딸기향립밤

[EndGame Macro @onechancefreedm]

The Oil Market Is Burning Through Its Safety Margin

The most important thing in this chart is not simply that global oil inventories are low.

It is that inventories are still falling after the system has already leaned on emergency reserves and floating storage to absorb the shock.

The red 2026 line sits well below the recent historical range and continues to trend lower. That suggests the market is steadily consuming the buffers that normally protect the system from the next disruption.

The Cushion Is Getting Thinner

The chart shows global onshore inventories around 3.34 billion barrels near week 38, down roughly 420 million barrels from the early 2026 peak and around 220 million barrels below the 5 year average.

That means the amount of flexibility available to absorb further outages is shrinking.

Not every barrel in storage is equally usable. Some inventory is operationally necessary. Some is in the wrong location. Some cannot reach the refinery that needs it quickly enough.

So a relatively modest decline in total inventories can represent a much larger decline in the actual safety margin.

Emergency Supply Has Bought Time

The chart also notes roughly 340 million barrels of global strategic reserve releases and about 140 million barrels drawn from floating storage around Hormuz.

Those numbers should not simply be added together as a precise shortage calculation because some categories can overlap.

But the underlying point is important.

Emergency stocks and tanker inventories can temporarily replace barrels that current production and shipping are failing to deliver.

They buy time.

They do not permanently solve the imbalance.

If the emergency release stops, current production and transportation must eventually replace those barrels or consumption has to fall.

Why This Looks Worse Than 2022

The comparison with 2022 matters.

During the first Russia shock, much of the global oil adjustment came from rerouting Russian exports rather than permanently removing the barrels from the market.

Today the problem is more physical.

More than 10 million barrels per day of Gulf production remained shut in during August, while global refinery throughput was more than 4 million barrels per day below a year earlier.

The system is dealing with lost production, impaired refining and disrupted transportation at the same time.

The U.S. Strategic Petroleum Reserve is also far smaller than it was before the 2022 releases, leaving less emergency capacity available if another major disruption occurs.

The 1979 Lesson

The real danger begins when buyers lose confidence that the next delivery will arrive.

During the 1979 oil shock, precautionary buying and inventory hoarding helped magnify the shortage.

Refiners, distributors and large consumers bought earlier and held more inventory because they feared future scarcity.

Individually those decisions were rational.

Collectively they made the market tighter.

That is the risk when inventories become this depleted.

This Winter Is the Next Test

The concerning part is that the draw may continue into winter.

Distillate inventories remain tight, refinery maintenance reduces output, agricultural demand remains elevated and heating demand rises.

A barrel of crude does not solve that problem by itself.

It still has to reach the right refinery, be processed into the right fuel and then be delivered where it is needed.

That is why crude prices alone can give a misleading sense of relief.

The real question is whether refinery throughput, shipping flows and finished fuel inventories recover before the remaining buffer is exhausted.

Inventories have bought the global economy time.

The red line shows how much of that time has already been used.

[Nix0n @UnTalNixon_exe]

¿Imaginas correr un modelo de 8B especializado 100% en ciberseguridad, sin ningún tipo de censura ni bloqueos molestos, directamente en tu laptop con una GPU de 8GB?

Dolphin3-Cyber-8B (en su versión GGUF optimizada en Q2_K a solo 3.2 GB) acaba de salir y es una bestia para el día a día.

Lo que trae bajo el capó:

• Cero negativas absurdas en preguntas sobre pentesting autorizado.

• Entrenado con un dataset custom: OWASP Top 10, MITRE ATT&CK, CVEs, ExploitDB y metodologías reales de pruebas de intrusión.

• Ideal para code review rápido, resolver CTFs, armar notas de bug bounty y diseñar modelos de amenaza.

(Tip: mantén el contexto cerca de los 2k tokens que fue lo que vio su LoRA para que no alucine).

Herramienta brutal para llevar un asistente ofensivo local sin depender de la nube.

Te dejo el enlace al repositorio y los pesos directos abajo en los comentarios 🛠️👇

[雨夹雪❄️ @mizorewww]

介绍比Jev快50倍，在你设备上跑的laya-mlx!

只在你的设备上占用最高1G内存

Laya是一个开源的类似于Jev的,基于文本输出概率的分类系统

我将其移植到MLX,并且做了一些性能优化!

