This week reports of potential performance breakthroughs by the Chinese AI app DeepSeek sent shockwaves through the market, challenging assumptions about AI compute efficiency and competitiveness. While it’s too early to verify these claims, the AppSOC team received over a dozen inquiries from security journalists and was widely quoted. Below is a summary of press questions and responses from the AppSOC team, including Chief Scientist Mali Gorantla, CEO Pravin Kothari, and CMO Willy Leichter. Links to the full articles are provided at the end.

The news and market reaction

• What about DeepSeek is making it the talk of Silicon Valley right now? Why has it spooked the market?
  AI has attracted a huge amount of investment, but taking a brute-force approach - throwing unlimited compute power at the is hugely expensive. If there's a far cheaper way to do this (which variants of Moore's Law say is inevitable), this causes headaches for short-term investments but should open the market for more innovation.  
• Why has this app surged in popularity?
  Any breakthrough that reduces the massive computing resources, and costs of large-scale AI will be popular. Lowering the costs to entry will benefit innovators, criminals, and anyone else using AI on a budget.
• Is this an overreaction or is there real disruption in the AI space?
  Short-term, it's probably an overreaction as there are likely gotchas or limitations to this new, more efficient approach. But the AI space is being built by disruption, massive investment, and quick pivots to new development models following breakthroughs. It’s good for all of us to avoid complacency.

‍How well does DeepSeek really perform?

• How difficult is it to verify DeepSeek's performance claims, since it's a Chinese company?
  Verifying these claims and doing apples-to-apples comparisons will definitely be difficult. The shortcuts that these Chinese developers found may have hidden flaws, massive security gaps, or other issues we can't anticipate yet. For people building AI on the cheap, this may not matter, but large businesses building AI applications will need much more assurance that China is likely to provide.
• How does DeepSeek stack up against AI players like OpenAI?
  It's too early to tell. Until DeepSeek is more carefully vetted, which will be difficult given its Chinese origins, it will be hard to tell if there are gotchas, security issues, or other flaws we can't predict. But assuming most of the claims are true, it will spur leading AI players to double down on their investments and look for their own efficiency breakthroughs.

Potential security issues

• What are the privacy risks of DeepSeek’s open-source AI models?
  The risks are unknown, but probably significant. By sharing resources across AI components, it seems likely that security flaws will emerge that will be exploited. With any AI rollout, if privacy is a concern, then you need to slow down, and test all your systems and assumptions carefully.
• How could DeepSeek’s AI be weaponized for cyberattacks or disinformation?
  Any development that lowers the cost of entry helps both innovators and cyber attackers. Powerful AI is a fact of life for attackers, defenders, and organizations trying to protect their own AI stacks from abuse. If AI costs drop dramatically, then risk levels will rise.
• Is DeepSeek a greater threat to US national security than TikTok?
  Potentially, yes. TikTok was a huge social phenomenon, but the value of user data on TikTok has its limits. Lowering the costs of building AI applications will help innovators, attackers, defenders, and everyone else in the cyber wars.

‍

Industry Competitiveness

• Will DeepSeek’s cost-effective approach force U.S. giants to slash prices?
  If real, these innovations will force U.S. giants to invest more in efficiency and prices will inevitably come down. But it's too early to predict a fire-sale on current AI resources.
• Will DeepSeek's performance claims influence how countries think about AI infrastructure? Why or why not?
  If these breakthroughs are legitimate, they will be very disruptive and spur more innovation. But we need to take a beat and carefully evaluate these claims.
• Does DeepSeek's ability to produce results comparable to leading US AI firms using chips less powerful than the most advanced Nvidia products undermine the rationale behind export controls on US technology?
  Yes. If these results are real, then companies benefitting from brute-force computing approaches will have to regroup. Export controls rarely work in the long term. Just like past controls on encryption, scarcity is the mother of innovation and breakthroughs outside the US.
• Does China’s state-backed AI strategy give DeepSeek an unfair edge?
  China has invested heavily in AI, and instead of focusing on massive computing power, they have trained massive armies of developers to attack these challenges. But on both sides, brute force doesn't always equal innovation. Fairness has nothing to do with it.

AppSOC was quoted in these published articles:

• TechNewsWorld: Chinese AI App DeepSeek Rattles Tech Markets, by John P. Mello
• Security InfoWatch: What is DeepSeek and what is all the fuss about?, by Steve Lasky
• SiliconANGLE: Tech stocks tank as Chinese startup DeepSeek stuns AI world, by Paul Gillin
• AI Business: Industry Weighs in on DeepSeek Launch, by Berenice Baker
• MarTECH: What is DeepSeek? A boon for marketers and a threat to Big Tech, by Constantine von Hoffman
• Unite.AI: DeepSeek-R1: Transforming AI Reasoning with Reinforcement Learning
• Medium: Decoding the LLM Titans: ChatGPT 4o1, Gemini 2.0, and DeepSeek R1 — A Deep Dive