😺Monday’s AI crash, explained

Welcome, humans.

This is hilarious: apparently, DeepSeek R1 knows how to “think” in the style of Donald Trump.

Funny enough, DeepSeek got Donald Trump’s attention for a different reason yesterday—it more or less crashed tech stocks (more on that in the main story). In response, Trump said DeepSeek is actually “very much a positive development” for AI, because “instead of spending billions and billions, you’ll spend less.”

I wonder if the Whale Bros had “Donald Trump actually knows who we are” on their bingo card for 2025…

Here’s what you need to know about AI today:

• DeepSeek hype caused a mini-tech sell off.
• Apple Intelligence now default on new devices.
• Meta AI got conversation memory on messaging apps.
• Trump announced Taiwan chip import tariffs.

Advertise in The Neuron here.

DeepSeek accidentally crashed the AI party (for a day, anyway)…

It was the best of AI times, it was the worst of AI times… for the stock market, anyway.

Here’s WTF happened yesterday. The market had a mild panic attack after DeepSeek showed you could build OpenAI-level AI for ~5% of the cost:

• NVIDIA crashed 17% in a single day and lost ~$600B in market cap (single-biggest day drop in U.S. history!).
• CEO Jensen Huang's net worth personally declined by ~$21B.
• The tech-heavy Nasdaq fell 3.1% + many AI-heavy stocks fell >8.7%.

Oh, and just to twist the knife, DeepSeek also overtook ChatGPT as the #1 free app in Apple’s App Store. Our favorite reaction? “I can't believe ChatGPT lost its job to AI.”

Then, because apparently crashing the market wasn't enough excitement for one day, DeepSeek casually dropped Janus Pro 7B, a new vision model that can understand and create images. Talk about a Monday.

Don’t worry about Jensen; he can wipe HIS tears with his hundred other billions

The latest victim of DeepSeek’s success? DeepSeek—it had to temporarily limit new users due to “large-scale malicious attacks.”

Analysts had a wide range of reactions to the news, from Morgan Stanley declaring the end of the “bigger is better” era in AI to Bernstein dismissing the “weekend Twitter panic” as overblown.

This saga will henceforth be known (to us) as the battle between the “Scale Bros” and the “Whale Bros.”

The “Whale” case:

1. Companies overspend on expensive GPUs instead of solving key technical challenges—the AI industry's “fix it in post.”
2. Smaller teams can can achieve impressive feats within constraints (200 people vs thousands), using 1/5 the GPUs at 1/50th the cost.
3. If the barriers to entry are 50x lower (or 95% lower, in the case of training), the prices you can sell your services at will be even lower than that—and innovation will thrive.

The “Scale” case:

1. Sam says “the world is going to want to us A LOT more AI”, and AI integrating into agents, robotics, and “physical AI” will need plenty of compute.
2. Jensen Huang notes that even if training gets cheaper, inference (actually running the model) “requires significant numbers of NVIDIA GPUs."
3. Consider DeepSeek's achievements with ~2,000 GPUs—imagine what Sam could do with his $500B Stargate project's estimated one million GPUs.

The likeliest case: both bros have valid points. Gary Marcus celebrated “there is no obvious moat” for the Scale Bros, and claimed that the race for AI supremacy is over—for now. Google more or less admitted the same back in 2023.

But “no moat” doesn't mean “no value”—it means value shifts from owning the technology to owning the user experience. Alex Karp, CEO of Palantir, says AI models are commodities; it's what you do on top of them that counts.

That might be where ChatGPT has created the most value to date. OpenAI can always build new models (or even, gasp, use open-source ones), but that 300M monthly active user base, brand trust, and ecosystem of tools they've built? That's much harder to replicate.

Interestingly, the companies that didn’t sell off today were the ones like Salesforce and Adobe whose products benefit from cheaper AI.

Remember, this is not investing advice—for educational use only!

FROM OUR PARTNERS

Here’s one AI tool we use every week at The Neuron.

Noah rambling… probably

A little well-kept secret for companies crushing it with AI is that only a handful of AI tools are actually worth using.

That’s why we—Noah & Grant—use ChatGPT and Claude, along with a killer product called Attention.

Here’s how it works:

1. We tell Attention exactly what to pay attention to in our meeting (goals, budget, etc).
2. Attention listens in to our call.
3. After, Attention outputs important insights, action items, and follow-up emails.

There are way, way more features, and it’s a must-have for all sales teams. (If you sell anything, literally drop what you’re doing and go book a demo with Attention ASAP).

Prompt Tip of the Day

Julian Goldie released this 20 minute guide for how to use DeepSeek R1 to build applications across three different methods.

Treats To Try.

1. Perplexity now lets you use DeepSeek R1 in your web searches; just click the Pro toggle and select R1 (daily limits apply)—demo.
2. Besides DeepSeek, there’s also Qwen2.5VL, a new family of open-source “agent” models from Alibaba that can control your screen like Operator—chat with it here.
3. Kimi is another “o1-level” Chinese AI model competing with R1 that can search the web and analyzes up to 50 files at once (like PDFs, documents, images), completely free with unlimited usage (github).
4. Add Me is a feature for Pixel phone users that lets you combine two photos of the same scene—this blog explains how it works.
5. Co.dev turns your app ideas into working websites by writing and deploying the code for you—and it automatically fixes errors and updates your database as you build.
6. Llamao lets you chat with an AI on your phone like you would ChatGPT but working completely offline to protect your privacy (free version runs on LLama 3.2 1B right now).
7. Artlas helps you plan museum visits and identify artwork by snapping photos, while a guide (Metti) answers your questions about any piece you encounter.

See our top 51 AI Tools for Business here!

Around the Horn.

• President Trump announced the US will place tariffs on semi-conductors imported from Taiwan in the near future.
• Apple Intelligence is now on by default for all new users with supported devices—except for AI summaries for news apps.
• Meta AI can now personalize its responses by remembering your conversations with it on WhatsApp + Messenger.

FROM OUR PARTNERS

Only 9% of workers are AI proficient enough to see real time savings. Where does your team stand?

Download Section School's AI Proficiency Report to benchmark your organization's AI readiness and get actionable insights for 2025. Plus, discover why company policies might be holding you back. Download for free here.

Under the Hood

1. Here’s a wild use of Deepseek: you can extract its reasoning and apply it to any other model (like GPT 3.5) via the deepseek-reasoner API—here’s how.
2. Onit lets you chat with AI models anywhere on your Mac, with support for local processing and multiple providers—currently Mac-only and early stage, so may be buggy (open-source ver).
3. Llama.vim is a plugin that provides AI-powered code completion similar to GitHub Copilot, but running locally on your computer (Neovim, VSCode).

A Cat's Commentary.