03-28 00:00 · 国际,新闻,BBC,国际
Noelia Castillo died on Thursday evening in a Barcelona hospital, after a protracted legal battle with her own father.
Spanish woman who died through euthanasia failed by state, say critics
Warning: This story contains references to sexual assault and suicide
The death by euthanasia of a 25-year-old Spanish woman after a protracted legal battle with her father has triggered debate about the role of the state in caring for her and why it took so long to implement her wish to end her life.
Noelia Castillo, who had been left paraplegic due to injuries suffered when she tried to take her own life in 2022, died on Thursday evening at a Barcelona hospital.
The Catalan regional government had granted her the right to assisted dying in 2024.
However, the process was suspended at the last moment after legal objections raised by her father, backed by campaign group Christian Lawyers.
The case has received enormous attention in Spain, with Christian Lawyers (Abogados Cristianos) attempting to block her death until the last moment.
After an 18-month legal battle, the European Court of Human Rights (ECHR) ruled this week in Noelia Castillo's favour and her death was eventually confirmed late on Thursday.
Castillo had spent much of her childhood in care homes and had recounted the impact on her mental health of her father's problems with alcohol and of being sexually assaulted by an ex-boyfriend and also by several men in a nightclub.
In a TV interview this week, she said nobody in her family had supported her decision to die by euthanasia, and her father "hasn't respected my decision and never will".
"I want to go in peace now and stop suffering," she told Antena 3 TV the day before she died.
Her mother had disagreed with her decision but joined her at the Sant Camil Barcelona clinic.
A former friend of Castillo, Carla Rodríguez, tried to enter the hospital to persuade her to change her mind, but told Spanish media that police had barred her from entering.
British pianist James Rhodes, who lives in Spain, issued an appeal to Castillo via social media to reconsider and offered to pay her medical costs until she felt "able to take this decision from a slightly more tranquil place".
Christian Lawyers warned that her case had underlined failures in her care.
"For a girl who obviously has had a very tough life, which we all regret, the only thing that could be offered to her by the healthcare system is death," said José María Fernández, of Christian Lawyers.
The opposition conservative People's Party (PP), which voted against a 2021 euthanasia law, had a similar response.
"The institutions that should have protected Noelia failed her," wrote PP leader Alberto Núñez Feijóo on social media. He added: "I refuse to believe that the state did not have the tools to give her care."
In a statement, the Catholic Church, which is close to the PP, said that Castillo's story "reflects an accumulation of personal suffering and institutional failures".
However, other observers have taken issue with the fact that her wishes were thwarted by legal obstacles put in her way by her father and Christian Lawyers.
"The desire to put an end to her suffering by using the right to euthanasia was… sabotaged by a legal crusade that added nearly two years of pain to her existence," noted the left-leaning El País newspaper in an editorial.
Alberto Ibáñez, a member of Congress for the left-wing Sumar platform, said that "19 doctors have supported her decision and we should be respectful of it", while adding that it was a "deeply complex" issue.
Spain is one of a handful of European countries, including the Netherlands, Belgium and Luxembourg, that have passed laws allowing euthanasia to be carried out by physicians.
Under Spain's 2021 law, any Spanish adult over 18 requesting euthanasia must be suffering from an incurable disease or "serious, chronic and disabling condition" and their decision must be taken free of external pressure.
The decision needs to have been made twice in writing and has to be certified by a doctor who then consults with another doctor.
The request then passes to a Guarantee and Evaluation Commission, which has the task of assessing if the conditions have been met.
According to government data, 426 requests for assisted dying were granted in 2024, the most recent year for which data is available. This was the first time that a case went to court for a judge to decide.
▸ 展开全文
03-28 00:00 · 国际,新闻,BBC,国际
Tiger Woods is charged with driving under the influence after rolling his car in a crash in Florida, police confirm.
Golf legend Tiger Woods has been released on bail following his arrest on charges of driving under the influence after rolling his car in a crash in Florida.
The Martin County Sheriff's Office said the 15-time major champion was also charged with property damage and refusal to submit to a lawful test.
