Technology

➚ MOVE FAST: Run. The EU has been a tougher regulator of big tech companies than the US. But so far, AI-related deals and partnerships have been cleared. The latest is Nvidia’s $700 million purchase of Run:ai.

➘ BREAK THINGS: Walk. Amazon drivers have gone on strike at seven delivery hubs over contract negotiations. The company said it won’t affect operations but the walkout could delay holiday shipments in several big US cities.

Google released its first “reasoning” model based on Gemini 2.0, called Flash Thinking, which uses techniques pioneered by OpenAI’s o1 model. The idea is that models can break prompts into different steps and “think over” the answers. Flash Thinking leaves an audit trail of its thought process, a new innovation helpful for understanding how the models work and potentially how to make them safer.

TechCrunch called out the model for making a simple mistake: Miscounting the number of Rs in Strawberry (which happened to be the code name for o1).

But I wouldn’t count Gemini out just yet. After last week’s flurry of announcements, I think Google DeepMind is pursuing a long-term path that is fundamentally its own, believing that in the long run, it will take the lead. And this is an ultra marathon. Nobody is on the verge of building “AGI” or making some giant leap ahead of competitors.

For that reason, I think Google’s models will be bad at some things and better at others. We kind of saw this happen between Gemini 1.0 and 2.0. The first version was terrible at coding, but then jumped ahead in the second version. It’s not entirely clear why at the moment, but likely relates to building a “natively” multimodal model.

David Singleton, Hugo Barra, Ficus Kirkpatrick, and Nicholas Jitkoff are the founders of /dev/agents, which recently raised $56 million in a seed round led by Index Ventures and Alphabet’s CapitalG. The alumni of Google, Meta, and Stripe launched a company to make agentic AI a reality.

Reed Albergotti: Today, you could have an agent on your own network, within your own corporate ecosystem. But at some point, they have to make the leap outside of that, and then it becomes really powerful — where my agent is talking to someone else’s agent to schedule a meeting. Is that where you guys are trying to develop the ecosystem?

David Singleton: If you think about it from a consumer-facing perspective, today, our devices just make us work way too hard to accomplish anything that we care about. Don’t get me wrong, smartphones are awesome. We helped invent a lot of the key components, and we got a lot out of them. But as we all feel the potential that AI brings, we’re really now starting to feel the limitations of the smartphone computing paradigm. Spend any time with ChatGPT and then go and use Google Home or just your phone, you get really frustrated. Like, why can’t it do that? That’s because app and website workflows are really kind of primitive, they’re very repetitive. They treat every user the same way, generically, not individually.

You can imagine today a company like DoorDash could build an AI agent where you go into the app and just say, ‘sushi,’ and it takes care of the rest. Then you’re looking at a screen, you’re still navigating the screen of apps.

Instead of going to the grid of apps to find the app for me, the agents should come to me instead. We think that means that agentic experiences will, for many workflows, actually replace apps today. And what we need is a system that is able to connect me and my personal context to all of the agents that could serve it. And that’s what we’re building.

What is an app in this ecosystem? As you were saying, ‘I don’t really want to open up this DoorDash app.’

The apps are agents, they will come to you when you need them. We will build the capability with the first-party experience to actually understand your context and then bring the right agent to you. And that’s just a very different way of structuring this ecosystem. It means that there are opportunities for developers to build quite niche experiences. As long as they can have the context that this could be useful for this person right now, we can actually have those quality experiences get in front of users and rise to the top. And that’s very difficult in today’s ecosystem. Unless you’re in the top chart of apps, you’re unlikely to get a lot of discovery. So we’re pretty excited about how this changes the game for developers.

What’s also different about this is the probabilistic nature of the technology. How do you ensure that these agents will do what they’re supposed to do and not go off the rails?

