2026-05-29 · Fri
generated 10:13:00
🌟 Today's Headline
Robinhood launches AI agents for automated stock trading and spending
Robinhood has launched a new beta feature enabling AI agents to automatically execute stock trades through dedicated brokerage accounts. Users can set budgets for agents, approve individual transactions, and establish limits on what actions the AI can take. The platform also introduced virtual cards allowing AI agents to make purchases within specified spending limits. This represents a significant step toward delegating financial decision-making to autonomous AI systems, letting users set high-level objectives while AI handles execution. The beta rollout demonstrates growing adoption of agentic AI in consumer finance, marking the convergence of autonomous AI systems with retail wealth management.
💬 Editor's Note
Robinhood crosses a new threshold: autonomous AI now handles both your investments and spending. As algorithms gain trading authority, the liability question becomes unavoidable. Progress or peril? The answer depends on who pays when the agent errs.
🔥Today's Highlights
10/10
Cognition secured $1 billion funding at $26 billion valuation, more than doubling from $10.2 billion eight months ago. Annual revenue run-rate skyrocketed from $37M to $492M. The breakthrough metric: Devin now writes 89% of Cognition's own codebase, up from 13% in December—eating their own cooking. The new SWE-1.6 model runs at 950 tokens per second, enabling real pair-programming workflows.
10/10
Tech
Anthropic released Opus 4.8, achieving new highs across Every's testing suites. Senior Engineer Benchmark: 63 points vs GPT-5.5's 62 and Opus 4.7's 33.5. Writing tests at high effort: 79.6 vs Sonnet 4.6 (74.5), GPT-5.5 (73), and Opus 4.7 (63), with fewest AI tells. First model to produce one-shot PowerPoint presentations with clear narrative structure.
9/10
New Product
阶跃星辰(Step)发布了开源大模型 Step 3.7 Flash,主打智能体(Agent)工作流的效率。该模型在 ClawEval-1.1(67.1分)和 SimpleVQA Search(79.2分)评测中排名第一。
9/10
New Product
llm-anthropic 发布 0.25.1 版本。主要更新包括:新增 Claude Opus 4.8 (`claude-opus-4.8`) 模型;为账户启用了该功能的组织新增了 `-o fast 1` 选项以使用快速模式;调整了各模型的默认 `max_tokens` 值,使其直接使用模型的最大…
9/10
New Product
不要只依赖基准测试;要看全面情况!
试试我们的新比较页面,它还能让你可视化模型性能:https://openrouter.ai/compare/openai/gpt-5.5/anthropic/claude-opus-4.7/anthropic/claude-opus-4.8
9/10
New Product
Grok Build 0.2.7 现已发布,包含 /usage、/login、跨子智能体共享终端,以及改进的图像理解功能。
所有更新请查看 https://x.ai/build/changelog
📊Topic Clusters
📌 Anthropic融资与生态扩张
Anthropic完成650亿美元H轮融资,同步发布Opus 4.8和Claude Code新功能,生态加速
📌 AI芯片供应链卡位
SK海力士、美光、三星、AMD等大厂纷纷推出AI芯片与高带宽内存新品,缓解行业成本压力
📌 多模态LLM竞争加剧
Google DeepMind、阶跃星辰、xAI、MiniMax等密集发布新模型版本,多模态能力成卖点
📖Worth a Deep Read
🕐 ~8 min read
· Tech
8/10
💡 Detailed technical reference
Biohub, backed by Mark Zuckerberg and Priscilla Chan's Chan Zuckerberg Initiative, released Evolutionary Scale Models (ESM), a suite of state-of-the-art protein language models for structure prediction and design. ESMFold2, trained on 2.8 billion protein sequences, achieves SOTA performance in predicting protein structures and protein-protein interactions, outperforming AlphaFold. Lab validation shows ESMFold2 designed protein binders against five cancer and immune disease targets with hit rates of 36–88%. The system includes ESM Atlas, mapping 6.8 billion protein sequences and 1.1 billion predicted structures to reveal evolutionary connections. This $500M Virtual Biology Initiative aims to compress drug discovery from years to months, directly accelerating therapeutics for cancer and autoimmune diseases.
🕐 ~8 min read
· Tech
8/10
💡 Detailed technical reference
Google DeepMind unveiled a single embedding model handling text, audio, video, and images simultaneously—a major infrastructure consolidation. This unified approach eliminates the complexity of maintaining separate embeddings for each modality, allowing organizations to build multimodal AI systems with less engineering overhead. The model is trained to understand semantics across all four input types with aligned representations. This represents infrastructure-layer maturation, paralleling the shift where agents are becoming products rather than features. For multimodal AI applications, it reduces deployment complexity and potentially improves cross-modal semantic alignment. The release signals that AI infrastructure is standardizing and consolidating across the stack, mirroring similar trends in model development where unified architectures outperform specialized variants.