Woods, 50, rolled his Land Rover after clipping a truck in Jupiter Island on Friday afternoon, according to Sheriff John Budensiek, who said officers had not suspected him of impairment by alcohol but possibly by an unknown substance.
Woods, who had to crawl out of the passenger door, passed a breathalyser test after the crash before refusing a urine test. No-one was injured.
He was kept in jail for eight hours - the minimum allowed under state law - before being released on bail.
Budensiek said the crash happened after Woods allegedly attempted to overtake a pressure cleaner truck at a "high rate of speed".
DUI investigators came to the scene and "Mr Woods did exemplify signs of impairment", Budensiek said.
"They did several tests on him," Budensiek said. "He did explain the injuries and surgeries that he's had and we did take that into account, but they did some in-depth roadside tests."
Woods was placed under arrest after the incident, which took place on Beach Road just before 14:00 local time (19:00 GMT).
Budensiek said investigators at the scene and at jail were "really not suspicious of alcohol being involved" and Woods passed a breathalyser test with "triple zeros".
Budensiek said Woods was "co-operative but was trying not to incriminate himself".
"He has a right to refuse that test," Budensiek said.
"There is a statute which he will be charged with for refusing to take that test, but we will never get definitive results as to what he was impaired on at the time of the crash."
US President Donald Trump said: "I feel so badly. He has got some difficulty. There was an accident. That's all I know.
"He's a very close friend of mine. He's an amazing person - an amazing man."
The charges against Woods are misdemeanours, not felonies.
The BBC has contacted his representatives for comment.
More trouble for Woods
Woods has played a limited schedule since the serious car crash in 2021 that left him with extensive injuries and fortunate to be alive.
In 2009 he hit a fire hydrant, a tree and several hedges in a bizarre collision outside his home.
The incident sparked accusations of extramarital affairs which led to the end of his seven-year marriage and the loss of lucrative sponsorship deals.
In 2017 police officers found Woods slumped at the wheel of his parked Mercedes-Benz not far from his Florida home.
A toxicology report found he had several legal medications in his system and marijuana's active ingredient.
He was sentenced to a year's probation after pleading guilty to reckless driving.
Another setback to Woods' comeback plans?
Woods has been recovering from rupturing an Achilles tendon in March 2025 and back surgery in October.
This week he competed in the TGL indoor league finals - his first competitive golf for more than a year - and had not ruled out playing in next month's Masters, a tournament he has a huge affinity for.
Woods has triumphed five times at Augusta National, which was the scene of his first major title in 1997 and his most recent in 2019.
That victory was arguably his greatest achievement in golf given the number of injuries he had to overcome.
However, the sight of him grittily limping through four rounds on that same course in 2022, 14 months after a life-threatening crash, also lives long in the memory.
Woods, whose 82 career wins are level with Sam Snead for the PGA Tour record, has not competed at a major since missing the cut at The Open in July 2024.
He last made the cut at the 2024 Masters, where he finished 60th.
'Playing in Masters looks increasingly unlikely'
This is another huge setback for one of the all-time great sporting figures, who has suffered massive injury problems for the best part of two decades.
Woods is battling to be fit for the Masters, which starts in less than two weeks.
We saw him hitting golf shots competitively in the indoor TGL league at the start of this week - that was the first time since July 2024 we had seen him playing golf.
He did not seem overly optimistic that he would be fit, but he was really trying to be at Augusta for the Masters.
That looks ever more unlikely given the scenario that is playing out in Florida.
▸ 展开全文
03-28 00:00 · 国际,新闻,BBC,国际
A hacker group shares Patel's purported resume and photos - information, the FBI says, is "historical in nature".
Iran-backed hackers breach FBI director Kash Patel's personal emails
FBI Director Kash Patel's personal email account has been hacked by an Iran-linked group, the agency has confirmed.
A group, known as the Handala Hack Team, shared Patel's purported resume and photos of him on its website on Friday along with a statement that says: "This is just our beginning."
The FBI said it was aware of "malicious actors" targeting Patel's email information. "The information in question is historical in nature and involves no government information."
The agency is offering up to $10m (£7.5m) for information that helps in identifying members of the Handala group.
