Altman respects the hustle

DeepSeek: The fast and the furious

Welcome back! Oh boy…Sam Altman praising DeepSeek? Wall Street going bonkers? Microsoft investigating? Feels like we’re in an AI soap opera this week. 🧼 

Let’s dig into the biggest stories. ➡️ 

DeepSeek’s meteoric rise (and the chaos it’s causing)

Via DeepSeek

DeepSeek didn’t just top the App Store this week—it upended the AI hierarchy overnight. Here’s how some of the biggest names in tech are responding:

  • OpenAI CEO Sam Altman acknowledges DeepSeek’s strength. Altman called R1 “impressive” and “invigorating,” suggesting competition really could spur innovation here​.

  • Microsoft is on the hunt, investigating whether DeepSeek violated terms in using OpenAI’s API to train its model​.

  • Cloud security firm Wiz calls out DeepSeek. Turns out, the AI platform left one of its critical databases exposed, revealing more than a million records.

  • Wall Street takes a hit. Nvidia’s market cap shrank $593 billion earlier this week as investors questioned AI chip demand​. But worth noting? Nvidia recouped most of those immediate losses by Friday.

The big picture: DeepSeek is rewriting the AI playbook by proving that billion-dollar investments aren’t the only way to build frontier models. But with questions about data sourcing, API violations, and Chinese regulatory influence, its future remains a question.

I'm curious: Do you think DeepSeek is a paradigm shift, or just another splashy launch?

Click one and tell me more.

Login or Subscribe to participate in polls.

Alibaba flexes with Qwen2.5-Max

DeepSeek isn’t the only Chinese AI powerhouse making waves. Alibaba just launched Qwen2.5-Max, a model that outperforms GPT-4o, Claude 3.5 Sonnet, and DeepSeek R1 in key benchmarks​.

Here’s the breakdown:

  • It’s efficienct. Qwen2.5-Max uses a mixture-of-experts model, achieving top-tier performance with 40–60% fewer computational resources than traditional LLMs.

  • It’s enterprise-ready. Qwen2.5-Max is optimized for business use, excelling in code generation (LiveCodeBench: 38.7%) and advanced reasoning (Arena-Hard: 89.4%)​.

Why it matters: Just a month ago, we thought OpenAI was winning the AI race. It’s only January, and China’s AI push is already setting the tone for 2025. All I can say is…buckle up.

Oumi bets on fully open-source AI

Ex-Google and Apple engineers founded Oumi, an open-source AI platform that removes the typical roadblocks researchers face​ and launched this week. How it works:

  • Fully transparent models. Oumi provides open weights, training data, and methodologies.

  • Modular AI development. Researchers can train on their laptops before scaling up to cloud or university clusters.

Why it matters: The demand for open AI research is massive—Oumi is a direct response to the frustration around locked-down LLMs. It could be the backbone of future AI breakthroughs.

An AI that lives on your desktop

Via Witsy

Witsy is a desktop assistant that provides access to generative AI models from the top AI providers. Basically, it’s a one-stop solution for genAI.

How you can use it: 

  • Write and refine with Scratchpad. Give AI instructions like “add a title” or “rewrite this” without switching windows.

  • Talk instead of type. Use voice chat to interact with AI hands-free.

  • Automate tasks. Set up shortcuts for frequent actions like emailing, copying data, or launching scripts.

Pricing: Free

A second brain for coding

Via Aider

Aider is an AI-powered coding assistant that integrates directly into developers’ workflows. The tool works within your local git repository, meaning it understands your entire codebase and can modify multiple files in context.

How you can use it: 

  • Real-time debugging. Identify and fix errors without switching between your editor and documentation.

  • Code optimization. Get AI-powered suggestions to improve efficiency and structure.

  • Custom model support. Train and integrate your own models for specialized coding needs.

Pricing: Free

Jobs, announcements, and big ideas

  • There was no shortage of crazy NVIDIA announcements out of CES. Here's what excited me most.*

  • Two major Apple Silicon flaws could expose sensitive data.

  • Quartz has been publishing AI-generated news.

  • Mistral’s new model hits 81% MMLU.

  • Join Scale AI as an AI Strategy Consultant to develop solutions for frontier tech.

  • Anthropic is hiring a Content Lead to create technical documentation for developers and AI practitioners.

  • Is this the AI Spotify of the future?

*This is sponsored advertising content.

Dos and don’ts of AI: Watch along as I share what you can and can’t copyright by using AI.

Thanks for reading! DeepSeek has been deep in this week’s news. Are the newcomers starting to unseat the giants? You tell me—hit reply!

—Matt (FutureTools.io)

P.S. This newsletter is 100% written by a human. Okay, maybe 96%.