Trends Artificial Intelligence
leads us back to one of our favorite quotes – Statistically speaking, the world doesn’t end that often, from former T. Rowe Price Chairman and CEO Brian Rogers. As investors, we always assume everything knowledge. It reflects how much that knowledge could be worth if applied effectively, even if it hasn’t yet generated revenue. Source: Microsoft, ‘Governing AI: A Blueprint for the Future,’ Microsoft Report you happen to be Indonesian. Today, there’s a language barrier and it will be very hard if you don’t know English to be able to get to a world stage. But with AI, it might be possible in the future where0 码力 | 340 页 | 12.14 MB | 4 月前3
DeepSeek-V2: A Strong, Economical, and Efficient
Mixture-of-Experts Language Modelthroughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock achieve top-tier performance among open-source models. The model checkpoints are available at h t t p s : / / g i t h u b . c o m / d e e p s e e k - a i / D e e p S e e k - V 2 . 0 20 40 60 80 100 Activated 200 250 300 DeepSeek-V2 DeepSeek 67B saving 42.5% of training costs Training Costs (K GPU Hours/T Tokens) 0 100 200 300 400 DeepSeek-V2 DeepSeek 67B reducing KV cache by 93.3% KV Cache for Generation0 码力 | 52 页 | 1.23 MB | 1 年前3
Google 《Prompt Engineering v7》modalities such as image prompts) is the input the model uses to predict a specific output. You don’t need to be a data scientist or a machine learning engineer – everyone can write a prompt. However, ambiguous, inaccurate responses, and can hinder the model’s ability to provide meaningful output. You don’t need to be a data scientist or a machine learning engineer – everyone can write a prompt. Prompt and higher costs. Prompt Engineering February 2025 9 Reducing the output length of the LLM doesn’t cause the LLM to become more stylistically or textually succinct in the output it creates, it just0 码力 | 68 页 | 6.50 MB | 6 月前3
Manus AI:Agent元年开启2cLevel 3¬G-•>Manus AI L®‰¯# §¨©ª°±²³{´µG SOTA œ=> • Manus AI ¶·fgG$%JKA+)€,¸¹!Lº»JK«Level 3¬°G-•¼½a‡¹T AI Ÿ >•)¾%‡ˆ¿ÀGÁ%ÂÃ,Ä Å'B|4ÆcÇ©ÈÉÊËcÌÍ•mÎÏJKG()A+> !"#$%Bloomberg*&'()6 Manus AI%2345 • ManusÐ!ÑÒÓ*GÔg<Õ5 "#$%Bloomberg*&'()7 Manus AI%6789: • 67,89:;<щ=>?Š@&ACEO,BC‡DF<Ñg[> • 2016 E:;zFW>GHIJ÷øGKfLMNOPQgRTýTUØV"WX>OPŸ !zFW>GHIJ÷øGKfLM, ÇYZ‰200[G\]¥,^E‰°_[G`a> • 2019 EbcCFW 3.0 µdeG÷øÕf$2°,67ËþæacCFWghX(–ò)bºde)€GáÛ> • *UŸzAI Agent‰+LÌžúïm•)€áÛ51¬LJKí!úït¡‡í,c-]Gí!)€ÙÚÆ¡‡mYG.0cÓ*æÆÇ 2¬L'¶ñ%2µ¶úï,t¡‡&‰÷/Gwþ'¶,Æ'I'¶Gµ¶í!æÆ,01ˆæ"ÚGŠ‹¾%‡LLMG'¶> !"#$%Bloomberg*&'()17 AI Agent%U[R\]+^Y_`+^Z 0 码力 | 23 页 | 4.87 MB | 5 月前3
OctoML OSS 2019 11 8attocates add(t1,t2) ncubator-tvm/pull/3560 了 e, Enables future optimizations fn emain() -,Tensor[tk,),f32] { and end-to-end dynamic Tet tl 引 -。 Let t2 3 invoke_tvn_op(add,(tl,t2),(outl,))3 Out1l loops, and offloading dynamic } allocation to devices. QQ octoML VM Memory Abstractions Old New t1: Tensor t1: Tensor t2: Tensor t2: Tensor t3: Tensor t3: Tensor Q octoML Coalesced t1: Tensor t2: Tensor t3: Tensor 13 Acknowledgments e The0 码力 | 16 页 | 1.77 MB | 5 月前3
