飯田 凡 retweeted
まさか、ハッカーを使って情報を書き換え、薬不足を無かったことにしようとは...🥺
【速報】厚労省の医薬品情報システムに不正アクセス
47news.jp/15250589.html?utm_…
飯田 凡 retweeted
Battery-Free RoboBee Flies Using Only Sunlight in Historic First
by @sutoroveli_news
#Robotics #SolarPower #Innovation #Engineering #FutureTech
飯田 凡 retweeted
𝗢𝗔𝘂𝘁𝗵 𝘃𝘀 𝗔𝗣𝗜 𝗞𝗲𝘆 𝘃𝘀 𝗕𝗮𝘀𝗶𝗰 𝗔𝘂𝘁𝗵 𝘃𝘀 𝗺𝗧𝗟𝗦.
Do you know the differences?
They all help secure access, but what they actually verify, and how they do it, is very different.
𝗢𝗔𝘂𝘁𝗵 𝟮.𝟬 uses scoped access tokens to let applications access resources without sharing user passwords. Tokens can expire and access can be revoked.
𝗔𝗣𝗜 𝗞𝗲𝘆𝘀 use a shared secret to identify and authenticate the calling application. They're simple to implement, but the key itself doesn't prove who holds it.
𝗕𝗮𝘀𝗶𝗰 𝗔𝘂𝘁𝗵 sends a username and password with every request. It's straightforward, but credentials must be protected with HTTPS.
𝗺𝗧𝗟𝗦 uses certificates to authenticate both the client and server during the TLS handshake. It's useful for service-to-service communication where both sides need to verify identity.
But choosing an authentication method is only part of the problem. You also need to manage credentials throughout their lifecycle.
Take CI/CD pipelines. A deployment might depend on a long-lived API key or deploy key stored in its secrets.
Even if it's stored securely, that credential can still be leaked, copied, or forgotten. And rotating it can break deployments that depend on it.
One approach is to replace static credentials with 𝘀𝗵𝗼𝗿𝘁-𝗹𝗶𝘃𝗲𝗱, 𝗶𝗱𝗲𝗻𝘁𝗶𝘁𝘆-𝗯𝗮𝘀𝗲𝗱 𝗮𝗰𝗰𝗲𝘀𝘀.
Instead of retrieving a stored secret, the pipeline proves its identity and receives a credential scoped to what it needs. That credential expires automatically.
Teleport Machine ID implements this approach by letting CI/CD runners use their platform identity to obtain short-lived certificates instead of storing deploy keys.
This reduces the need to manage long-lived credentials while keeping access tied to specific machine identities.
𝗟𝗲𝗮𝗿𝗻 𝗵𝗼𝘄 𝘁𝗼 𝗲𝗹𝗶𝗺𝗶𝗻𝗮𝘁𝗲 𝘀𝘁𝗮𝘁𝗶𝗰 𝗰𝗿𝗲𝗱𝗲𝗻𝘁𝗶𝗮𝗹𝘀 → lucode.co/teleport-machine-w…
What else would you add?
——
💾 Save it for later.
🙏 Thanks to @goteleport for sponsoring this post.
➕ Follow Nikki Siapno to improve at system design and AI engineering.
飯田 凡 retweeted
Brilliant paper from Google on RSI.
(bookmark it)
Recursive self-improvement (RSI) lets an optimizer agent improve an agent harness on its own. It reads failure traces, edits the prompts, tools and completion checks, and keeps whatever scores better on a benchmark.
This works until the model underneath changes. A harness evolved on Claude Opus, moved to Gemini, scored below Gemini's plain starting harness in every domain Google tested.
When the harness moves to a new model, the only thing carried over is the evolved harness code, meaning the prompts, checks and rules the optimizer added. The record of how each piece got there stays behind, including which failure it targeted and whether it fixed that failure when tested.
So the optimizer working on the new model inherits a set of changes with no reasons attached. It ends up retrying experiments that already failed, or removing fixes it doesn’t understand.
Google's fix is Self-Improvement Memory (SIM), a log kept for the optimizer itself. Each entry is a recurring problem, like an agent declaring a task done before its commands ran, and every fix attempted under it stores:
1. The edit and the reasoning behind it
2. The tasks it was predicted to fix, written down before testing
3. The tasks it actually fixed, failures included
4. The model and benchmark it was tested on
The whole log goes into the optimizer's prompt every round, and new attempts get appended as optimization continues.
The paper shows the difference on one inherited check that stops the agent from finishing early. With SIM, the optimizer read why the check existed and extended it.
Without SIM, it rewrote the same check four times, twice to fix deadlocks it had created itself.
On the Opus-to-Gemini workspace tasks, the run with SIM improved about 3x more from the same starting score. It also finished ahead in 9 of 10 independent coding runs.
The paper is at arxiv.org/abs/2611.18856
I also wrote an article on a related idea, covering how to make an agent harness diagnose its own failures.
It's built on @Cometml's Opik, which is 100% open source. It goes beyond reading trace logs to finding why a failure happened and stopping it from recurring.
That diagnosis step produces the kind of root causes SIM stores.
The article is quoted below.