视频中就是这个模型在我的本地M3Max上玩贪吃蛇

这个模型能够以每秒决策60次的速度玩贪吃蛇！

https://t.co/31KGUiNunb

[SuSu_酥酥👅 @NFT_Chen]

💥炸了！开源本地版 Laya 响应速度吊打 Jev ，毫秒级决策快到飞起！

左边 Laya 是本地 421M 开源决策模型，右边 Jev 1.13.0 是云端 API。

同一盘贪吃蛇、同一套 typed decision，30 秒自由跑下来：

🔹Laya：分数 46、长度 52，决策 86.5 次/秒，P50 约 9ms
🔸Jev：分数 1、长度 7，决策 3.2 次/秒，API 往返 317ms

M5 Pro 实测中位耗时：本地 15.3ms vs 云端 298.1ms，接近 20 倍

Laya 推理内存大约 1GB，毫秒级决策，不靠网络

而云端 Jev 每一步都在等往返，本地每一步都在当场拍板！

GitHub：https://t.co/iksiWcCUbX

#开源Jev #Laya #本地部署 #SystemOne #贪吃蛇对战AI #Jev

https://t.co/4tZHU6kg17

[CV.YH @0xCVYH]

Laya é um open source do Jev que já veio acima dele.

tá cada vez mais rápido

https://t.co/XVyFYPxVht

[Quant Science @quantscience_]

🚨 Someone built an AI that reads candlestick charts the way GPT reads English.

Trained on 12 billion records from 45 exchanges.

Outperforms every model by 93%. Live BTC demo. Free.

[PrismML @PrismML]

Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.

Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.

Ternary Bonsai 2 27B is available today under Apache 2.0.

[宮田 |Cloudflare x ClaudeCode @miyataArcHack]

Cloudflareが公開した「security-audit-skill」が実用的。

AIがコードから脆弱性の候補を探し、別のAIが「本当に問題か？」を検証する仕組み。

結果は「確認済み・要検証・棄却」に分けて記録。調査が足りない箇所も管理できる。https://t.co/8MnrBBMxoS

[Atria ASI @AtriaASI]

Introducing Atria Dawn Preview:
From Research Questions to Verifiable Results.

Atria Dawn Preview is an agentic foundation model built for long-horizon tasks, helping researchers and engineers turn open-ended questions into executable, verifiable, and reproducible outcomes.

Explore Atria Dawn Preview:
🌐 Website: https://t.co/MDDspFiqYO
💻 GitHub: https://t.co/tNmLocadjT
🤗 Hugging Face: https://t.co/cyZod5oHbw
🤖 ModelScope: https://t.co/6BNi5vHFDl
🚀 Try Atria:
EN：https://t.co/Ap5aO2gQeu
ZH：https://t.co/cG16cQdx4j

[Emre Savcı @mstrYoda_]

CloudFlare AI Agentlar için security-skill yayınlamış, direkt yapıştıralım 👇

https://t.co/NMN5oH2ijv

[alex @alextalksai]

Give your ai agent eyes to see the entire internet for free

Read & search
- Twitter,
- Reddit,
- YouTube.
- GitHub,
- Bilibili,
- XiaoHongShu

One CLI, zero API fees.

[سلطان الفردان @SultanAlFardan]

أداة تخليك تفهم أي مشروع برمجي خلال دقائق:

Understand Anything تحلل المشروع بالكامل باستخدام وكلاء ذكاء اصطناعي، وتحوله إلى خريطة معرفة تفاعلية توضح لك:

• الملفات
• الدوال
• الكلاسات
• الاعتماديات
• وكيف كل جزء مرتبط بالثاني

وتقدر بعدها تبحث داخل المشروع وتسأل عن أي جزء من الكود بلغة طبيعية.

مفيدة جدًا إذا استلمت مشروع ضخم وما تعرف من وين تبدأ.

[world paradise @worldparadiseX]

📍 Passo Giau, Italy 🇮🇹

[Tony Simons @tonysimons_]

🚨 Hermes Agent just gained something wild.

Dream Loop.

Give Hermes a visual idea.

It generates concept art.

Builds it.

Takes a screenshot of what it actually built.

Looks at the screenshot.

Judges it against the original concept.

Then keeps working until it gets there.

Utter chaos! 🤘

[山羊赚钱笔记 @shanyanggm]