Iranian-backed hackers were reported to have breached Patel's private communications in 2024, weeks before he was appointed to lead the FBI. It is not clear if that breach was different from the one claimed by the Handala group on Friday.
Photos Handala claims to have taken from Patel's email account have been circulating on social media with the group's logo added as a watermark.
The photos show Patel at various unidentified locations, including standing beside a vintage convertible, smiling next to a jet, smoking and sniffing cigars, taking a selfie next to a bottle of liquor, and posing in what appear to be restaurants and hotels.
The BBC has not independently verified the leaked documents.
Cynthia Kaiser, senior vice-president at Halcyon Ransomware Research Center, told the BBC that Friday's release was likely from a historical breach.
"The emails look very old and that makes me believe that this is likely a compromise that occurred from other groups in another time period, and is recycled today," Kaiser, who has worked at the FBI's Criminal, Cyber, Response, and Services Branch, said.
The Handala group said in its statement announcing the hack that the "so-called 'impenetrable' systems of the FBI were brought to their knees within hours by our team. This is the security that the US government boasts about?! This is the cyber giant that thinks threats and bribes can silence the voice of resistance?!"
Experts say this kind of operation on a senior US government official may not take much sophistication to achieve.
"Personal accounts don't have the same level of protection and alerting as government systems, so these are often an attractive target for hackers," said Dave Schroeder, director of National Security Initiatives at the University of Wisconsin–Madison.
"Handala consistently tries to gain this type of access because it serves their interests to claim hacks of prominent people and organizations," Schroeder added.
Last week, the US justice department seized several Handala domain names it says were involved in hacking schemes linked to the Islamic Republic of Iran.
The department said Iran's Ministry of Intelligence and Security (MOIS) had been using the Handala websites to spread "terrorist propaganda", conduct "attempted psychological operations targeting adversaries of the regime", claim credit for hacking activity, and call for the killing of journalists and dissidents.
The domain used to carry out the hack against Patel was registered the same day the justice department announced it had seized the four domains associated with the group, on 19 March, CBS News, the BBC's US partner, reported.
Handala said its hacking of Patel's email account was in retaliation for the FBI's seizure of its websites, as well as for the FBI offering a reward of $10m for information on similar malicious attacks.
The incident saw the company's employee login defaced with a message claiming data had been erased in a "wiper" attack by the Iran-backed group of hacktivists.
In a post at the time on their now-suspended X account, Handala claimed it had wiped "over 200,000 systems, servers and mobile devices", and extracted "50 terabytes of critical data".
The group said the Stryker cyber-attack was "in retaliation for the brutal attack" on an Iranian girls' school at the start of the war, in which more than 160 people were killed, as well as "in response to ongoing cyber assaults against the infrastructure" of Iran and its allies.
▸ 展开全文
03-27 05:47 · AI,技术,HackerNews,AI
Chroma Context-1: Training a Self-Editing Search Agent
Using search systems in conjunction with a large language model (LLM) is a common paradigm for enabling language models to access data beyond their training corpus. This approach, broadly known as retrieval-augmented-generation (RAG), has traditionally relied on single-stage retrieval pipelines composed of vector search, lexical search, or regular expression matching, optionally followed by a learned reranker. While effective for straightforward lookup queries, these pipelines are fundamentally limited: they assume that the information needed to answer a question can be retrieved in a single pass.
In practice, many real-world queries are not satisfiable in a single-stage. Answering a question often requires a chain of intermediate searches in which the output of one search informs the next, a process known as a multi-hop retrieval.
To solve this, leveraging LLMs for multi-turn agentic search has become a viable approach to answering multi-hop retrieval queries. Rather than issuing a single query, an LLM agent iteratively decomposes a high-level question into subqueries, retrieves evidence, and refines its search strategy across multiple turns. Concurrently, it has been shown that smaller-parameter language models, trained on moderate-scale corpora, can serve as effective search agents with performance comparable to substantially larger models. Running frontier-scale models for multi-turn search incurs high cost and latency, which motivates offloading this task to a smaller, purpose-trained model.
