Good morning. It’s Monday, March 17th.

On This Day In Tech History: In 2016, AlphaGo, an AI developed by DeepMind, won its fourth game against world champion Go player Lee Sedol, clinching the series 4-1 in Seoul

In today’s email:

• Figure’s BotQ Factory

• China’s AI Price War

• Siri’s AI Revamp Delayed

• 4 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Figure’s BotQ factory to scale robot production from 12,000 to 100,000 humanoid units annually

California-based Figure is going all-in on high-speed humanoid production with its BotQ factory, engineered to churn out 12,000 units a year. The facility swaps slow CNC machining for injection molding, die-casting, and stamping—slashing part fabrication from weeks to seconds. Meanwhile, an in-house AI system, Helix, ditches traditional conveyor belts, orchestrating humanoid workers to transport materials and handle repetitive assembly.

With a fully integrated Manufacturing Execution System (MES) tracking every stage, a Warehouse Management System (WMS) optimizing logistics, and Product Lifecycle Management (PLM) ensuring precision, Figure is streamlining production like an industrial symphony. The shift from Figure 02 to the production-ready Figure 03 trims complexity for faster scaling. Read more.

China’s AI price war heats up as Baidu’s Ernie X1 undercuts DeepSeek, plans open-source Ernie 4.5

Baidu just turned up the heat in the AI arms race with Ernie X1 and Ernie 4.5, gunning for DeepSeek and OpenAI with lower costs and bold performance claims. X1 reportedly matches DeepSeek-R1 while cutting API costs in half, while Ernie 4.5 refines logic, code generation, and error reduction—allegedly outpacing GPT-4o and keeping stride with GPT-4.5.

With Qianfan cloud pricing slashing OpenAI’s rates, Baidu is making it harder to justify premium Western models. Ernie 4.5 is also set to go open source in June, following DeepSeek’s strategy of democratizing access while squeezing competitors’ profit margins. Meanwhile, X1’s cutthroat pricing pressures AI incumbents banking on enterprise dominance.

As US firms lobby for stricter regulations on Chinese AI, Baidu is embedding its models across search and consumer products, tightening the competition. Read more.

Siri’s AI makeover delayed—because 80% isn’t good enough for Apple

Apple’s grand AI upgrade for Siri won’t be landing until at least 2026, after internal tests exposed accuracy issues hovering between 66% and 80%. The overhaul, meant to supercharge Siri with deeper app integrations and personal data handling, now risks slipping further as Apple wrestles with scaling generative AI without breaking its reputation for polished user experiences.

To regain momentum, Apple has restructured its AI division, with senior managers now reporting directly to AI chief John Giannandrea. Unlike OpenAI and Google, which release AI features with known flaws, Apple prioritizes high reliability, complicating its rollout across hundreds of millions of devices. Meanwhile, competitors continue to iterate rapidly, intensifying pressure on Apple to deliver a functional, competitive AI assistant without compromising user trust. Read more.

• Google and Meta’s ‘open’ AI model licenses stir concerns over control and compliance

• ​​Google expands AI stack with Gemini upgrades and Chirp 3 integration

• AI talent wars intensifies as elite research scientists command fortune-level bids

• Y Combinator startups are fastest growing, most profitable in fund history because of AI

• AI at SAP: "The real challenge lies in transforming your own work"

• OLMo 2 32B sets a new standard for true open-source LLMs with public code, weights, and data

• “I currently believe leading AI labs are on track to have the first fully automated researcher prototypes by the end of 2025.“ - Herbie Bradley on 𝕏

• Anthropic’s plan to win the AI race

• "This is the year that AI gets better than humans at programming forever" - OpenAI CPO Kevin Weil

📄 PLADIS: Pushing the Limits of Attention in Diffusion Models at Inference Time by Leveraging Sparsity

📄 ReCamMaster: Camera-Controlled Generative Rendering from A Single Video ​​

📄 Charting and Navigating Hugging Face's Model Atlas

📄 Adversarial Data Collection: Human-Collaborative Perturbations for Efficient and Robust Robotic Imitation Learning

📄 API Agents vs. GUI Agents: Divergence and Convergence

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!