←Search results

Search “DeepSeek”

By relevance · Total 1368 results Clear search

Tuesday

  1. Zyphra Research

    Zyphra Inference Wave 2: Accelerating MoE and Long-Context Inference Serving on MI355X

    Zyphra has released ZOE, its second-generation inference optimization, which on AMD Instinct MI355X reduces GLM-5.3's time to first token (TTFT) at 512K context from 178.8 seconds to 64.6 seconds (2.8x), and from 42.3 seconds to 11.8 seconds (3.6x) at 90% prefix cache hit rate. Its CSA top-K selection HIP kernel raises DeepSeek-V4-Pro's single-user decode throughput at 1M context from 43.5 to 61.1 tokens/s (1.40x). AMD MXFP4 quantization of GLM-5.3 shrinks the model from 756 GB to 410 GB, with benchmark scores such as AIME25 within 3 points of FP8.

    ce optimization, which on AMD Instinct MI355X reduces GLM-5.3's time to first token (TTFT) at 512K context from 178.8 seconds to 64.6 seconds (2.8x), and from 42.3 seconds to 11.8 seconds (3.6x) at 90% prefix cache hit rate. Its CSA top-K selection HIP kernel raises DeepSeek-V4-Pro's single-user decode throughput at 1M context from 43.5 to 61.1 tokens/s (1.40x). AMD MXFP4 quantization of GLM-5.3 shrinks the model from 756 GB to 410 GB, with benchmark scores such as AIME25 within 3 points of FP8.

    Matched fieldsSummary、Body

    #lab#Product updates#Reasoning#English text is being prepared.
  3. Alibaba Cloud Developer Community AI

    Development Methods and Tech Stack for an AI English Writing Agent

    An AI English writing agent is implemented through the collaboration of a multi-agent pipeline, and can be split into agents for outline and idea generation, grammar and spelling, discourse logic, style polishing, and evaluation and scoring, built on LangGraph or PydanticAI. The recommended tech stack is Python (FastAPI) paired with PydanticAI, LangGraph, and LlamaIndex; the main reasoning model can be DeepSeek-V3, Claude 3.5 Sonnet, or GPT-4o, with GPT-4o-mini and Qwen-Turbo for lightweight validation. The system also uses RAG to connect to authoritative corpora such as the Oxford and Longman dictionaries and the Academic Phrasebank, and sets up anti-hallucination and academic integrity guardrails.

    hing, and evaluation and scoring, built on LangGraph or PydanticAI. The recommended tech stack is Python (FastAPI) paired with PydanticAI, LangGraph, and LlamaIndex; the main reasoning model can be DeepSeek-V3, Claude 3.5 Sonnet, or GPT-4o, with GPT-4o-mini and Qwen-Turbo for lightweight validation. The system also uses RAG to connect to authoritative corpora such as the Oxford and Longman dictionaries and the Academic Phrasebank, and sets up anti-hallucination and academic integrity guardrails.

    Matched fieldsSummary

    #hot-signal#Tutorials / practice#Agents#RAG
  7. cnBeta

    Nobel laureate Levitt: AI will erase the age gap among scientists, and in 30 years a young person may accomplish in one year what predecessors did in a lifetime

    Michael Levitt, winner of the 2013 Nobel Prize in Chemistry, predicted at the Second International Conference on AI Scientists (ICAIS 2026) that by 2056, young scientists aided by AI agents may accomplish in one year the work his generation did in a lifetime, and said this prediction may not even be aggressive enough—the efficiency gain could far exceed 30 times. He believes AI will erase the age gap among scientists, and stressed that the coexistence of multiple models such as DeepSeek, Kimi, Claude, and ChatGPT is the guarantee for humanity to use AI safely.

    cted at the Second International Conference on AI Scientists (ICAIS 2026) that by 2056, young scientists aided by AI agents may accomplish in one year the work his generation did in a lifetime, and said this prediction may not even be aggressive enough—the efficiency gain could far exceed 30 times. He believes AI will erase the age gap among scientists, and stressed that the coexistence of multiple models such as DeepSeek, Kimi, Claude, and ChatGPT is the guarantee for humanity to use AI safely.

