与「DIVE」相关的搜索结果

Why Tokenization Could Be the Most Important Economic Innovation of Our Lifetime | Bruce Fenton @BruceFenton is a longtime Bitcoin advocate since 2012, securities professional, and one of the early pioneers of tokenized securities. He is the founder and CEO of Chainstone Labs and founder of Atlantic Financial. Alongside Overstock founder @PatrickByrne, Bruce was involved in some of the earliest efforts to bring securities and financial-market infrastructure onto blockchain rails. In 2018, he tokenized equity in his own company, Chainstone Labs, on the Ravencoin blockchain—years before tokenization became one of Wall Street’s biggest narratives. In this episode, Bruce breaks down naked short selling, the plumbing behind traditional securities markets, and why he believes much of the problem is ultimately a ledger problem. We dive into questions such as: Who actually owns the stocks sitting in your brokerage account? How can shares be lent and rehypothecated? What role do brokers, clearing firms, DTCC, and Cede & Co. play in determining who owns what? And could blockchain replace an opaque system of claims and intermediaries with a transparent ledger showing exactly where an asset is and who owns it? We unpack what “tokenized stocks” actually means—and the important distinction between a token that simply gives you exposure to a stock’s price and a security issued natively onchain, where the token itself represents the actual share and the ownership rights that come with it. Bruce discusses how tokenization could allow smaller businesses to tokenize ownership and have their stock traded globally without needing to be listed on a traditional exchange like the NYSE, as well as what a future without KYC could look like. Finally, we get into Wall Street’s embrace of tokenization, the SEC’s evolving approach to tokenized securities, and one of the biggest questions surrounding the future of this technology: who ultimately controls these new financial rails? Remember to subscribe and hit the bell “🔔” icon to get notifications 0:00 Bruce Fenton & Patrick Byrne: Pioneers of Tokenized Securities 3:36 Naked Short Selling Explained 6:28 Who Actually Owns the Stocks in Your Brokerage Account? 12:00 Tokenization Solves Wall Street’s “Ledger Problem” 15:48 What Are You Actually Buying With Robinhood’s Tokenized Stocks? 16:48 Real Tokenized Stocks: When the Token IS the Share 21:46 Tokenization Could Open U.S. Markets to the Entire World 22:52 Why Companies May No Longer Need the NYSE 26:29 Could Tokenized Markets Exist Without KYC? 29:31 Bruce Tokenized Equity on Ravencoin in 2018 30:53 Will Blockchain Make Clearing Houses & DTCC Obsolete? 33:17 Could Tokenization Have Prevented the GameStop Crisis? 38:19 Wall Street Is Embracing Tokenization — Who Gets the Power? 40:20 The SEC’s New Tokenized Securities “Innovation Exemption” 42:57 Why Robinhood’s Tokenized Stocks Don’t Qualify 47:01 Could Everything Eventually Be Tokenized? 51:02 The Dystopian Side of Tokenization: Surveillance & Asset Seizure

Kling 4.0 is coming this October. Kling 4.0 Flash is live now for Ultra Yearly subscribers. Kling 4.0 brings video creation to a new level of visual realism, creative control, and narrative completeness. 🎞️ Upgraded Audio & Visuals: Stable dynamic motion, high-quality stereo audio, more accurate lip sync, up to 4K resolution, and 10-bit HDR output. 🔮 Omni Reference: Richer reference options with up to 15 multi-modal references, more consistent results, and enhanced video editing. 🎥 Seamless Storytelling: Multi-keyframe control supporting up to 10 keyframes, native 30-second generation. 🌍 Diverse Possibilities: Video extension, support for multiple languages, accents, and dialects. The stage is set. You call the shots!

