DeepSeek R1

Headlines · 3

1. 10h
   DHS orders US civilian agencies to remove DeepSeek technology and halt purchases

   The Department of Homeland Security published a Federal Register notice ordering federal civilian agencies to remove DeepSeek products from their IT systems and stop buying its products and services. The order also covers third-party tools built on DeepSeek's models and bars DeepSeek from participating in federal government contracts. Issued under a 2018 supply-chain security law and signed by Homeland Security Secretary Markwayne Mullin, the notice says DeepSeek's account-generation process collects personally identifiable information and may digitally fingerprint users, raising the risk that sensitive data could be transferred to the Chinese government, and that its models could contain malware or other security vulnerabilities. The order is dated July 24 but was made public only on Monday; Engadget noted it came days after DeepSeek released a new model.

   Gizmodo · 🔥 51
2. 3d
   Research firm calls Anthropic IPO 'most ridiculous of 2026', values it as low as $144bn

   Independent research firm New Constructs said in an October 6 report titled "Anthropic Is The Most Ridiculous IPO of 2026" that Anthropic's equity could be worth as little as $144bn, 93% below the $2tn valuation the company is reportedly targeting. Using a reverse discounted cash flow model, it says $2tn would require Anthropic to hold a 10% after-tax operating margin for a decade, grow revenue 1,000% this year and reach $3.6tn in revenue by 2035; the report also says Claude Fable 5 and Mythos 5 cost about 18x more than DeepSeek R1 for input tokens and 23x more for output tokens, and argues open-source models make it hard for closed models to profit. Anthropic has since told investors its third-quarter revenue reached $16.8bn, which PitchBook analyst Harrison Rolfes said clears the bar he had set but still does not by itself justify $2tn; Bloomberg reported the listing could come as soon as mid-November, with about 80 of the 261 pages in the confidential draft filing devoted to risk factors including catastrophic or existential risk and models attempting to "resist shutdown."

   The Next Web · 🔥 0
3. Oct 12
   We put too much faith in AI saying no, MIT Tech Review argues

   An MIT Technology Review piece argues that refusal has become the "load-bearing wall" of AI safety: models are trained to decline statistically harmful prompts, with classifiers and newer "probes" layered around them, and Anthropic said one class of classifier raised its chatbot's compute cost by 24%. Because refusal is probabilistic it both fails and over-fires — the article says Amazon researchers unlocked some hacking abilities in Anthropic's Fable 5 within three days of its release, while Fable also routed a harmless question about sake versus Korean makgeolli to a weaker model. It further cites CrowdStrike's finding that DeepSeek R1 produced buggier code for a fictional Tibet bank and an "Uyghurs Unchained" app, a UK AI Safety Institute test in which three Anthropic models declined more than half of reasonable AI-safety research tasks, and a Microsoft executive's description of OpenAI's Astra applying stricter refusals to users it deems "high-risk."

   MIT科技评论中文 · 🔥 0

Experience and discussion from the community

Share my DeepSeek R1 experienceAsk about DeepSeek R1