A key factor driving the cost and latency of agentic search is the growth of the context window. As the agent gathers information over multiple turns, its context window fills rapidly with retrieved documents, many of which may be tangential or redundant. This bloated context not only increases computational cost but can also degrade downstream performance due to increasing the presence of distracting information. One promising direction to address this is self-editing context, in which the agent actively decides which retrieved information to retain and which to discard, allowing it to continue long-horizon search tasks more efficiently and more accurately within a bounded context window.
Building on these insights, we trained Chroma Context-1, a 20B parameter agentic search model on over eight thousand synthetically generated tasks. Context-1 achieves retrieval performance comparable to frontier LLMs at a fraction of the cost and up to 10x the inference speed. Context-1 operates as a retrieval subagent: rather than answering questions directly, it returns a ranked set of supporting documents to a downstream answering model, cleanly separating search from generation. The model is trained to decompose a high-level query into subqueries and iteratively search a corpus across multiple turns. As the agent's context window fills, it selectively discards irrelevant results to free capacity and reduce noise for further exploration.
In this work we present our synthetic data generation pipeline, agent harness, and training methodology alongside a comprehensive evaluation of Context-1 across a range of retrieval benchmarks. Our results demonstrate that a purpose-trained 20B model can reach the Pareto frontier of retrieval performance with respect to cost and latency, matching or exceeding frontier models that are orders of magnitude larger at a fraction of the compute.
We present the following:
- A staged training curriculum that first optimizes for recall before shifting toward precision, training the agent to progressively narrow from broad retrieval to selective retention. We release the weights of this model to the public under a permissive Apache 2.0 license.
- A context management strategy in which the agent selectively edits its own context during search, discarding irrelevant passages to free context capacity for further exploration and to reduce the effects of context rot.
- A scalable synthetic task generation pipeline that uses a human-aligned LLM judge to minimize the need for human annotation while maintaining task quality. We release the full codebase for this pipeline to support reproducibility and further research.
The limitations of single-shot retrieval have driven substantial exploration into agentic search systems, in which reasoning is interleaved with retrieval to resolve queries that require satisfying multiple constraints jointly or following a chain of dependent clues across documents. These systems vary in their termination strategy: some run for a fixed number of turns, while others terminate dynamically based on a learned sufficiency signal. By shifting control of the retrieval strategy to the model itself, these systems can reformulate queries based on intermediate results, decide when to explore versus exploit, and terminate search based on a confidence assessment. These systems model search as a sequential reasoning task, in which the right next query depends on what has been found so far. Be
▸ 展开全文
03-27 05:47 · AI,技术,HackerNews,AI
Agent-to-agent pair programming
What if you could let Claude and Codex work together as pair programmers, talking to each other directly? One of them as the main worker and the other as a reviewer.
It is amusing how the best agentic workflows often look a lot like human collaboration. Researchers at Cursor discovered this in their work on long-running coding agents. That work led them to create a multi-agent workflow with a main orchestrator assigning tasks to workers. This is similar to how most human teams operate. Claude Code “Agent teams” and Codex “Multi-agent” features work similarly, with subagents reporting back to the main agent. And in the future, subagents could interact with each other, like humans do.
I wanted to pursue the idea of mimicking human collaboration with multiple agent harnesses and another workflow used by programmers: pair programming. While building a code review agent using Claude and Codex side-by-side, I found something interesting: they gave different feedback -- but even when they gave the same feedback, it wasn’t annoying: it was in fact a very strong signal. Our team addresses 100% of the feedback when both reviewers agree. Code reviews are great because they happen on a multiplayer app where humans and agents collaborate, but they are slowing down the feedback loop and can become noisy.
That’s why I built loop
: a dead-simple CLI that launches claude
and codex
side-by-side in tmux, with a bridge that lets them talk to each other. It makes this feedback loop faster and more natural, while preserving context across iterations. It’s interesting because it enables the agents to be more proactive, since the interaction between them is more natural (and I expect that to only get better as the models get better too). Because loop
runs the interactive TUIs, you can stay in the loop, steer, answer questions, and follow up if needed.
The future of agentic workflows may look less like magic automation and more like familiar teamwork. And I’m sure that there are some great observations to apply to this pair programming workflow. Some open questions around how to make the human handoff and PR review easier:
- Should we split the work across multiple PRs?
