The Humans in the Loop: DeepSeek Aftermath
This Week in AI for Devs: What DeepSeek Means for AI
Welcome to The Humans in the Loop, your executive summary of AI news for devs. I’m Andrew, the [human] author, and I’d love your feedback on this newsletter. Please feel free to email me at andrew@heavybit.com with your thoughts. Thanks!
Our top story: Life Post-DeepSeek R1
As we covered last time, newcomer model DeepSeek R1 made waves with reasoning abilities that rival that of OpenAI’s o1 model—but as an MIT-licensed open-source model trained at a fraction of the cost. As you already know, the news sent shockwaves through the media and the markets. NVDA’s stock took a significant hit after DeepSeek’s impressive performance on modest hardware resources cast doubt on the need to invest heavily into the chip leader’s priciest products. It also caught several of the top AI vendors scrambling, though they’ve apparently come back in force. GOOG has announced plans to invest $75B in AI this year (presumably to compete with MSFT’s planned $80B AI budget). OpenAI has since revealed several new products, including its impressive DeepResearch agent, which performs research for users, as well as government- and education-focused versions of ChatGPT (the latter of which will be adopted by an estimated 500K students at Cal State University). It has also struck an eye-watering $3B deal with Softbank’s Masayoshi Son—with the investor claiming that AGI will arrive “much earlier” than anticipated. MSFT, meanwhile, has welcomed DeepSeek with open arms, adding support for R1 to its Azure Cloud platform. The US Federal Government, viewing DeepSeek as a competitive and security threat from China, has been less receptive, with the House introducing a bill to ban it. While DeepSeek definitely made an impact, there was already a great deal of money in the ground and deals already in motion. This rollercoaster isn’t close to coming to the end of the tracks.
Up next: The bottom line in AI news for busy devs.
💻 Development
How a GH Engineer Uses LLMs
GitHub engineer Sean Goedecke explains his daily use cases: Code completion for faster prototyping and faster adoption of new languages.
DeepSeek In-Depth: Illustrated, Demo vs. o1, plus FAQ
ML expert Jay Alammar explains DeepSeek’s reasoning; DeepSet’s demo shows the models’ performance side-by-side; Stratechery also covers FAQ.
Replit’s AI Agent Now Free and on Mobile
Founder Amjad Masad makes it official: Replit’s AI app builder is now free-to-use, and now has a mobile version as well.
[GH] smolmodels - Build ML Models in Natural Language
This new project lets users “build ML models using natural language and minimal code.” Maybe we didn’t need that ML PhD after all.
🤔 Interesting AI Projects, Research, and Updates
GOOG DeepMind Team: Understanding Model Scaling
This extensive article explains how to think about model scaling vs. resource and memory constraints.
Research: UI-TARS - Open-Source Computer-Use Model
Thinking about computer-use applications? This project resulted in a tweaked model that does mouseclicks and keypresses. Try it here.
Course: How Transformer LLMs Work
DeepLearning’s latest free course explains transformer-based architecture.
Research: Vintix - Improved Decision-Making
This research explores in-context reinforcement learning as a method for creating robust decision-making systems.
💼 Hiring and Community
Startups Hiring This Week:
- Member of Technical Staff, AI → Coframe
- Chief Cloud and AI Architect → Finrec.io
Mid-Markets Hiring This Week:
- Software Eng, AI Team → Volley
- AI Robotics Research Eng → Nimble Robotics
Enterprises Hiring This Week:
- Principal ML Scientist → DNA
- Principal Software Eng, AI Data Platform → GM
Sighted in the Wild: First Job Listing for an AI Agent
Humans need not apply.
💡 Spotlight: Newly-Launched AI Startups
- Observo AI: $15M from Lightspeed for agentic data pipelines
- Prior Labs: $9.3M from Balderton for AI-driven tabular data processing
- Doti: $7M from F2 for AI-powered internal knowledge bases
- Riley: $3M from Vertex for AI-powered product insights
- Hestus: $1.5M from Liquid2 for AI-CAD
🏭 Industry: M&A, Launches, Trends
GOOG Brings Reasoning to Gemini
GOOG is the latest vendor to break into reasoning, bringing the “Gemini 2.0 Flash Thinking” update to the latest models in the Gemini family.
AMZN Tackles Hallucinations With Automated Reasoning
The Cloud giant is looking into automated reasoning—which uses math proofs to verify outcomes—to resolve hallucinations for AWS customers.
MSFT Poaches NotebookLM Scientists
Perhaps showing how limited the AI talent pool still is, MSFT AI CEO Suleyman nabs former colleagues Marco Tagliasacchi and Zalán Borsos.
⚖️ Copyright, IP, Licensing, and Regulation
US Copyright Office: No Protections for Non-Human AI Content
The USCO confirms it will not offer copyright protection for any content created 100% by AI without human input.
CA SB-243: AI Guardrails to Protect Children
State legislators propose bill to provide anti-addiction and anti-suicide guardrails to AI products to protect kids. Probably a good idea.
GOOG Calls for Public-Private NatSec, Quietly Erases Weapons Pledge
The AI giant proclaims the need for more Stargate-like public-private AI partnerships for national security. It also quietly recants its pledge to not use AI for weapons or surveillance purposes.
About the Author
Hi there. My name is Andrew, and I work at Heavybit, the leading VC for developer-first startups. As the Editorial Lead, my goal is to find the most valuable and important AI news for developers and founders. The idea is to curate and bottom-line emerging trends from our perspective in 10+ years of coaching and funding developer-facing companies. Email me and let me know what you think.
That’s all for this edition. In the meantime, please feel free to follow The Humans in the Loop on Twitter and LinkedIn.