🕐 ~3 min read
· Tutorial
7/10
💡 Can be adapted into tutorial material
报告显示,AI正深刻改变开发工作形态。开发者周均代码产出从约3.6K行增至8.6K行,更大规模的PR(千行以上)占比上升。AI智能体在单次会话中的工具调用数增加约30%,正在处理更复杂的任务。同时,被接受的AI代码在60分钟后的留存率从约76%提升至约81%,表明更多AI生成内容进入了实际代码库。这些趋势共同指向AI已从个人辅助工具,演进为推动开发向更大规模任务与自动化基础设施发展的核心力量。
🕐 ~3 min read
· Tutorial
7/10
💡 Can be adapted into tutorial material
特斯拉声称其全自动驾驶软件(FSD)安全性最高可达人类的10倍,但路透社调查发现此数据经不起推敲。参与训练FSD的员工表示该技术远未成熟,其安全演示高度依赖人工。统计方法被11位交通安全研究人员指出存在缺陷,例如与更广泛的联邦事故数据进行不恰当比较。相比之下,竞争对手Waymo采用了更严谨的统计方法。目前,特斯拉FSD仍需驾驶员主动监督,安全部署可能还需数年。
🕐 ~3 min read
· Industry
7/10
💡 Industry trends and analysis
📂Browse by Category
New Product
最好的设计工作不会在聊天框里发生。你需要空间来探索想法、创建变体并进行迭代。
认识新的 Replit Canvas。
你的智能体设计工具,用于构建精美的网站、应用、营销资产等。
好消息!印度用户现在可以上传视频(来自相册或已保存文件),并使用Gemini Omni进行编辑和转换。快来试试,并告诉我们你的想法。
在 @OpenHandsDev 上使用 MiniMax M2.7 进行免费智能体编程?
是的,请给我!限时提供 ⚡👀
Industry
Anthropic 宣布完成由 Altimeter Capital 等领投的 650 亿美元 H 轮融资,投后估值达 9650 亿美元。公司表示其旗舰模型 Claude 的企业部署持续增长,年化收入已突破 470 亿美元。
Apple 正尝试将大型 Gemini 模型集成到 iPhone 中,以支持全新的 Siri 功能。由于模型规模庞大,本地处理可能无法完全实现,因此一个云端组件很可能是必然的选择。
Tech
hexoai开源了SIA(自我改进AI)框架。该框架展示了AI智能体不仅能优化其外部工作流(harness),还能通过任务反馈直接更新自身的模型权重,从而在领域知识和能力上实现自主提升,而非仅依赖人类提供的提示或工具改进。
SGLang 与 AMD 团队合作,通过一系列全栈优化,使 AMD InstinctTM MI355X GPU 在运行 DeepSeek-R1 大模型推理时实现了极具竞争力的总拥有成本。在 129 tok/s/user 的交互延迟下,其成本为每百万 token $0.169,比 NVIDIA B20…
Josef Chen and Kaikaku released Epicure, a multilingual ingredient-embedding model trained on 4.1 million recipes across 7 languages, encoding 1,790 ingredients into 300 dimensions—compressing 'all of human cooking' into just 2MB. The Epicure Flavour Explorer lets users explore ingredient pairings interactively, available as a HuggingFace Space and as an MCP endpoint for use in AI agents.
Tutorial
如何用四步保障你的vibecoded应用安全 🔒
速度若无安全加持,便是隐患。以下是使用Replit发布应用时,如何避免留下后门的方法。
🧵展开阅读 ↓
Guide on designing APIs for AI agents, emphasizing that agents cannot read documentation, infer intent, or handle edge cases like human developers. Requires fundamental rethinking of API patterns for agent-first systems.
freeCodeCamp published a six-hour comprehensive course on Figma covering UI/UX design workflow from fundamentals to professional-grade, high-fidelity design.
📭Skip Today
Auto-filtered. Here's why — so you know you're not missing out:
v2.1.156
→ Already covered, no new facts today
Learn Figma for UI/UX Design
→ Score too low (≤4)
How to Build a Hybrid Cloud Platform with Google Cloud Services and On-Premise Kubernetes Infrastructure
→ Score too low (≤4)
0.136.0-alpha.1
→ Minor alpha/beta/rc release, no new feature
Anthropic raises $65B in Series H funding at $965B post-money valuation
→ Already covered, no new facts today
Apple working to cram massive Gemini model into iPhone to power new Siri
→ Already covered, no new facts today
Big release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both …
→ Already covered, no new facts today
📎 Long Tail (3) · click to expand