NVIDIA 发布 Skill2Env：用“集体技能”强化智能体 NVIDIA 研究者们把社区公开的 Agent Skills 编译成可执行 RL 训练环境的数据流水线：3.4k 个 Skills 变成 8k 个带程序化测试和行为量规的终端任务；仅 300 步 RL 训练就让 Qwen3.8-27B 在 Terminal-Bench 2.1 上提升 4.7 个百分点，且模型行为显著向源 Skills 的方法论对齐。 开源项目： 核心洞察：公开 Agent Skills 是一个被忽视的监督来源 Agent Skills 是“教智能体做某件事”的文件夹：一个 SKILL.md 加上可选的脚本、参考资料和资产。论文指出，把公开 Skill 语料当作数据来读，它同时提供三样东西： · 任务分布的采样：人们真正想让智能体处理的任务分布（有人愿意花时间写下工作流，说明这活儿值得自动化）； · 真实世界的锚点：指向真实的仓库、数据集、工具和工件； · 结果测试表达不了的质量标准：领域专长、默认参数、常见坑、“好结果长什么样”。 # 数据流水线：四阶段编译，验证靠构造 1. Plan（分解）：容器化的 Codex 规划器读取完整 Skill 包、联网调研相关公共资产，把 Skill 拆解成若干可验证的 workflow，每个附带元计划（场景、初始世界、预埋缺陷、难点来源、解法草案、验证策略）、资产建议和“任务轴池”（任务原型 × 验证器模式 × 人物画像）。 2. Diversify（多样化）：宿主从轴池采样一组组合，加上复杂度、指令语气、请求者专业水平。关键设计是轴池以 workflow 为条件：研究型 workflow 配“证据可追溯”验证和研究者画像，而不是从全轴乘积空间乱抽，这让多样化保持 sensible。 3. Create（构造）：全新创建者 Codex agent 在 Docker 内工作，尽可能用真实素材（钉在特定 commit 的开源仓库、真实版本化文档、官方 API 规范）；需要联网服务的场景改造成本地替身（stub 服务器、录制回放 fixture、PATH 上的假 CLI、种子数据库），求解时绝不依赖网络。创建顺序被严格固定：先建世界 → 写指令 → 写测试 → 写量规 → 最后才写参考解，测试先于解法冻结，保证解法必须迁就评分契约而非反过来。 4. Verify（验证）：宿主端无模型参与的接收门：静态检查（布局、符号链接、Dockerfile 安全、基础镜像按内容摘要钉死）+ 两个容器内试跑：Oracle（参考解）必须全指标满分，NOP（什么都不做的 agent）必须全指标零分。任一失败即拒绝。 值得注意的一个反直觉选择：不做 teacher 模型预验证（不像部分工作用强模型试解、解不出就丢弃任务）。理由有二：这会把任务难度上限压到验证器能力，且成本翻倍；而 group-based RL 的在线动态过滤（rollout 无优势的 prompt 自动不产生梯度）天然淘汰过难/过易任务。 # 数据画像：广、贵、且忠实于源 规模与成本：7,971 个任务，用 GPT-5.6 Sol（xhigh 推理档）生成，API 花费超 9 万美元。（脚注：出于法律原因，公开发布的数据集改用 Kimi-K3-max 在同一流水线下生成。） 领域分布：13 个领域中，软件工程仅占 22.5%，AI/ML 10.5%，商业/金融/法律/HR 10.5%，营销 9.3%……论文对比了 TMax-15K、Terminal-Bench、DeepSWE 等，Skill2Env 是唯一全覆盖 13 域、且非技术知识工作占大头的语料。 忠实度探针（很聪明的设计）：用任务指令+量规作查询、对 3.4k 个 SKILL.md 做 TF-IDF 检索，73.2% 的任务 top-1 命中真实源 Skill，94.6% 进 top-10（随机 0.03%）。单用量规也有 68.5% top-1，证明量规携带的是 Skill 专属方法论而非泛泛建议。 SFT 数据：用 GLM-5.3 对每个任务 rollout 两次，得到 15,968 条轨迹，平均奖励 0.74，中位轨迹 19 次模型调用 + 23 次工具调用。 S2EBench：考虑到公开基准饱和，从 SkillHub 另外生成、逐条人工审核（指令无歧义、忠实于源 Skill、测试公允）后的 79 任务私有 held-out 基准。 # RL 实验：基础设施 + 极简配方 基础设施（论文明确说“现代 agentic RL 首先是基础设施挑战”）：Molt（PyTorch 原生全异步训练，Ray + vLLM + FSDP2）+ Polar（agent rollout 层：rootless Apptainer 沙箱、代理回传 token ID 和采样时 log-prob、prefix merging 把 harness 的多次补全缝合成训练轨迹）。 配方（刻意走“简单路线”）：GRPO 组归一优势 + DPPO 的 binary-KL 信任域掩码（δ=0.05，超出阈值的 token 直接丢弃，无需参考模型，还能防训练-推理失配）；G=8 rollouts/组，批 64，lr 1e-6 恒定，无 KL 惩罚、无熵奖励、无 SFT 热启动，每任务 65k 上下文。 量规校准奖励：开量规时，额外由 GPT-6 Astra 做 LLM-as-Judge（带“宪法”：惩罚无脑循环、reward hacking、答非所问；hacking 实证 = -5 分），总奖励 r = r_V + λs/5（λ=0.2），即 judge 最多把程序化奖励拉动 ±0.2。量规是校准可执行结果奖励，而非取代它，这是与“Rubrics as Rewards”一系的定位差异。 # 四项发现（论文最有信息量的部分） 发现 1：小规模 RL 即有跨域迁移。 仅 300 步、只用 2,400 任务子集训一个 epoch：S2EBench pass@1 +4.3（均分 +18.5），Terminal-Bench 2.1 +4.7（49.4→54.1）。训练集与 TB 无重叠（13-gram Jaccard < 0.8），且训练集从未针对 TB 调过，论文将其解读为规划、工具使用、收尾能力的通用提升而非任务族记忆。这让 27B 本地模型显著缩小了与云端前沿模型的差距。 发现 2：量规校准 RL 在基准上落后于纯结果 RL，一个诚实的负结果。 量规版在 TB 2.1 只有 50.1（纯结果版 54.1）；训练中量规版的程序化奖励长期停在 0.5–0.6，judge 分项从头到尾无上升趋势，两个奖励在训练分布上互相拉扯。论文不把它当作对量规奖励的终审判决（两者优化不同目标，而基准只考结果那一半），并给出两个疑因：λ=0.2 的加性形式让失败任务仍能拿正奖励、judge 看不到文件系统等设定均未调优；以及更本质的，Skill 写下的方法论可能本来就不是最大化基准通过率的分布。 发现 3：行为确实向 Skill 对齐，量规的价值所在。 200 个任务的成对偏好测试（judge 拿源 SKILL.md 当标准，比较匿名化的 base 与 RL 轨迹）：纯结果 RL 已被偏好 54.5% vs 33.5%；量规版被偏好 73.0% vs 24.0%。这说明量规奖励买到的东西在结果基准上看不见，但对“怎么做事”影响实质，对网页开发、报告综合、开放研究这类难验证任务尤其重要。 发现 4：GLM-5.3 蒸馏 SFT 反而伤害 Qwen。 在 GLM-5.3 轨迹上做 SFT：27B 上 TB 2.1 掉到 45.8；4B 上直接崩塌（TB 18.7→3.4，出现思维/工具调用死循环）。归因：教师的 interleaved-thinking + 工具调用风格与学生自身 post-training 不兼容，模仿覆盖了学生依赖的行为模式却带不来教师的能力。与 TMax 报告的“SFT 混合数据劣化已后训练的 Qwen”互相印证。因此论文所有 RL 结果都从未修改的原始 checkpoint 出发。

