el producto #425 🚀
Alexa+, How to use LLMs and AI prototyping, Goodbye Skype, New models: GPT 4.5 and Claude 4.7, Podcasts soaring on YouTube, Ad-supported Office, Meta's AI app & more
Hi folks 👋
Welcome to a new edition of el producto
🎰 The week in figures
$91.5B: Stripe’s valuation climbs to $91.5B in new tender offer as existing investors buy shares from employees, offering them liquidity while the company delays IPO plans. This marks a jump from last year’s $70B valuation but remains below its 2021 peak of $95B
$60B: Revolut is facing investor pressure to authorize a new secondary share sale at a $60B valuation, significantly up from $45B just six months ago
$35B: Meta in talks to raise $35B for more data centers
$1B: Thinking Machines Lab, a new startup by former OpenAI CTO Mira Murati, reportedly aims to raise its first $1B at a $9B valuation
1B: More than 1B people are now watching podcasts on YouTube every month
$480M: Lambda, the AI cloud computing firm raised a $480M Series D at $2.5B
$100M: Mercor, the AI recruiting platform raised a $100M Series B at $2B valuation
$100M: Genspark, the AI search engine startup raised $100M at $530M valuation. Their LLM-powered search serves 2M users with Deep Research features
+78%: Nvidia released their latest earnings report on Wednesday, besting Wall Street’s estimates for revenue with $39.33B. Sales are up 78% and still rising as the company remains at the center of the AI accelerator race. Nvidia anticipates $43B in Q1 revenue. That’s around 65% YoY growth. All eyes are now on the company’s next-gen AI processors, known as Blackwell
📰 What’s going on
OpenAI has released a research preview of GPT-4.5, its largest and most advanced language model yet, which reportedly outperforms previous models in natural interaction, knowledge, reasoning, and hallucination reduction. GPT-4.5 was trained using "new techniques for supervision" combined with supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). GPT-4.5 is available to ChatGPT Pro users on web, mobile, and desktop, with a rollout to Plus, Team, Enterprise, and Education users in the coming weeks
Anthropic, has released the industry’s first “hybrid AI reasoning model”—Claude 3.7 Sonnet—which is designed to complete “real-world tasks” (like complex coding or creating legal documentation), and lets users choose whether they want the model to give quick, real-time answers to their questions, or think for longer and deliver more considered, ‘thought-out’ responses. Anthropic has also released a “limited research preview” of an agentic coding tool—Claude Code—pitching it as an “active collaborator that can search and read code, edit files, write and run tests, commit and push code to GitHub, and use command line tools,” as it moves towards providing users with a more simplistic experience—essentially building one model that can do it all. Anthropic shared a roadmap of their three-stage vision for Claude:
Perplexity has revealed that it’s working on an AI-powered web browser—called Comet—which will be able to perform “agentic searches.” The browser will reportedly be able to search the web in real-time, locate information, understand more complex instructions, perform tasks, and make autonomous decisions
Grok 3’s new thinking mode can accidentally leak the model’s system prompt, and it’s created quite an awkward situation for Elon. Users discovered this when asking who spreads the most misinformation on X—and the AI kept identifying its own creator before having an existential crisis about whether it was allowed to say so
Google’s Gemini coming to iOS? Reporter Aaron Perris spotted some interesting code in Apple’s iOS 18.4 beta, suggesting that a Google Gemini integration may be coming soon. A line-up of Apple Intelligence features are expected to debut in iOS 18 or 19
Meta plans a standalone AI app for Q2 2025 to compete with ChatGPT
Meta also announced Aria Gen 2 glasses, an upgraded research device with advanced sensors that enables researchers to explore machine perception, contextual AI, and robotics applications
Baidu is set to launch the next generation of its AI model, Ernie 4.5, in mid-March, promising enhanced capabilities in reasoning and multimodal processing. Baidu CEO Robin Li acknowledged that DeepSeek's popularity prompted the company to shift towards an open-source approach, with plans to make Ernie 4.5 open source from June 30
Amazon showcased Alexa+ with generative AI and smart home features, a "complete re-architecture" of the AI assistant as more of a serious productivity tool. Expected to begin rolling out in early access this month
Apple prepares to launch a foldable iPhone alongside another innovative design in 2026, with the foldable device reportedly nearing final development stages
Microsoft is quietly testing a free, ad-supported version of Microsoft Office for Windows. Users could access PowerPoint, Word, Excel, and other classic apps for free, so long as they’re willing to tolerate a persistent banner ad along the side of their screen and occasional 15-second commercial interruptions
Goodbye, Skype: Recall the last time that you used Skype. Has it been a minute? If it has, you’re not alone. So much so that Microsoft is killing off the venerable voice and text service later this year. Teams will replace it
Instagram is considering launching a standalone Reels app. As AI makes it easier than ever to build features - this risks feature bloat. Could app unbundling be one way to address this?
Waymo accelerates: Alphabet’s self-driving project now serves 200k paid rides each week. That’s still a fraction of Uber’s volume — which is in turn a fraction of the larger market — but still a 33% gain since we last heard from Waymo towards the end of 2024. Waymo is adding markets and capacity as Tesla gears up for its own competing service
Crunchbase relaunched as an AI-powered prediction engine, moving away from historical data to provide users with predictive intelligence on private companies' growth, funding rounds, and acquisitions
Trello is getting a major update that transforms it into a personal to-do task app with a new inbox that lets you organize all of your to-dos in one place
Quora’s Poe intros AI-generated apps: Poe’s new App Creator Tool allows users to generate a new app based on a simple text description, including basic specs and the AI models on which they want the app to rely (for example, OpenAI’s o3-mini or Google’s Veo 2.) The tool is powered by Anthropic’s Claude 3.7 Sonnet model. Poe Apps can run as a pure visual or alongside a chatbot window; the code allows for manual adjustments and tweaks as well
📚 Good reads
How to use LLMs: Andrej Karpathy released a new video in his “general audience” series on language models and how to use them, with over 15 tips for prompting and best practices when using AI tools
AI prototyping masterclass for PMs: step-by-step guide of Bolt, Lovable, v0, Replit, and Cursor. A video tutorial building an app with features, live, right in the podcast, by Aakash Gupta and Colin Matthews
Are AI agent marketplaces the future (for real?) Gigi Levy-Weiss explores how AI-driven agents could reshape marketplace dynamics. The piece assumes that marketplaces will remain defensible through differentiation and trust. However, if generalist AI agents can dynamically optimize across services, traditional moats—like branding, user lock-in, and integrations—may erode faster than expected. Instead of reinforcing existing models, AI could force companies to constantly re-earn their place in the ecosystem. In any case, as always with NFX, this is a thought-provoking read
Deep Research in the age of AI: knowledge as a competitive edge. Information wants to be free. But in an AI-driven world, who benefits from knowledge, and how do we retain its value? Ben Thompson explores the tension between AI’s ability to aggregate, synthesize, and commoditize information and the growing attempts to build walls and toll gates around proprietary research. As AI models grow more powerful, some will double down on proprietary knowledge and exclusive access, while others will embrace a more open, decentralized approach
What is vibe coding? Developers are embracing a new way of building software that’s more conversation than code. Claude 3.7 may be its bigger enabler so far
That’s a wrap for this week! 🌟
I’d love to hear your thoughts—what stood out to you, and how are you thinking about integrating these insights into your product strategy? Reply to this email or drop a comment on Substack to share your take. And if you found this valuable, forward it to a fellow PM or product enthusiast who’d enjoy the read
See you next week! 👋
Angel