OpenAI 《A practical guide to building agents》reservation, committing a code change, or generating a report. Applications that integrate LLMs but don’t use them to control workflow execution—think simple chatbots, single-turn LLMs, or sentiment classifiers—are context, considering subtle patterns, and identifying suspicious activity even when clear-cut rules aren’t violated. This nuanced reasoning capability is exactly what enables agents to manage complex, ambiguous try swapping in smaller models to see if they still achieve acceptable results. This way, you don’t prematurely limit the agent’s abilities, and you can diagnose where smaller models succeed or fail0 码力 | 34 页 | 7.00 MB | 6 月前3
普通人学AI指南. . . . . . . . 6 1.4.1 上下文窗口 . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.4.2 单位 B 和 T . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2 AI 工具梳理 6 2.1 问答 . . . . . . . . . . . . . 1.4.2 单位 B 和 T 在 AI 大模型中,常用的两个单位是 B 和 T。 B(十亿,Billion):在英文里是 Billion 的缩写,表示十亿。对于 AI 大模型 来说,B 一般用于描述模型的参数数量。例如,具有 50B 参数的模型代表这个 模型有 50 亿个参数。Ollama3 有尺寸 8B 和 70B,Phi-3-mini 有 3.8B 参数等。 T(万亿,Trillion):在英文里是 Trillion 的缩写,表示万亿。在 AI 大模型 中,”T” 常用来表示模型在训练中处理的 Token 数量。Token 是指模型处理的 基本单元,可以是一个单词、子词,或者字符等。 在大规模预训练语言模型的训练中,通常会提到模型是在多少个 Token 上 进行学习的,以表明模型的训练规模和数据量。例如:LLaMA3 语言模型使用 了超过 15T 个 token 进行训练。 2 AI 工具梳理 大家有没有觉得0 码力 | 42 页 | 8.39 MB | 8 月前3
OpenAI - AI in the Enterprisedelivering more relevant and responsive customer experiences. 3 AI in the EnterpriseBut leveraging AI isn’t the same as building software or deploying cloud apps. The most successful companies are often those improvement, all while maintaining satisfaction scores on par with human support. These results didn’t happen overnight. Klarna achieved this performance by continuously testing and refining the assistant mindset, backed by rigorous evaluations, and safety guardrails. The companies seeing success aren’t rushing to inject AI models into every workflow. They’re aligning around high-return, low-effort use0 码力 | 25 页 | 9.48 MB | 5 月前3
清华大学 普通人如何抓住DeepSeek红利7 K w S v L C q Y 4 Y V 1 T 8 0 u m B k k m O x d k C i y K r j i 6 n p Y d O w t v B 4 G 0 G p y 8 U I q e T 9 M 6 Deepseek的能力图谱 直接面向用户或者支持开发 Natural Questions等)来生成问题。可以从多个数据集中组 合问题,以达到10万个的问题数量。 这 些 数 据 集 包 含 大 量 的 问 答 对 , 例 如 使 用 d a t a s e t s 库 (Hugging Face的datasets库)来加载SQuAD数据集 (Stanford Question Answering Dataset),这个数据集 是一个著名的问答数据集,基于维基百科数据生成,并且数0 码力 | 65 页 | 4.47 MB | 8 月前3
清华大学 DeepSeek+DeepResearch 让科研像聊天一样简单25:177-185. Brnmark C, Lakowitz T Hollander J(2011) Predator-induced morphological plasticity across local populations of a freshwater snail. PLOS ONE 6:e21773. Bronmark C. Lakowitz T,Hollander J(2011) Predator-induced 门 槛 , 推 动 AI技术民主化。 重塑定价逻辑 DeepSeekV3模型以557.6 万 美 元 的 训 练 成 本 , 实 现 了 与 G P T - 4 o 相 当 的 性 能 , 生 成 速 度 提 升 至 6 0 T P S 。 这 种 “ 低 成 本 高 性 能 ” 模 式 不 仅挑战了OpenAI、Google 等 巨 头 的 市 场 地 位 , 还 迫 使 行 8 5 % ),重塑了 A I 服 务 的定价逻辑。 推动研发转型 D e e p S e e k 的 全 栈 开 源 策 略 ( 模 型 权 重 、 训 练 代 码 均 采 用 M I T 协 议 ) , 吸 引 了 全 球 开 发 者 参 与 , 形 成 了 强 大 的 社 区 生 态 。 这 种 开 放 模 式 加 速 了 技 术 迭 代 , 削 弱 了 闭 源 巨 头 的0 码力 | 85 页 | 8.31 MB | 8 月前3
共 20 条
- 1
- 2













