas list. * Carta’s Q1 state of the private markets report. * Vitalik Buterin on Ethereum 3.0’s roadmap. * a16z breaks down the cost of AI compute. * Nearly half of all YouTube viewing in the US is on TV screens. * Former OpenSea employee Nathaniel Chastain was found guilty of wire fraud for flipping NFTs. * Vanity Fair on the Hollywood writers strike. --- That’s it for this week. If you aren’t already subscribed to Command Line, you can do so here. I’ll be back next Thursday. In the meantime, if you have any feedback on this edition or fun theories about the coming AI apocalypse, let me know. Thanks for subscribing.

Inside Google’s AI culture clash

Inside Google’s AI culture clash

What’s in store for the DeepMind/Google Brain merge. Plus: this week in AI news.

Start your free trial now to continue reading

Already a subscriber?Sign in

Disney’s password-sharing crackdown has begun

The Raspberry Pi 5 is finally here

The iPhone 15 Pro case market is a minefield right now

Watch Linda Yaccarino’s wild interview at the Code Conference

The Fitbit Charge 6 looks like a return to form

More from Command Line

The chaos at Unity

Command Line: The mailbag issue

How Big Tech really feels about the EU’s crackdown

Why Google’s co-founder left retirement to work on AI

Inside Google’s AI culture clash

Inside Google’s AI culture clash

What’s in store for the DeepMind/Google Brain merge. Plus: this week in AI news.

Share this story

Start your free trial now to continue reading

Already a subscriber?Sign in

Disney’s password-sharing crackdown has begun

The Raspberry Pi 5 is finally here

The iPhone 15 Pro case market is a minefield right now

Watch Linda Yaccarino’s wild interview at the Code Conference

The Fitbit Charge 6 looks like a return to form

More from Command Line

The chaos at Unity

Command Line: The mailbag issue

How Big Tech really feels about the EU’s crackdown

Why Google’s co-founder left retirement to work on AI