43 posts tagged by Digg
AI
DeepSeek's V5 would train almost entirely on Huawei chips to reduce reliance on US hardware, one commentator argues.
AI
Longer contexts, more test runs and revised grading underpin its new comparison with GSQ-RCO.
AI
DeepSeek's open-source Super-Agent prototype can monitor developer group chats and suggest coding solutions without being asked, its team says.
Technology
Its nearly $16 billion round reportedly drew backing from Tencent, CATL and Chinese state-owned institutions.
Technology
Reuters attributes the Astra distillation claim for Kimi K3.1 Pro and V4.1-Flash 1017 to The Information.
Technology
Reuters names Moonshot's Kimi K3.1 Pro and DeepSeek's V4.1-Flash 1017 in its account of The Information's report.
AI
The Information links Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017 to OpenAI’s model, Reuters reports.
World & Business
Reuters says The Information reported heavy use of OpenAI's GPT-6 Astra outputs by DeepSeek and Moonshot AI.
World & Business
Reuters attributes the reported approvals to a CCTV-affiliated account as China weighs proposed export controls on 23 AI technologies.
Technology
TheAigentic says frontier AI labs are lining up for the same money at once.
Technology
A translation posted by latentspacepod says some components of DeepSeek's production system are not yet fully open source.
AI
A post speculates that an unnamed man won't sell, saying he needs total control to continue the work.
AI
A user estimates 42 GB200 NVL72 racks could suffice at what they call production throughput, using off-peak prices only.
World & Business
DeepSeek says the framework powers DeepSeek-V4.1-Flash; its repository provides offline inference but omits components needed for a full service.
AI
DeepSeek's README offers offline inference for DeepSeek-V4.1-Flash but lists additional systems needed for a complete service.
AI
A user calls the result impressive but says the system is multiple times less efficient than Sol.
AI
A user credits DeepSeek directly for technological acceleration and predicts people will generally underestimate its contribution.
AI
Techmeme, citing The Information, reports DeepSeek plans to use Ascend 950DT for pre-training as early as the end of 2026.
Technology
The Wall Street Journal reports the order also bars future purchases of DeepSeek products and services.
AI
A commentator suggests MiniMax is also using smaller models and links the shift to compute scarcity and reinforcement learning.
Technology
A tech news roundup says the order cites national security and data protection risks.
AI
DeItaone cites a notice barring government agencies from purchasing DeepSeek-related products.
AI
One post calls a linked DeepSeek post ‘crazy pareto mogging’; another quips, ‘*when removing the models that beat them.’
Technology
The company says the model improves benchmarks, task speed and output length over its predecessor.
AI
DeepSeek says the model is live on its website, app and API, and optimized for DeepSeekHarness and Tencent CodeBuddy.
Technology
Techmeme relays a Bloomberg report citing sources that Tencent and CATL would be the round’s biggest contributors.
AI
A poster says they're certain the UK could build an AI lab at that level, though they acknowledge it wouldn't be easy.
AI
A post says the team has many openings and links to a technical write-up described as covering sandbox infrastructure for large-scale agent training.
AI
A post describes DeepSeek as wrestling with that question.
AI
A user argues the restrictions were a mistake, describing the claimed lack of U.S. dependency as “exactly as predicted.”