We’ve personally built a bunch of agents, sat with developers who’ve been building agents. We’ve gotten a pretty good handle on how you can use the core of what we’ll build on our SDK to make sure that you have very dependable performance. This is possible, so we will help solve for that. It is also the case that some of the most sensitive things that you might want to do, we as a system will still make sure that users have the opportunity to say, ‘I want to approve those things.’ There’ll be certain actions where an agent can propose them to the core operating system to say, ‘Let’s present this deterministically to the user.’ We think that by combining those things, we’re going to be able to create great consumer experiences and also great developer productivity, which is important.

Read here for the rest of the conversation, including Singleton’s view on how agentic AI will be monetized.  →

The amount that North Korean hackers stole in 2024 through crypto hacks — more than double what they collected last year, according to research firm Chainalysis. The news comes as Bitcoin recently surpassed the $100,000 mark and the US ushers in a crypto-friendly administration.

Semafor announces the launch of The CEO Signal from Semafor Business, an exclusive, invitation-only membership for chief executives of the world’s largest companies.

Helmed by veteran Financial Times editor Andrew Edgecliffe-Johnson, the initiative builds on the success of Liz Hoffman’s Semafor Business and sets a new standard for how global leaders connect, learn, and navigate future challenges. Focusing on exclusivity over scale, the platform will debut as a weekly briefing in January 2025 offering candid, practical insights and interviews tailored for global CEOs who are short on time and seeking actionable intelligence.

You can request an invitation for the debut edition here.

Congress-elect members taking their seats for the first time in January own at least $3.8 million and as much as $9.1 million in shares of key tech stocks, financial disclosures show. Elected members report their stock ownership in ranges rather than exact numbers, so the definite amounts aren’t publicly known. Still, the multimillion-dollar figures reveal how the new members’ stock portfolios could benefit from the policies they will have a hand in molding.

With companies racing to develop the largest and most advanced AI models, the coming years are critical in shaping the future of tech regulation. Congress must balance privacy and safety concerns with rules that still allow the US to maintain its lead in AI innovation over China.

Gil Cisneros, a California Democrat elected to the House, leads the group with between $1.3 million and $2.9 million in holdings across Microsoft, Alphabet, Meta, Apple, and Amazon. Pennsylvania Republican Rob Bresnahan and Texas Democrat Julie Johnson own as much as $1.4 million and $1.2 million, respectively, across the five stocks.

Some sitting and former Congressional members own far more in tech shares. Former Speaker of the House Nancy Pelosi has famously loaded up on stocks, with her shares of Apple alone worth more than $25 million.

Members of Congress are not required to divest their stocks in the industries they oversee, but they are expected to avoid conflicts of interest and cannot trade on nonpublic information. President Joe Biden, as well as other lawmakers and advocacy groups, have pushed for stricter rules. “Nobody in the Congress should be able to make money in the stock market while they’re in the Congress,” Biden said in an interview recently.

Funds that hold similar portfolios to sitting members of Congress outpaced the Dow Jones Industrial Average’s growth this year. The tech-heavy ETF aligned with Democrats’ holdings beat the S&P 500.

YouTube is working on tools to help its biggest influencers identify and control their AI lookalike. In partnership with talent firm Creative Artists Agency, the platform will launch tools allowing celebrities to manage videos that use the technology to recreate their appearance and request its removal.

Celebrities, politicians, and public influencers have long been concerned with AI deepfakes, particularly during the 2024 election. (Keanu Reeves puts a clause in his movie contracts that prevents studios from digitally changing him.)

YouTube said it will first test the tools with “award-winning actors and top athletes from the NBA and NFL” early next year as part of a much larger AI testing effort. It has already started trying out similar tools to identify AI-generated voices of famous singers, and recently allowed music labels to request the removal of deepfake voices.

Private investment firms are circling a major life insurer, intensifying a race for Americans’ nest eggs, Semafor’s Liz Hoffman scooped.

Prosperity Life, owned by Elliott Management, held sale talks this month with potential buyers including TPG and European investment firm JAB, people familiar with the matter said. Wall Street has been hoovering up insurance and retirement businesses, which generate cash that they can invest for decades.

To keep up with the news (and the scoops) from Wall Street, subscribe to Semafor’s Business newsletter. →