| In this edition, how Google Gemini’s leadership change indicates a greater shift in the AI race, and͏ ͏ ͏ ͏ ͏ ͏ |
|  | Technology |  |
| |
|
 | Reed Albergotti |
|
Hi, and welcome back to Semafor Tech. Our scoop today on Google’s AI leadership change (see below) is a good opportunity to talk about a big vibe shift that has been occurring in the AI race. When ChatGPT first came out, most of the focus was on the model capabilities. Was GPT-4 going to be AGI? Was Anthropic going to overtake OpenAI? With the advent of “reasoning” models, starting with OpenAI’s o1 last year, it’s now possible to build a wide range of solid, useful products that are increasingly “agentic,” meaning they can take some actions on their own. But the “secret sauce” of successful AI products like Cursor, Harvey, and even ChatGPT is really all the little tricks after the models are trained to get them to work reliably and in a way that makes it easy for people to use. That’s not to say there isn’t a high-stakes race to create the best models in the world. There is, and model capabilities will take center stage again as foundation model companies hit new inflection points in data, compute power, and research. The vibe shift is apparent at Google. By almost all measures, it became the leader in the AI model race with Gemini 2.5 last week, and it seemed like, outside of the tech industry, the world hardly noticed. Instead, they were enamored by cartoon images produced by ChatGPT’s latest image creation tool. Gemini 2.5 is a technological marvel, but its impact is going to come through Google’s ability to infuse technology in a way that gets users hooked on the product in the long term. It’s a two-track effort for the company. Google DeepMind researchers are making breakthroughs in robotics, science, etc. — areas with more transformative potential that could carry Google into the next era as the business model of the web is disrupted. The leadership change at Gemini shows, however, that it’s all about turning foundation models into products right now. ➚ MOVE FAST: Circle. USDC issuer Circle has filed to go public, in what would be one of the most prominent crypto listings in recent years as blockchain assets enjoy the Trump administration’s favor and relaxed regulatory stance. It’s Circle’s second IPO attempt, after a SPAC merger failed in 2022 on regulatory delays and a broader crypto market crash. ➘ BREAK THINGS: Spin. Intel will spin off certain undisclosed assets that aren’t part of its core business, Lip-Bu Tan said in his first public address as CEO, Bloomberg reported. The news indicates Intel’s turnaround plan under the new chief executive is taking shape as the chip firm plays a big game of catch-up. |
|
Berkeley Engineering/YouTubeReading minds. In the latest breakthrough in AI-powered speech for paralyzed individuals, California researchers have developed a new technology that can communicate a person’s intended verbiage by reading neural activity in near real-time. The technology intercepts signals in the brain where thought turns into speech and decodes the message using an algorithm similar to those powering voice assistants like Siri and Alexa, the authors from UC Berkeley and UC San Francisco said. The system takes one second to decode the speech — significantly faster than the eight seconds it took when the researchers tested it in 2023. Quicker decoding can help paralyzed individuals and others with conditions impacting their speech communicate their needs, connect with friends and family, and participate in the natural flow of conversations. Researchers are working on bringing expressivity to the output voice to reflect changes in tone, emphasis, and pitch. |
|
Google is shaking up the leadership of its consumer AI apps as the focus of the AI race shifts from the underlying models to the products built around them, according to memos reviewed by Semafor. Sissie Hsiao, who led Google’s effort to create an AI chatbot, originally called Bard and now dubbed Gemini, will step down immediately. Josh Woodward, who leads Google Labs and oversaw the launch of NotebookLM, the company’s popular tool that turns text into a podcast-like show, will replace her. Google/YouTubeIn a memo to the staff, Google DeepMind CEO Demis Hassabis said the move will “sharpen our focus on the next evolution of the Gemini app.” He said Woodward will remain head of Google Labs while shaping the next chapter of Gemini. Hsiao, in a separate staff memo, called her time as head of the team “chapter 1” of the Bard story, and said she was optimistic in handing the baton to Woodward for “chapter 2.” Hsiao plans to take a “short break” and return to Google in a new role. Hsiao, Woodward, and Hassabis declined to comment through a spokesperson. Hsiao, a 19-year veteran of Google, jumped into action after ChatGPT upended the tech industry, forcing Google to accelerate the rollout of AI technology it had created. While Google researchers had pioneered the transformer-based architecture that allowed the creation of large language models that powered ChatGPT, its unpredictable nature led the company to keep its chatbot experiments mostly under wraps. In the competitive rush, mishaps illustrated why. The launch of Bard was criticized for “hallucinations,” and its Gemini image creation tool became a target of conservatives when it generated embarrassing results, inserting photos of women or people of color when prompted to create images of Vikings, Nazis, and the Pope. But Google quickly moved past those early errors with successful product launches. Last week, Gemini 2.5 blew away AI benchmarks held by competitors like OpenAI and Anthropic. It took the lead in Chatbot Arena, where users vote on their favorite large language model responses. |