卧槽兄弟们，GitHub 上真的有一堆免费到离谱的项目。

很多能力已经能直接干掉你正在月付的软件。

1. TradingAgents
AI 多 Agent 量化交易框架
https://t.co/3T1sPPQ6TL

2. LibreChat
一个界面接入 ChatGPT、Claude、Gemini 等多个模型
https://t.co/HZJQEij5h1

3. HyperFrames HeyGen
开源的视频生成引擎
https://t.co/o0OnS8iJMG

4. Fincept Terminal
开源版彭博终端
https://t.co/n5COGtvHc4

5. MoneyPrinterTurbo
AI 一键生成短视频 https://t.co/Po8sGObI7q

6. Agentic Inbox Cloudflare
开源 AI 邮件助手
https://t.co/S7XDWi1yTD

7. VoxCPM
AI 声音克隆工具
https://t.co/wkSGMchuq2

8. Flowsint
开源 OSINT 情报分析工具 https://t.co/RwFvhurB0P

9. agent-skills
Claude Code 技能库 https://t.co/c2NifSzKM1

10. Nango
开源 API 集成平台
https://t.co/tNbwrMEzZn

兄弟们，这些真不是玩具项目。

很多你还在交月费的软件，GitHub 上已经有人做出开源平替了。

一句话：

别只会收藏 AI 工具网站了。

真正狠的东西，很多都藏在 GitHub 里。

[山羊赚钱笔记 @shanyanggm]

10个GitHub仓库，免费得不像话！

1.AutoHedge
四个Agent搭一套自动交易：定方向、验证、风控、下单。原帖写着pip装上就能跑。交易有风险，当工具看，别当稳赚。
https://t.co/gD9ini2FVQ

2.Vibe-Trading
一堆金融Skill和专家Agent一起吵策略，你能看着它们辩论。
https://t.co/bXMHTsC6Ou

3.Fincept Terminal
笔记本上的彭博平替。多数据源、多投资人风格的Agent。彭博一年几万刀，这个开源。
https://t.co/n5COGtv9mw

4 LibreChat
ChatGPT、Claude、Gemini、DeepSeek一个界面。自己架，数据在自己手里。
https://t.co/HZJQEiixrt

5. Open Higgsfield AI
两百多个生成模型塞进一个本地工作室。文生图、图生视频，订阅可以先停。
https://t.co/B4mnmqGhX6

6.Open-LLM-VTuber
离线Live2D助手。能看屏幕、听声音、记住你。心里那层思考会单独显示出来。
https://t.co/6LThgqciAU

7.Claude Ads
给Claude Code用的广告体检。谷歌、Meta、YouTube、LinkedIn、TikTok、微软广告一起查。代理公司按月收这个钱。
https://t.co/bqijnsprRn

8.Agentic Inbox
Cloudflare开源邮箱。Agent读信、拟回复，邮件留在你自己的账号里。
https://t.co/S7XDWi1145

9.Camofox Browser
给Agent用的无头浏览器，往不好被当成机器人的方向做。输出无障碍树，token能少一截。
https://t.co/aXBsNFu3hP

10.Hyperframes
HeyGen开源的视频框架。Agent写HTML，它吐MP4。同样的HTML，出来的片子一样。
https://t.co/o0OnS8ibX8

这些不是玩具！

你还在月付的，这儿不少有平替。

先装一个，塞进自己的流程里。

[Sandhya @agenticgirl]

Alibaba open-sourced the code reviewer it says has served tens of thousands of developers and found millions of defects internally. Open Code Review does not leave the whole review loop to an LLM: deterministic code handles file coverage, bundling, rule matching and comment positioning, while the agent handles reasoning and repository context. On its benchmark of 200 PRs across 50 open-source repos, Alibaba reports higher precision and F1 than Claude Code with the same model, using roughly 1/9 the tokens, with lower recall as the trade-off. GitHub Repo: https://t.co/vSXeaq0DzQ

[Smartpig @Smartpigai]

这可能是我见过最离谱的金融数据库之一。

FinanceDatabase 收录了 30万+ 金融资产：

股票、ETF、基金、指数、外汇、加密货币……

还能按国家、行业、交易所、市值直接筛选。

做美股研究、量化、金融数据分析的，建议收藏：

https://t.co/d9zeNwAbzl

一行安装：

pip install financedatabase

[章工 @435hz]

推荐Agent-Reach，专门解决 Agent 联网找资料难的问题。
https://t.co/HGWXiq92ZG
以前让 Claude Code / OpenClaw / Hermes Agent 抓资料，经常卡在 API、订阅、反爬、平台限制上。
它直接把 yt-dlp、twitter-cli、xhs-cli 等工具打包好，网页、YouTube、RSS、公众号、微博、小红书、V2EX 都能跑。

[Zibá @zibxx]

Tras 3 años usando IA he escrito más de 15.000 prompts.

Solo uso 5 cada semana. Sin excepción.

Estos 5 nunca los borro. Funcionan para TODO.

Uno para decidir. Uno para escribir. Uno para aprender. Uno para planificar. Uno para crecer.

Los 5 mega-prompts que necesitas para tu vida entera 🧵👇

(Si solo guardas un hilo este año, que sea este 🚨).