Previously on AI - Episode 3

DeepSeek’s Impact on the AI Game

DeepSeek is taking up all the headlines this week with the release of their model that competes with OpenAI’s top models with a significantly smaller (reported) training budget of $5.6 million and maybe a little alleged subterfuge.

You can find think pieces everywhere about how DeepSeek is going to blow up NVIDIA’s game since, theoretically, models can be trained using less of NIVIDIA’s chips, or how DeepSeek is going to lift up all ships and create an even bigger boom for AI companies.

My theory, on vibes alone, lands somewhere in the NVIDIA will probably continue to grow and gain back some of its 17% loss after the initial DeepSeek news hit. The news hasn’t seemed to impact Mark Zuckerberg’s wallet.

It is still an interesting look at how the chip restrictions the U.S. has put on China has led to adaptations that allowed for comparative models to be built with smaller budgets and less resources. Constraints can lead to creativity and maybe a little corporate hijinx.

🤿 Dive Deeper

• Stratechery’s great explainer

• DeepSeek R1 on HuggingFace

• HuggingFace’s Open-R1 Reproduction of DeepSeek-R1

The Name’s GPT … ChatGPT

DeepSeek has, at least temporarily, replaced AI Agents as the biggest buzzword of the AI moment. But what exactly is an AI Agent? Are we dressing up a LLM in a suit and tie and having it take down international criminals?

AI agent is essentially used to indicate AI based tools that can help to automate various tasks over multiple applications. Where you might tell ChatGPT to write you a blog post, an AI agent could be used to generate a blog post, navigate your browser to MidJourney to create an image, then login to your website and publish it.

Its really just fancy marketing speak for us dumdums to understand AI tools are going to be doing more things across different platforms using pre-defined workflows or winging it to buy your groceries.

🍸 Agentpalooza

• OpenAI announced its Operator preview for Pro users

• Claude Sonnet Announces Computer Use

• Block releases Goose, an open source AI Agent

• Anthropic on building effective agents

Links

📚 Read - DeepSeek Disruption Has Its Upside
Matt Levine has an interesting take on disruptors ability to fund themselves by shorting the thing they’re disrupting. Did DeepSeek do that? 🤷‍♂️ 

📺 Watch - Inside Anthropic's Race to Build a Smarter Claude and Human-Level AI
Pre-DeepSeek listen to Anthropic’s CEO talk about the future of Claude and AI.

🎧 Listen - Acquired Podcast’s History of Nintendo
The news on DeepSeek’s innovation in spite of constraints reminded me of this podcast episode covering the history of Nintendo. Including how the constraints for the original NES led them to innovate in surprising ways.

Term of the Day

Next Time On

In upcoming newsletters I plan to dive into coding assistants like GitHub Copilot.

What interests you? Hit reply to share your thoughts on this and future topics.