Latest news with #PhilippSchmid
South China Morning Post
21-03-2025
- Business
- South China Morning Post
ByteDance advances DeepSeek work in AI reasoning with open-source project led by intern
TikTok owner ByteDance, which has invested heavily in artificial intelligence (AI), has unveiled a new system that claims to improve on the work done by DeepSeek in training AI reasoning models. Advertisement DAPO, or Decoupled Clip and Dynamic Sampling Policy Optimisation, is a scalable reinforcement learning algorithm that helps a large language model (LLM) achieve better complex reasoning behaviour such as self-verification and iterative refinement, according to a research paper published earlier this week by ByteDance and Tsinghua University's Institute for AI Industry Research. The algorithm outperformed the reinforcement learning approach in DeepSeek's R1 reasoning model, scoring 50 points in the American Invitational Mathematics Examination (AIME) 2024 using Alibaba Group Holding's Qwen2.5-32B base model, compared with 47 points attained by R1 when applying the same Alibaba model, the paper showed. Alibaba owns the South China Morning Post. Notably, DAPO achieved the better result with 50 per cent fewer training steps. TikTok owner ByteDance has invested heavily in artificial intelligence. Photo: Digitimes The achievement drew positive academic and industry comments. Google DeepMind engineer Philipp Schmid, who shared the project on X, said the new method was 'better than' DeepSeek's 'group relative policy optimisation (GRPO)' in reinforcement learning. GRPO is one of DeepSeek's training methods that enables a model to learn by comparing different actions and making updates with a 'group' of observations.
Yahoo
31-01-2025
- Business
- Yahoo
Not Just DeepSeek - Alibaba Unveils AI Model To Rival OpenAI's Operator
On Monday, Chinese e-commerce juggernaut Alibaba Group Holding's (NYSE:BABA) cloud unit released a new family of AI models, Qwen2.5-VL, that can parse files, comprehend videos, count objects in images, and control a PC. This model is similar to the one powering OpenAI's recently launched Operator. Qwen2.5-VL model claimed to beat OpenAI's GPT-4o, Anthropic's Claude 3.5 Sonnet, and Google's Gemini 2.0 Flash in various video understanding, math, document analysis, and question-answering evaluations, TechCrunch reports. Don't Miss: Deloitte's fastest-growing software company partners with Amazon, Walmart & Target – Built on the trusted network of Fortune 500 companies, this blockchain company partners with Salesforce to uproot lengthy and expensive B2B transactions, and you can invest with just $100. Qwen2.5-VL claimed it can analyze charts and graphics, extract data from invoice and form scans, and comprehend multiple-hour-long videos. It can also recognize IPs from films and TV series. Qwen2.5-VL can interact with software both on PCs and mobile devices. It can launch the app for Android and can book a flight from Chongqing to Beijing, TechCrunch cites from Hugging Face tech lead Philipp Schmid's video tweet on X. In January, Alibaba Cloud launched new AI tools and LLMs at its developer summit. Alibaba's cloud revenue grew 7% to $4.22 billion in second-quarter. Trending: Nancy Pelosi Invested $5 Million In An AI Company Last Year — Meanwhile, U.S. tech stocks plunged in premarket trading Monday as Chinese open-source artificial intelligence platform DeepSeek R1 triggered market jitters over sustainability of the AI technology investment in the U.S. tech firms. Nvidia Corp (NASDAQ:NVDA) lost $600 billion in market cap on Monday. DeepSeek's open-source AI model, developed for under $6 million, reportedly outperformed leading U.S. models like those from OpenAI. Reportedly, Microsoft Corp (NASDAQ:MSFT) committed $80 billion in AI infrastructure spending for 2025, and Meta Platforms Inc (NASDAQ:META) earmarked $60-65 billion. For context, the Biden administration had slapped multiple semiconductor technology embargoes on China, restricting the country from accessing sophisticated AI technologies from Nvidia, Taiwan Semiconductor Manufacturing Co (NYSE:TSM) citing national security threats. Investors can gain exposure to stocks of companies domiciled in China through the iShares China Large-Cap ETF (NYSE:FXI) and the KraneShares Trust KraneShares CSI China Internet ETF (NYSE:KWEB). Read Next: This 12,000 RPM Spinning Battery With Over $100 Million In LOIs Could Be The Missing Link For Green Energy — Here's Why Early Investors Are Flocking To Invest Before Funding Closes 'Scrolling To UBI' — Deloitte's #1 fastest-growing software company allows users to earn money on their phones. You can invest today for just $0.26/share with a $1000 minimum. Photo by Ascannio via Shutterstock Up Next: Transform your trading with Benzinga Edge's one-of-a-kind market trade ideas and tools. Click now to access unique insights that can set you ahead in today's competitive market. Get the latest stock analysis from Benzinga? This article Not Just DeepSeek - Alibaba Unveils AI Model To Rival OpenAI's Operator originally appeared on © 2025 Benzinga does not provide investment advice. All rights reserved.