- Should we share the PLAN.md in git or in the PR description?
- Should we share a screenshot or video recording as a proof of work?
Letting the agents loop can result in more changes than expected, which are usually welcome -- but unfortunately it makes the human review harder.
A lot of people are using multiple agent harnesses for a variety of reasons: to avoid vendor lock-in, to use and contribute to an open-source project, to max out their subscriptions, or to get different perspectives, strengths, and results. Multi-agent harness apps should probably treat agent-to-agent communication as a first-class feature. I’d love to see them adopt this approach.
Try it out: https://github.com/axeldelafosse/loop
Thanks to Léna Deloizy Delafosse, Will Horn, Tian Wang and Ferruccio Balestreri for reading drafts of this.
▸ 展开全文
03-27 05:47 · 中东,局势,BBC,中东
The price of a wide range of goods - from food, to smartphones, to medicines - could be affected by the US-Israel war with Iran.
How the Hormuz closure could affect food, medicines and smartphones
The interruption of oil and gas supplied through the Strait of Hormuz due to the US-Israel war with Iran has dramatically pushed up global energy prices.
Petrol has gone up already and UK domestic heating bills are almost certain to follow.
But it's not just fuel that's been impacted by the conflict. A host of other vitally important chemicals, gases and other products normally enter international supply chains via the Hormuz Strait.
BBC Verify has found that the price of a host of goods - ranging from food, to smartphones, to medicines - could be affected, as the number of ships passing through the Hormuz Strait has dropped from well over 100 a day before the war to just a handful.
Here is what could be impacted.
Fertilisers (Food)
Petrochemicals are derived from oil and gas and they are produced in great quantities for export by countries in the Gulf region.
And one of the most important is fertiliser, vital for global agricultural production.
Analysts have warned that a shortage of these ferilisers is likely to be particularly damaging to agricultural production now because March and April are the northern hemisphere's planting season and less fertiliser use now by farmers will impact yields for later in the year.
"A relatively brief closure could disrupt an entire growing season, with food security consequences that persist long after the strait reopens," according to researchers at the Kiel Institute.
The Institute's work suggests a full closure of the Strait of Hormuz could push up global wheat prices by 4.2% and fruit and vegetable prices by 5.2%.
Helium (Microchips)
It's a byproduct of the production of natural gas and is used in the manufacture of semiconductor wafers, which are then processed into the microchips used in computers, vehicles and household appliances.
Helium is also used to cool the magnets in Magnetic Resonance Imaging (MRI) scanners used in hospitals.
Qatar's giant Ras Laffan plant, which produces the gas, has shut down production after Iranian missile and drone strikes.
And the Qatar government has warned it will take three to five years to repair the damage, raising fears about supplies.
"MRI machines require somewhere between 1,500 to 2,000 liters helium to cool the magnets. Every time you do a scan, a little bit of that boils off or evaporates.
"People like to think helium's predominant use is in data centers, semiconductors and cooling for the AI and data industry. But we can't forget that helium is quite important for MRIs and for other medical users," he told BBC Verify.
Petrochemical derivatives (Medicines)
Derivatives from petrochemicals - such as methanol and ethylene - are vital materials in the global production of pharmaceuticals, including painkillers, antibiotics and vaccines.
The countries of the Gulf Co-operation Council - Saudi Arabia, Qatar, Oman, the United Arab Emirates, Kuwait and Bahrain - are estimated to account for around 6% of global petrochemical production capacity.
These nations primarily use the Strait of Hormuz to export these chemicals to the wider world, with around half going to Asia.
Sulphur (Metals/Batteries)
Its main use is as an agricultural fertiliser, but it is also vital for metal processing.
Sulphur is used to make sulphuric acid, which is used for processing copper, cobalt and nickel and also for the extraction of lithium.
All of those metals are needed for the production of batteries, which are used in everything from domestic appliances to electric vehicles to military hardware like drones.
Analysts warn that if sulphur supplies remain interrupted, the result will likely be higher prices for consumers of products containing batteries.