Today at @WeAreDevs, Docker and the @linuxfoundation are announcing a collaboration around the Docker Sandbox Kit Specification: an open standard for declaring what an agent may do, where it may reach, and what it may touch. Open source under Apache 2.0. Read the deep dive:

Introducing Ondo Intelligent Portfolios, the first three portfolios powered by BlackRock. Ondo Intelligent Portfolios introduces a new onchain product category: curated investment portfolios delivered as single onchain transferable tokens. The first three portfolios are based on portfolio strategies developed by BlackRock for Ondo, marking the first time eligible onchain investors can access exposure to such strategies through a single token. 1. BLKHIon: Ondo High Income Powered by BlackRock 2. BLKDIGon: Ondo Diversified Growth Powered by BlackRock 3. BLKGRWon: Ondo High Growth Powered by BlackRock Diversified, professionally constructed strategies have historically required brokerage accounts and traditional fund structures. Now, delivered as peer-to-peer transferable tokens from Ondo, these onchain portfolios become accessible to eligible non-US investors in permitted jurisdictions through the wallets, exchanges, and DeFi applications they already use. “Tokenization creates new ways for portfolio strategies to be delivered through digital infrastructure. Diversified portfolio strategies can be incorporated into tokenized investment products, enabling eligible investors to access diversified allocations through a single instrument. It shows how established portfolio construction approaches can be delivered through new channels and technologies.” - Lisa O’Connor, Global Head of the Model Portfolio Solutions team and Co-CIO for Global Solutions within the Multi-Asset Strategies group at BlackRock Ondo Intelligent Portfolios can unlock novel capabilities: → Programmatic rebalancing → Full composability with DeFi → Complete transparency onchain → Multiple asset classes in a single token This is just the start for Ondo Intelligent Portfolios. The infrastructure is now in place for leading financial institutions to bring their asset allocation expertise onchain.

Tired of the endless scroll? Meet Dreambeans, an experiment from @GoogleLabs that curates personalized collections of stories to inspire your day based on your day based on your enabled connected apps, now including Gemini. Dive deeper into any story, take actions like making reservations or appointments, and bookmark or share your favorites. 🫘 Dreambeans is now available to all accounts in the U.S. (18+) on @Android and iOS. Learn more ↓

It’s officially football season 🏈 We're introducing new features in Search so you can dive deeper into the game — including a Live Game Feed, more detailed stats, and the ability to link your fantasy football account for more tailored recommendations. 🧵↓

Grateful to @cicada_mm for featuring us in their research blog. Dive into the full piece for our thinking on scaling data volume without sacrificing quality, how we prioritize robot embodiments, and what foundation-model scale actually means.

Interesting split in @okx X-Perp activity: Crypto is still highly concentrated: • BTC: 55.6% • ETH: 14.3% • SOL: 5.1% → Top 3 = 75% of crypto volume Equities are much more diversified: • SNDK: 22.8% • SPCX: 18.7% • BEAT: 15.1% → Top 3 = 56.6% Meanwhile, gold dominates metals at 70.7%, while CL accounts for 69.4% of commodity volume. Nice to see OKX evolving into more than just a crypto venue. Try it via my affiliate link: