https://sputnikglobe.com/20250126/china-drops-powerful-ai-model-thats-free-fast-and-better-for-humanity-than-silicon-valley-rivals-1121502542.html
China Drops Powerful AI Model That’s Free, Fast and Better for Humanity Than Silicon Valley Rivals
China Drops Powerful AI Model That’s Free, Fast and Better for Humanity Than Silicon Valley Rivals
Sputnik International
The US and China are in an all-important race for AI supremacy, with America's outspending of the PRC multiple times over and restrictions on the Asian nation's ability to obtain sophisticated computing hardware seemingly having little impact.
2025-01-26T18:35+0000
2025-01-26T18:35+0000
2025-01-26T18:35+0000
science & tech
china
hangzhou
openai
meta
microsoft
artificial intelligence
chatgpt
asia
https://cdn1.img.sputnikglobe.com/img/07e4/0b/02/1080954125_159:0:3800:2048_1920x0_80_0_0_a46ad46edf1590cce83a64cc2d2bef4d.jpg
Advanced large language model DeepSeek R1 is taking users by storm, wowing reviewers and earning praise from AI-phobes.The Hangzhou-based tech startup’s new model beats OpenAI’s o1 on math and reasoning benchmarks, and blows Meta’s* Llama 3.1 and OpenAI’s GPT-40 out of the water in coding and complex problem-solving.The model is free to run locally, with access to its API priced at a fraction of competitors’ rates.The setup reportedly cost $5.6 million to train (vs $78 million for GPT-40), and uses performance-capped chips due to US restrictions, which also saw the use ban the delivery of more powerful processers to China.Instead, DeepSeek R1 harnesses its power from superior compute efficiency.How Does It Reason?Piquing users’ curiosity is the way in which the tool generates responses, in a process nature.com dubbed “analogous to human reasoning,” and thus “more adept than earlier language models at solving scientific problems.”That’s great news for scientists engaged in data analysis, pattern recognition and predictive modeling across a broad array of fields, from astronomy and medicine to the earth sciences.Best of All? It’s Actually ‘Open’Unlike other commercially available models, which experts have dubbed “essentially black boxes,” DeepSeek R1 is open source, allowing users fearful of AI turning into Skynet on them to study how it works and even build on it.DeepSeek founder Liang Wenfeng told Chinese media last year that “research and technological innovation,” not profit, was the company’s priority, and that his ultimate goal is artificial general intelligence.If the mission succeeds and an open-source AGI is born, humanity could be saved from a privatized, weaponized, and monopolized AI wiping us out.* Banned in Russia for extremism.
https://sputnikglobe.com/20241001/china-developed-ai-model-with-trillion-parameters-on-own-chips---reports-1120368696.html
https://sputnikglobe.com/20250125/trumps-stargate-project-gateway-to-us-ai-dominance-or-dystopian-cyberpunk-nightmare-1121490901.html
china
hangzhou
Sputnik International
feedback@sputniknews.com
+74956456601
MIA „Rossiya Segodnya“
2025
News
en_EN
Sputnik International
feedback@sputniknews.com
+74956456601
MIA „Rossiya Segodnya“
Sputnik International
feedback@sputniknews.com
+74956456601
MIA „Rossiya Segodnya“
what is deepseek, is deepseek chinese, is deepseek better
what is deepseek, is deepseek chinese, is deepseek better
China Drops Powerful AI Model That’s Free, Fast and Better for Humanity Than Silicon Valley Rivals
The US and China are in an all-important race for AI supremacy, with America's outspending of the PRC multiple times over and restrictions on the Asian nation's ability to obtain sophisticated computing hardware seemingly having little impact.
Advanced large language model DeepSeek R1 is taking users by storm, wowing reviewers and earning praise from AI-phobes.
The Hangzhou-based tech startup’s new model beats OpenAI’s o1 on math and reasoning benchmarks, and blows Meta’s* Llama 3.1 and OpenAI’s GPT-40 out of the water in coding and complex problem-solving.
The model is free to run locally, with access to its API priced at a fraction of competitors’ rates.
The setup reportedly cost $5.6 million to train (vs $78 million for GPT-40), and uses performance-capped chips due to US restrictions, which also saw the use ban the delivery of more powerful processers to China.
Instead, DeepSeek R1 harnesses its power from superior compute efficiency.
“We should take the developments out of China very, very seriously,” Microsoft CEO Stya Nadella said at the WEF’s annual meeting in Davos this week, days after DeepSeek’s latest model dropped.
1 October 2024, 01:44 GMT
Piquing users’ curiosity is the way in which the tool generates responses, in a process nature.com
dubbed “analogous to human reasoning,” and thus “more adept than earlier language models at solving scientific problems.”
That’s great news for scientists engaged in data analysis, pattern recognition and predictive modeling across a broad array of fields, from astronomy and medicine to the earth sciences.
Best of All? It’s Actually ‘Open’
Unlike other commercially available models, which experts have dubbed “essentially black boxes,” DeepSeek R1 is open source, allowing users fearful of AI
turning into Skynet on them to study how it works and even build on it.
DeepSeek founder Liang Wenfeng told Chinese media last year that “research and technological innovation,” not profit, was the company’s priority, and that his ultimate goal is artificial general intelligence.
If the mission succeeds and an open-source AGI is born, humanity could be saved from a
privatized, weaponized, and monopolized AI wiping us out.
* Banned in Russia for extremism.