Additional reporting by Tom Edgington and Joshua Cheetham
Graphics by Phil Leake
▸ 展开全文
03-27 00:04 · 科技
Claude Code skills for fetching curated content from high-quality sources. AI news, mental models, GitHub trending, and more.
A collection of Claude Code skills for fetching curated content from high-quality sources.
# List available skills
npx skills add geekjourneyx/mind-skills --list
# Install all skills
npx skills add geekjourneyx/mind-skills --all
# Install specific skill
npx skills add geekjourneyx/mind-skills --skill ai-daily
npx skills add geekjourneyx/mind-skills --skill fs-street
npx skills add geekjourneyx/mind-skills --skill trending-skills
npx skills add geekjourneyx/mind-skills --skill github-topics
# Install to specific agent
npx skills add geekjourneyx/mind-skills --all -a claude-code
Fetch AI industry news from smol.ai RSS feed.
昨天AI资讯
2026-01-13的新闻
pip install feedparser requests
Fetches articles from Farnam Street blog, covering mental models, decision-making, and learning.
昨天的文章
今天的FS文章
2024-06-13的文章
pip install feedparser requests
Fetches skills.sh trending rankings via Playwright.
今天技能排行榜
Top 10 skills
技能榜单
# Basic rankings
pip install playwright
playwright install chromium --with-deps
# For skill details (optional)
pip install beautifulsoup4 lxml requests
Fetches GitHub topic trending repositories via GitHub API.
今天 claude-code 话题排行榜
Top 10 GitHub 项目
热门仓库
anthropics/claude-code 介绍
pip install requests
Note: With token: 5,000 requests/hour; Without token: 60 requests/hour.
mind-skills/
├── README.md
├── CHANGELOG.md
├── CLAUDE.md
├── skills/
│ ├── ai-daily/
│ │ ├── SKILL.md
│ │ └── scripts/
│ │ └── fetch_news.py
│ ├── fs-street/
│ │ ├── SKILL.md
│ │ └── scripts/
│ │ └── fetch_blog.py
│ ├── trending-skills/
│ │ ├── SKILL.md
│ │ └── src/
│ │ ├── config.py
│ │ ├── skills_fetcher.py
│ │ └── detail_fetcher.py
│ └── github-topics/
│ ├── SKILL.md
│ └── src/
│ ├── config.py
│ ├── github_fetcher.py
│ └── readme_fetcher.py
└── requirements.txt
Design Philosophy: Stateless skills following Unix principles
┌─────────────────────────────────────────────────┐
│ AI Assistant Layer │
│ (Scheduled tasks, persistence, analysis) │
└─────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────┐
│ Stateless Skills │
│ ┌──────────┐ ┌───────────┐ ┌─────────────┐ │
│ │ ai-daily │ │ fs-street │ │ trending │ │
│ │ RSS获取 │ │ RSS获取 │ │Playwright│ │
│ └──────────┘ └───────────┘ └─────────────┘ │
└─────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────┐
│ External Data Sources │
│ smol.ai │ fs.blog │ skills.sh │ GitHub │
└─────────────────────────────────────────────────┘
如果该项目帮助了您，请作者喝杯咖啡吧
- 作者:
geekjourneyx
- X (Twitter): https://x.com/seekjourney
- 公众号: 极客杰尼
关注公众号，获取更多 AI 编程、AI 工具与 AI 出海建站的实战分享：
MIT
- Fork the repository
- Create your feature branch
- Follow the existing skill structure