    Matched fieldsSummary、Body

    #hot-signal#English text is being prepared.#Agents
  12. TechCrunch AI

    Tenstorrent raises $1.5B at a $6.2B valuation to expand open-weight AI chips

    AI chip startup Tenstorrent has completed a $1.5 billion funding round at a valuation of $6.2 billion, more than double its $2.6 billion valuation at the end of 2024. The round was led by Trace Capital and Innovation X, with participation from Cadence Ventures, Eclipse, Fidelity, SanDisk, and T. Rowe Price. CEO Jim Keller said the company decided to remain independent and continue raising funds after demonstrating that its Galaxy servers can run DeepSeek models, and that the new capital will be used to expand chips for open-weight model inference and video generation.

    a valuation of $6.2 billion, more than double its $2.6 billion valuation at the end of 2024. The round was led by Trace Capital and Innovation X, with participation from Cadence Ventures, Eclipse, Fidelity, SanDisk, and T. Rowe Price. CEO Jim Keller said the company decided to remain independent and continue raising funds after demonstrating that its Galaxy servers can run DeepSeek models, and that the new capital will be used to expand chips for open-weight model inference and video generation.

    Matched fieldsSummary、Body

    #tech-media#Industry news#English text is being prepared.
  14. IT Home · Artificial Intelligence

    Turing Award winner Joseph Sifakis calls for establishing global AI regulatory standards

    Joseph Sifakis, winner of the 2007 Turing Award, called for establishing global standards and regulatory standards in the field of AI at the "2026 Sustainable Global Leaders Conference," and said this is a shared responsibility of humanity. Citing ChatGPT and DeepSeek scoring higher than humans on medical exams as an example, he pointed out that large models do not understand social norms and value systems and therefore cannot replace doctors. He also warned that large companies may use regulatory gaps to shirk responsibility and seek commercial gain.

    or establishing global standards and regulatory standards in the field of AI at the "2026 Sustainable Global Leaders Conference," and said this is a shared responsibility of humanity. Citing ChatGPT and DeepSeek scoring higher than humans on medical exams as an example, he pointed out that large models do not understand social norms and value systems and therefore cannot replace doctors. He also warned that large companies may use regulatory gaps to shirk responsibility and seek commercial gain.

    Matched fieldsSummary、Body

    #analysis#English text is being prepared.#Safety / alignment
  15. 36Kr · AI

    Young people today are paying to "hand-craft" AI personas

    AI role-play prompts have spawned a niche, clearly priced business on social platforms. Custom AI personas generally cost 5–8 yuan, single-character prompts usually stay within 10 yuan per 1,000 characters, delivery takes 2–3 days, and rush orders or buyouts cost double. Users try models such as DeepSeek, Doubao, Gemini, and ChatGPT, and some keep paying because they find DeepSeek "the most human-feeling." The business is still at a low price point, popular settings are easily copied, and disputes remain over ownership of settings and the boundaries of originality.

    ocial platforms. Custom AI personas generally cost 5–8 yuan, single-character prompts usually stay within 10 yuan per 1,000 characters, delivery takes 2–3 days, and rush orders or buyouts cost double. Users try models such as DeepSeek, Doubao, Gemini, and ChatGPT, and some keep paying because they find DeepSeek "the most human-feeling." The business is still at a low price point, popular settings are easily copied, and disputes remain over ownership of settings and the boundaries of originality.

