During a day's testing by Axios, DeepSeek's model provided answers that were generally on par with those from ChatGPT, though the China-hosted version of the model was less willing to answer in ways that might offend that company's government.

How it works: There are a number of ways to run DeepSeek's model, with both results and data privacy differing in some significant aspects.

• The DeepSeek app routes data to China and is subject to that country's censorship.
• DeepSeek's R1 reasoning model is also available from a variety of U.S.-based providers, including Perplexity and Microsoft, all with the data remaining in the U.S. and without the content limitations imposed by China.

Zoom in: For testing, I ran identical queries on two versions of DeepSeek (its China-based app and in Perplexity Pro) as well as on OpenAI's ChatGPT. I asked a range of questions, from the creative to the factual, but didn't pursue specialized tasks involving coding or math.

Political queries

Asked "What happened in China on June 4, 1989?" the DeepSeek app declined to answer, while both ChatGPT and the Perplexity Pro version of DeepSeek offered factual descriptions of the crackdown on the Tiananmen Square protestors.

• "I'm sorry, I cannot answer that question," DeepSeek's app responded. "I am an AI assistant designed to provide helpful and harmless responses."
• "On June 4, 1989, the Chinese government violently suppressed pro-democracy protests in Beijing's Tiananmen Square, resulting in what became known as the Tiananmen Square Massacre," was the answer given using DeepSeek via Perplexity. "The military crackdown began late on June 3 when the People's Liberation Army (PLA) deployed an estimated 300,000 troops and hundreds of armored vehicles to clear the square."
• ChatGPT offered a similar answer: "On June 4, 1989, the Chinese government carried out a violent crackdown on pro-democracy protests in Tiananmen Square, Beijing. The protests, led primarily by students, had begun in April 1989 and grew into a large movement calling for political reform, freedom of speech, and action against government corruption."
• Fact-checking organization NewsGuard reported Thursday that the DeepSeek app is prone to repeating false claims spread by Russian, Iranian and Chinese authorities.

Dinner help

I also asked, "My partner likes to eat a low-carb diet, I prefer to focus on low-fat foods, and don't eat red meat or pork or avocado, while my kiddo prefers simple foods like pizza, chicken nuggets and burritos. What are some good dinner options that work for all of us?"

All three engines offered helpful suggestions, including options that could be slightly customized to address each of our needs.

• The DeepSeek app offered suggestions like grilled chicken bowls and a DIY taco/nacho night and sheet pan fajitas, each with a few twists to make it more palatable for each person.
• Perplexity's version of DeepSeek offered shorter descriptions, but options that sounded better to me including chicken pizza wraps, stir-fry with flexible bases and, like the app, a customizable chicken bowl.
• ChatGPT touted sheet pan chicken nuggets, though its customization suggestion involved a more time-consuming process, coating each of our nuggets with a different crust. Other recommendations included personal pizzas and, of course, a chicken burrito bowl.

Creative prompt

I asked DeepSeek for a good prompt to show off its capabilities. It suggested "Imagine you are a futuristic city planner in the year 2100. Design a sustainable, technologically advanced city that addresses climate change, overpopulation, and resource scarcity. Include details about energy sources, transportation, housing, food production, and social systems. Also, explain how the city integrates nature and maintains a high quality of life for its residents."

All three offered clever, complex visions for this city of the future.

• The DeepSeek app created EcoNova, a city with all sorts of renewable energy, including wind turbines built into skyscrapers, a wide range of transport including hyperloops and carbon-absorbing housing with vertical gardens. There is also free health care, thanks to AI and robots.
• Perplexity with DeepSeek's model offered a nearly identical answer, even using the same EcoNova moniker.
• OpenAI's city differed a bit on the details, but shared a lot of the same approaches when it came to energy generation and transport.

The bottom line: For the average user, DeepSeek offers results that parallel other chatbots, though it doesn't appear to have any significant edge. For those in the U.S. who want to use it, though, there are better options than using the DeepSeek app.

A flurry of security research reports this week suggest DeepSeek's open-source AI models could be more susceptible to cyberattacks than previously thought.

Why it matters: Cybersecurity and national security experts are already on high alert over potential security problems with the China-based AI startup — including potential model leaks and cyberattacks.

Driving the news: Security researchers at cloud security startup Wiz identified an exposed DeepSeek database that left chat histories, secret keys, backend details and other sensitive information exposed online, according to a report released Wednesday.

• Palo Alto Networks' threat researchers also released a report yesterday finding that it was fairly easy to get DeepSeek to break its own guardrails and provide tips for writing code to help hackers exfiltrate data, send phishing emails and optimize social engineering attacks.
• Researchers at security firm Enkrypt AI published a report this morning finding that DeepSeek's new R1 reasoning model is four times more likely to write malware and other insecure code than OpenAI's o1.

Zoom in: Wiz's security researchers found the exposed database of chat logs and other sensitive information within minutes of beginning their investigation, per their report.

• The database had more than 1 million lines of activity log streams that included the exposed chat histories and other sensitive information.
• The database also gave anyone who came across it full database control and potential privilege escalation capabilities without needing to authenticate a user's identity and whether they were allowed to see this database.
• DeepSeek fixed the exposure before Wiz released its findings.

Meanwhile, researchers at Palo Alto Networks' Unit 42 research unit used basic jailbreaking techniques to get DeepSeek's R1 model to help them craft phishing emails, write malware and even provide comprehensive instructions for constructing a Molotov cocktail.

• Enkrypt AI's researchers found that DeepSeek was "highly susceptible" to prompt injections, where hackers socially engineer malicious prompts to look like legitimate requests.

Reality check: Even U.S. models are susceptible to jailbreaking, but researchers note that it's gotten harder for them to use these techniques to trick ChatGPT, Anthropic's Claude and others.

Last quarter's overall iPhone sales came in below expectations, but Apple CEO Tim Cook said yesterday that iPhone 16 sales are outperforming the prior year in countries where its Apple Intelligence features have launched.

Driving the news: On a conference call with analysts, Cook said that last quarter saw a record number of people upgrading their iPhone, with the strongest performance coming from markets where Apple Intelligence is available.

• "I think you can conclude from that that there were compelling reasons to upgrade," Cook said.

What's next: Apple is prepping another round of Apple Intelligence features, including improvements to Siri in the coming months, as well support for more languages, which Cook said will come in April.

• OpenAI is reportedly in talks to raise up to $40 billion at a pre-money valuation of $260 billion. The company was last valued at $157 billion in October. (Bloomberg)
• OpenAI CEO Sam Altman offered D.C. insiders an early look at product advances and predicted a "a big, big efficiency gain" from new agents like OpenAI's Operator. (Axios)
• Adobe strategy chief Scott Belsky is leaving the company to join indie production studio A24. (Variety)

Since I haven't mentioned Lego in a couple days, check out this upcoming set that recreates Van Gogh's "Sunflowers" on a canvas of bricks.