|
 CIOs and IT leaders don’t just follow trends — they drive them. That’s why IT leaders at global organizations subscribe to CIO Upside. Get exclusive insights on the innovations and strategies shaping the future of tech leadership — subscribe for free today. |
|
 How much Isomorphic Labs, Alphabet’s AI-powered drug discovery unit, has raised in its first external funding round, led by OpenAI-backer Thrive Capital. Under Demis Hassabis, Isomorphic will spend the money on research and hiring. |
|
President Trump and senior officials are expected to meet today to discuss a potential deal to purchase TikTok’s US operations, with the deadline just days away for its Chinese parent to sell it or face a(nother) ban on American soil, The Wall Street Journal reported. The bid, which must be approved by Beijing, reportedly comes from a consortium of US investors including Oracle and Blackstone.  The deadline, extended to Saturday by the Trump administration, comes as the president looks to tighten trade restrictions in a far-reaching tariff plan expected to be announced today. While TikTok’s parent has repeatedly stated the platform isn’t for sale, Chinese authorities have signaled they are open to a deal but may leverage the app in wider negotiations around tariffs, according to the Journal. Such actions could force Trump to ease tariff plans against China and bend to other demands if he wants to keep his campaign promise to save TikTok. In attendance with Trump will reportedly be Vice President JD Vance, Commerce Secretary Howard Lutnick, National Security Advisor Mike Waltz, and National Intelligence Director Tulsi Gabbard. TikTok’s value and each party’s investment size have not yet been decided. |
|
Generated using Amazon Nova with the prompt “fans cheering at a baseball game.”Sub-nova. As part of Amazon’s Monday launch of an agent that can take web actions on a user’s behalf, the company also expanded access to its text, image, and video generation models called Amazon Nova — a product limited to select few at its December release, so now we tested it out. To be this late in the model game, the generated results must be exceptional. But in our review of Amazon’s first family of models, that’s not the case. Roughly half the time, Amazon’s image and video generation tools produced distorted faces, arms, and hands; body parts that move through others; or objects that appear out of nowhere. Its image editing tools rarely did what was asked, and often increased the distortion. Results also missed details from the prompt instructions, especially when asked for mythical or supernatural creatures. Amazon Nova hasn’t drummed up much attention online, and maybe that’s a good thing. Aside from viral moments like ChatGPT’s Studio Ghibli-style images last week, AI models aren’t drawing much applause at the moment — just look at the ho-hum attention Gemini 2.5 got last week, despite blowing OpenAI and Anthropic out of the water on benchmarks. Why then is Amazon trying to compete in this world? It’s not really a foundation model company, and it doesn’t need to spend boatloads of cash trying to be. Its AI profits will largely come from selling AI-enhanced products and AWS compute. What matters now is the application layer, and that’s where its browser-control agent could shine. Until models matter again, though, releasing tools that fall behind rivals’ standard performance only hurts Amazon’s brand. |
|
 Dara Khosrowshahi, CEO, Uber, will join top global leaders at Semafor’s 2025 World Economy Summit, taking place April 23-25, 2025, in Washington DC. As the first major gathering since the new US administration took office, the summit will feature on-the-record discussions with 100+ CEOs. Bringing together leaders from both the public and private sectors — including congressional leaders and global finance ministers — the three-day summit will explore the forces shaping the global economy and geopolitics. Across twelve sessions, it will foster transformative, news-making conversations on how the world’s decision-makers are tackling economic growth in increasingly uncertain times. April 23-25 | Washington, DC | Learn More |
|
Steven Ferdman/Getty ImagesThe CEO of Flutterwave, Africa’s most valuable unicorn, didn’t rule out selling to a bigger player in an interview with Semafor’s Alexis Akwagyiram and Alexander Onukwue. The Nigerian-founded global payment processing platform, headquartered in San Francisco, is keeping options open as it focuses on reaching profitability this year, CEO Olugbenga ‘GB’ Agboola said. With a valuation of more than $3 billion, after raising around $500 million since its 2016 launch from investors like Visa and Tiger Global, the company wants to provide “stability” and be sustainable in the long run to achieve its profitability goal. “We want to go aggressive on that as a firm,” Agboola said. Going public is one of the options Flutterwave has been discussing over the past few years, but with a soft market for IPOs, a sale could be a convincing option for investors looking to collect a return from their stakes in the startup. The move wouldn’t be without precedent, Alexis notes: Paystack, a Nigerian Flutterwave contemporary, was acquired by US fintech giant Stripe in 2020 for around $200 million. If Flutterwave can reach profitability soon and then goes down the sale route, it could potentially sell for a lot more based on its size and private market valuations. |
|
|