- Submit a pull request
- Skills Directory - Discover more skills
- Vercel Skills CLI - Official CLI
- Agent Skills Spec - Specification
▸ 展开全文
03-27 00:04 · 科技
Agent Trending是一个聚焦 AI Agent 的开源研究项目，系统整理了 2025.01—2026.03 期间 8701 篇学术论文，围绕 15 个核心方向完成分类、去重、标注与趋势分析。项目包含月度论文索引、细分方向论文集、15 份技术趋势报告，以及 1 份全局趋势总报告和 1 份下一代 Agent 原型设计文档，适合研究者、工程团队和产业决策者快速把握 AI Agent 的技术演进与落地脉络。
Agent Trending 是一个面向 AI Agent 领域的大规模论文研究与趋势分析开源项目。
适合谁读
基于 15 个细分方向、8701 篇学术论文的跨领域深度分析报告。
过去 15 个月（2025.01—2026.03），AI Agent 领域经历了一次历史性的范畴跃迁，从"更聪明的聊天机器人"变成了"能在真实世界执行复杂任务的自主系统"。我们系统性地追踪了这一过程中 7700+ 篇学术论文的演进脉络，覆盖了从基础训练、认知架构、工具使用、多智能体协作到安全对齐、商业落地的完整技术栈，最终凝练为这份全景式趋势报告。
Agent 领域的论文月产出在一年间从约 200 篇飙升至 700+ 篇，信息过载让从业者很难建立全局认知。我们希望通过这份开源报告，帮助研究者快速定位高价值研究方向，帮助工程团队做出更有依据的技术选型，帮助商业决策者理解真实的技术成熟度而非营销叙事。
报告识别出 八大跨领域共振趋势：
- 强化学习全面取代监督微调，成为 Agent 能力获取的核心引擎
- 分层闭环架构成为系统设计的事实标准
- MCP / A2A 等通信协议催生"Agent 互联网"基础设施雏形
- 安全性从附加特性升级为系统生存前提
- 记忆与认知能力成为 Agent 从工具进化为助手的分水岭
- 软件工程成为全领域的首要验证场景
- 评测范式从结果打分转向过程审计
- 物理世界与数字世界 Agent 加速融合
报告同时给出了商业化冲击波排序、五大共性瓶颈分析，以及面向研究者、技术团队和投资者的差异化战略建议。
下表汇总了 15 个细分方向的论文集、技术趋势报告，以及从各方向执行摘要中提炼出的核心趋势、面临挑战与未来推演。
下表按月份汇总论文收录情况，并给出对应阶段的一句话趋势演化总结。
本仓库主要基于公开可获取的学术论文信息、公开链接以及作者的研究整理与分析撰写而成。
- 论文标题、摘要信息、链接等原始内容的版权归原作者及原出版方所有
- 本仓库中的综述、归纳、结构化整理与趋势判断为仓库维护者的原创工作
- 如你认为仓库中存在不适合公开展示的内容，请通过 Issue 或邮件联系处理
当前版本以资料整理和结构化发布为主，后续可能继续补充：
- 新月份追踪
- 方向报告迭代
- 导航页和索引页
- 英文化摘要
欢迎通过 Issue 或 Pull Request 参与改进，例如：
- 修正文档错误
- 补充遗漏论文
- 优化目录和导航
- 补充引用与来源说明
具体约定见 CONTRIBUTING.md。
如果这个仓库对你的研究或写作有帮助，欢迎在引用或转载时注明仓库地址与原作者信息。建议引用仓库中的具体文件，而不是仅引用仓库名。
本仓库采用 CC BY 4.0 许可协议。你可以转载、改编和分享，但需要保留署名。
▸ 展开全文
03-27 00:04 · 科技
Rust trading bot for Polymarket prediction markets. Trades BTC and ETH binary (Up/Down) markets (15m and 1h) using outcome prices (win probabilities from Gamma) and token price trend. Places batches of limit orders on the side that is both winning by outcome price and rising in price over N datapoints.
Automate your NBA market execution on Polymarket with a momentum workflow tuned for live basketball games, tight order books around tip-off and in-game swings, and disciplined trade management.
Private bot for serious traders who want a professional edge without babysitting charts during 48-minute games and overtime.
▸ 展开全文
03-27 00:04 · 科技
This project is an AI-powered anomaly detection system that monitors social media trends, physical environmental conditions, and stock market behavior in real time. It correlates signals across multiple domains using machine learning and LLMs to detect abnormal patterns and explain why anomalies occur and what they may indicate.
This contains everything you need to run your app locally.
View your app in AI Studio: https://ai.studio/apps/temp/3
Prerequisites: Node.js
- Install dependencies:
npm install
- Set the
GEMINI_API_KEY
in .env.local to your Gemini API key - Run the app:
npm run dev
▸ 展开全文