Welcome to the Women Who AI Newsletter

Hello there, and welcome to the first entry of the Women Who AI Weekly Newsletter!

We often hear that it’s difficult to stay on top of everything in AI. How do you know what to pay attention to and how it’s relevant to your work?

We hope this newsletter helps. The team behind Women Who AI is made up of startup founders and engineers, here to separate signal from noise in this rapidly evolving field for women founders building and scaling startups.

Look out for us in your inbox every Monday morning (or unsubscribe at the bottom of this email if you’re not interested in these; no hard feelings!)

This Week’s “Deep” Dive: DeepSeek-R1

DeepSeek-R1 made a lot of noise last month. What's actually going on here?

First, if you haven't already, you can try out DeepSeek-R1 here (free, hosted on US servers). Watching it <think> is quite cute and helps us learn to use other reasoning models, such as GPT-o1, more effectively.

Let's cut through some of the hype. The buzz around DeepSeek comes from its ability to match GPT-o1's performance on many benchmarks while using significantly less computing power to train and run. This efficiency breakthrough comes from their innovative use of Reinforcement Learning—where instead of just predicting the next word in text, the model tackles challenging tasks like writing Python code or solving math problems. It receives immediate feedback (positive points for correct answers, negative for mistakes), and after countless iterations combined with traditional language model training techniques, we get DeepSeek-R1.

This is a good thing! And contrary to some rumors, DeepSeek didn't emerge from a side project. It was developed by a team of established mathematicians and researchers who have been publishing related work for years—all documented in the DeepSeek-R1 paper.

What makes DeepSeek particularly interesting is that when using the open-source version, you're interacting directly with the model itself. With Claude or GPT, there might be layers of programming mediating your interaction—you can't really know. But with the open-source DeepSeek (not the hosted version on their site), that ambiguity disappears.

Can I run DeepSeek on my laptop?

Absolutely! If you've got a bit of time and curiosity, you can run DeepSeek locally using ollama. Follow this tutorial to get it up and running on your own computer in under 15 minutes!

What about censorship?

Yes, this is something you can verify yourself. Perplexity dives into this topic thoroughly in their blog post announcing their open source "unbiased" version called R1-1776. Check it out here to learn both about the original DeepSeek-R1's biases and the steps Perplexity took when creating their open-source alternative.

An interesting workaround comes from the Sundai Project on bypassing censorship through language switching. Their key insight: translating user queries into Russian before sending them to DeepSeek results in more permissive responses, as the model applies stricter content filters to English inputs. Learn more here.

Want to get in the weeds?

Start with the original paper. If you're not accustomed to academic publications, here's a pro tip: skip the mathematical equations on your first read-through to grasp the key concepts. The DeepSeek paper is surprisingly accessible and, at times, genuinely delightful, like in the Table description below.

At a hackathon last weekend, I followed this tutorial to fine-tune DeepSeek on new data. Reach out to Strong Compute for access to their systems to work through the tutorial yourself.

For the truly curious, Hugging Face is developing a fully open version of DeepSeek-R1 from scratch. What makes this special? You can explore the code and data at every stage and even contribute directly to the project.

New Models! Claude 3.7 and GPT-4.5

The AI landscape continues to evolve rapidly with two major model releases in late February 2025. As women founders in AI, understanding these tools can help you leverage them effectively, without getting caught in the hype cycle.

What's New and Why It Matters

Claude 3.7 Sonnet (Feb 24) and GPT-4.5 (Feb 27) bring different strengths to the table:

• Claude 3.7 Sonnet can now show its thinking process before answering (look for the cute <think> animation). This makes it especially good at tackling complex problems and writing code. Plus, it comes with a new terminal tool called Claude Code for developers.

• GPT-4.5 feels more natural in conversation with better emotional awareness. It's noticeably better at understanding what you're asking for and responds more concisely. It's particularly strong at creative writing and handling nuanced requests.

Practical Differences for Founders

Beyond the Marketing Hype

The most important thing to remember: these tools serve you, not the other way around.

Rather than getting overwhelmed by each new release, consider your specific needs:

• Need help building or debugging code? Try Claude 3.7 with its coding improvements.

• Creating customer-facing content or conversational interfaces? GPT-4.5's enhanced EQ might be worth exploring.

• Working on a tight budget? Claude 3.7 maintains the same pricing as previous models and offers a free tier.

The best way to understand these models is to experiment with them on your own specific tasks.

Remember that while these models are impressive, they're tools to enhance your work—not replace your expertise and vision as founders. Use them strategically when you need them, rather than getting caught up in every new release.

Further Reading

• Full Claude 3.7 Sonnet announcement

• Full GPT-4.5 announcement

Upcoming Hackathons

Hackathons are a great place to activate a team and build the product you've been dreaming about. If you're interested in joining a Women Who AI Team for one of these events, reply directly to this email, and we'll send a follow-up connecting everyone interested.

• Virtual, March 7-9: Prompt to Product Hackathon - Register here

• SF, Saturday, March 8: Hack #1: Personal AI Assistants - Register here

• Boston, Sunday, March 9: Sundai Club Financial Inclusion - Apply ASAP at Partiful

• NYC, Saturday, April 20: Build for Your Friends Hackathon - Register here

Founders Hiring Now

Women's Health AI Startup (MIT) — Looking for an AI Engineer to revolutionize women's health through predictive analytics and digital health solutions. Ideal candidates should have experience building end-to-end AI systems, strong skills in computer vision and multi-modal models, and passion for healthcare innovation. This paid opportunity offers the chance to build impactful solutions addressing long-overlooked health challenges for women, with potential to join the founding team.

Interested? Email [email protected] with your resume to apply.

About Women Who AI

Women Who AI emerged from countless conversations in our MIT WhatsApp group where we found ourselves sharing the same resources, answering similar questions, and collectively navigating the unique challenges women face in the AI space. We decided it was time to formalize these exchanges into something that could benefit our broader community. Whether you're building your MVP or scaling an established AI business, our goal is to provide you with information you can actually use—no hype, no jargon, just valuable insights that help move your business forward.

Reply With Questions

We want this newsletter to address the real challenges you're facing. Is there a specific AI development you'd like explained? Jargon we included but didn't properly explain? A business problem you're trying to solve with AI? Or perhaps a funding question specific to women founders? Reply directly to this email with your questions, and we'll tackle them in next week's edition.

Share the Knowledge

If you found value in today's newsletter, please consider forwarding it to other women in your network who are building, or thinking about building, in the AI space. The more we grow this community, the stronger our collective impact becomes.

Here's to building the future of AI, together.

Lea & Daniela