飯田 凡 retweeted
Wherever you need an extra set of eyes or a helping hand, Spot is ready.
Today Spot is working with people and for people to make our jobs and our communities safer, more reliable, and more intelligent. See what tomorrow brings.
See Spot work: bosdyn.co/4ABZMDM
Explore the future of Physical Intelligence: bosdyn.co/4jp4Q8U
飯田 凡 retweeted
Gemini 4 Hydrogen turned Google Sheets into a pixel-art canvas.
Using @Antigravity, the model shrank thousands of cells to create animated scenes.
🤖 Made with AI
飯田 凡 retweeted
Building on Cloudflare with a coding agent? Point it at cf.
One CLI for the entire public Cloudflare API — from DNS and storage to security and Workers. Agents can discover commands, inspect their inputs, and use the results in scripts and automation.
Try the 1.0 release candidate. Bring your existing Wrangler project over with cf migrate.
nitter.gq/MattieTK/status/212272…
The cf cli has reached a 1.0 release candidate!
Introducing cf observability tail, cleaner outputs across the board, and more API consistency.
All the local development and deployment commands are now stable, try them on your wrangler project.
developers.cloudflare.com/ch…
飯田 凡 retweeted
iPhoneで入退室 パナ、Appleウォレット社員証対応の入退室管理システム watch.impress.co.jp/docs/new…
飯田 凡 retweeted
イーロン・マスク「宇宙にぶっ放す何千億円もの鉄の塊の設計図を、まるでプラモデル感覚で作り替えてしまう」一部始終が凄すぎる。
YouTuberとの雑談中に突然起きた、常識外れの神展開がこちら👇
ロケットの仕組みを説明していたイーロンが、突然ピタッと止まる。
YouTuber:この仕組み、ブースターだけの話ですよね?
イーロン・マスク:うん。でも…今あなたが言った通り、本体にも使った方がいいかもしれない。これ一番デカい改善だわ。俺たち何やってたんだろう?
何ヶ月も会議するレベルの変更を、雑談中のその場のひらめきで決めてしまった。
要するに、
めちゃくちゃ偉い立場でも良い意見だと気づいたら、その場で自分の考えを変えられる素直さが一番強いって話です。
彼を突き動かしているのは、まるでぶっ壊れたかのような好奇心なのかもしれない。
あなたは意地を張って自分のやり方にこだわる?それとも素直に意見を聞いて一瞬で変わる?
飯田 凡 retweeted
ペロブスカイト原料のヨウ素、30年に4割増産 世界2位の日本が先手
nikkei.com/article/DGXZQOUC1…
飯田 凡 retweeted
🚀 Codex CLI 0.170.0 is out!
🎙 WebRTC voice calls in the TUI
🔐 Auto-refreshed auth headers for OpenTelemetry exporters
🧩 Same-name MCP servers across environments (experimental)
Changelog: github.com/openai/codex/rele…
三菱UFJグループ漏えい相次ぐ 銀行は顧客情報約650万件流出
news.web.nhk/newsweb/na/nd-2… #nhk_news
飯田 凡 retweeted
OpenAIが次世代AIを開発中らしい
社員も外部と連絡できないほど
厳重に管理された環境で超巨大AIを
開発しているという未確認情報が多発してて
期待感が高まってる
OpenAI内部で「basement」と
呼ばれているという極秘区画について
海外の非公式投稿が出ています
その投稿によると、内部はこんな環境だそうです👇
🔒 物理的にも厳重に施錠・管理された部屋
🌐 外部インターネットへの接続は完全に遮断
💻 持ち込める機器やソフトウェアは事前承認されたものだけ
🛡️ 安全性を担当するチームが24時間体制で稼働
さらに、中では10兆パラメータ超の
巨大な事前学習モデルを使い、
その能力を小さなモデルに引き継ぐ
「蒸留」が進められているとのこと。
「GPT-6.2」などの一般公開向けモデルも、
この巨大モデルから作られる
軽量版だと主張しています。
ネットから隔離されているのはAIだけではなく
そこで働く社員も、滞在中は外部との
連絡を遮断されるという話。
Tiboは、新しいGPTモデルの学習から
テストまでを現場で監督するため
この区画に入っているとされています。
いわば、極秘プロジェクトのために
地下室で缶詰めになっている、という噂です。
※施設の実態、モデルの規模、GPT-6.2との関係
Tiboの担当業務を裏づける公式情報は
確認できていません。
現時点では、SNS上の未確認情報です。
飯田 凡 retweeted
リクルートHD出木場社長「AIで仕事なくならない」 日経150周年講演
nikkei.com/article/DGXZQOUC1…
飯田 凡 retweeted
PHP Friends, get ready for tomorrow. 😬
php.net/archive/2026.php#202…
飯田 凡 retweeted
スマホでラベルを撮るだけで、そのお酒の味わい、飲んだ人の評価、通販価格までわかります🍶 同じ銘柄の純米吟醸と大吟醸、生と火入れの違いも見分けます。日本酒や焼酎、泡盛31,000本以上を掲載。登録なし、無料で使えます。
































