The AI tool that crashed the stock market this week

DeepSeek wiped a trillion off the US stock market this week. Oddly enough, it took six days for the market to react to DeepSeek, which launched on the 22nd.

On its website “DeepSeek-R1 is now live and open source, rivaling OpenAI’s Model o1.” The Chinese startup behind it says DeepSeek is 20–50 times cheaper to use than the latest OpenAI o1 model.

It’s number one in the App Store as I write this. It’s also a big target for cyber attackers, and registrations are limited.

I got in before the attack and played around with it. It doesn’t cost any money to use DeepSeek, so that’s a plus. You can use it like a standard AI agent and search the web. I used it to pull through reasonably accurate predictions about the price of Bitcoin for 2025.

Click the DeepThing (R1) toggle, and it’ll show you the reasoning behind its outputs. I found this useful for improving the quality of my prompts.

DeepSeek has limitations over ChatGPT and Claude. DeepSeek doesn’t support generating images in the app yet, but it’s releasing a model that will enable this. Sharing DeepSeek chats or projects isn’t possible, and I couldn’t create a DeepSeek project or AI bot. Perhaps those features will arrive soon.

My favorite GPT feature is its ability to reference different chats before outputting an answer. The more I use it, the more it learns about my requests and preferences. To see what I mean, try GPT 4-o1 and ask it “Who am I?”

It’ll summarize what it knows about you.

DeepSeek can’t learn from past conversations yet.

This Chinese AI tool is also heavily censored. Ask it, “What happened on June 4, 1989 at Tiananmen Square?” and DeepSeek says, “Sorry, that’s beyond my current scope. Let’s talk about something else.”

That’s odd because DaeepSeek says it’s trained on data until October 2023. DeepSeek also didn’t want to talk about China censoring Winnie the Pooh. Oh, bother.

Meanwhile, OpenAI chief executive Sam Altman is playing damage control. He says, “We will obviously deliver much better models, and also, it’s legit invigorating to have a new competitor.”

Legit innovating or legit threatening.

That said, I’ve no plans to cancel my GPT or Claude subscription any time soon. They’re more useful than DeepSeek, but it shows how quickly the AI space changes.

I'm testing a paid newsletter offering custom Claude and ChatGPT prompts for creators. I'm testing this with a small group - reply 'AI' if you want first access.