    Matched fieldsSummary、Body

    #hot-signal#English text is being prepared.#Agents

Wednesday

Monday

  1. Artificial AnalysisSelected

    MiniMax-M3.1-Flash scores 42 on the Artificial Analysis Intelligence Index

    MiniMax-M3.1-Flash scores 42 on the Artificial Analysis Intelligence Index, 13 points higher than MiniMax-M3, at less than half the latter's per-task cost. It scores 48% on Humanity's Last Exam and 22% on CritPt, both higher than GLM-5.3-Flash and Gemini 3.8 Flash, and slightly lower than DeepSeek V4.1 Flash 1017. The model has a 1M-token context window, supports text and image input, is priced at $0.15/$0.60 per 1M input/output tokens, and its weights are currently closed-source; MiniMax announced that it will open the weights on October 23.

    nalysis Intelligence Index, 13 points higher than MiniMax-M3, at less than half the latter's per-task cost. It scores 48% on Humanity's Last Exam and 22% on CritPt, both higher than GLM-5.3-Flash and Gemini 3.8 Flash, and slightly lower than DeepSeek V4.1 Flash 1017. The model has a 1M-token context window, supports text and image input, is priced at $0.15/$0.60 per 1M input/output tokens, and its weights are currently closed-source; MiniMax announced that it will open the weights on October 23.

    Matched fieldsSummary、Body

    #editorial#Evaluations / benchmarks#Reasoning#DeepSeek
  3. Prime Intellect

    prime-rl 0.10.0 Released: Faster Training, New RL Loss IPO, and a Unified Evaluation Stack

    Prime Intellect has released prime-rl v0.10.0 and verifiers v0.4.0, adding the RL algorithm IPO and setting it as the default, replacing the previous DPPO-based default loss. In tests comparing Qwen3-235B-A22B SFT against torchtitan, prime-rl achieved a 1.15× speedup (3732 vs 3253 tokens/s/GPU), and the MFU of full GLM-5.3 SFT rose from 22.3% to 40.1%. The evaluation CLI has moved from verifiers into prime-rl, with added support for models such as DeepSeek V4 Flash, Qwen3.8 Flash Next, and Nemotron-H.

    prime-rl 0.10.0: faster training, our new RL loss IPO, and one unified eval stack

    Intellect has released prime-rl v0.10.0 and verifiers v0.4.0, adding the RL algorithm IPO and setting it as the default, replacing the previous DPPO-based default loss. In tests comparing Qwen3-235B-A22B SFT against torchtitan, prime-rl achieved a 1.15× speedup (3732 vs 3253 tokens/s/GPU), and the MFU of full GLM-5.3 SFT rose from 22.3% to 40.1%. The evaluation CLI has moved from verifiers into prime-rl, with added support for models such as DeepSeek V4 Flash, Qwen3.8 Flash Next, and Nemotron-H.

    Matched fieldsSummary、Body

    #company#Product updates#Open-source ecosystem#English text is being prepared.
  4. Artificial Analysis full articleSelected

    Artificial Analysis launches HealthBench and HealthBench…

    Artificial Analysis has launched leaderboards for OpenAI's two health dialogue benchmarks, HealthBench and HealthBench Professional; the former contains 5,000 conversations with general users and health workers, while the latter consists of 525 conversations all aimed at clinicians. On HealthBench Professional, GPT-6 Astra (max) ranks first with a length-adjusted score of 66.0%, GPT-6.1 Sol (max) is at 65.1%, and Claude Sonnet 5.5 (max with fallback) is at 63.0%. On HealthBench, Claude Sonnet 5.5 (max with fallback) leads at 62.1%, while DeepSeek V4.1 Flash (max) is at 60.9%; Claude Opus 5.5 and Claude Fable 5.1 average about 5,900 characters per answer and lose more than 11 points after length adjustment.

    le the latter consists of 525 conversations all aimed at clinicians. On HealthBench Professional, GPT-6 Astra (max) ranks first with a length-adjusted score of 66.0%, GPT-6.1 Sol (max) is at 65.1%, and Claude Sonnet 5.5 (max with fallback) is at 63.0%. On HealthBench, Claude Sonnet 5.5 (max with fallback) leads at 62.1%, while DeepSeek V4.1 Flash (max) is at 60.9%; Claude Opus 5.5 and Claude Fable 5.1 average about 5,900 characters per answer and lose more than 11 points after length adjustment.

    Matched fieldsSummary、Body

    #company#Evaluations / benchmarks#OpenAI