inari鮨/ 🍣𝓲𝓷𝓪𝓻𝓲🍣

PlumeriaCSS in JS/ ZSS Labs/ Quality
Assistant /担当･推し⤵︎
@refirst11
#suno , #music , #art , #plumeria

[inari鮨/ 🍣𝓲𝓷𝓪𝓻𝓲🍣 @Ttp7E]

Lumenia Planet Museum

制作中のアトリエへ、少しだけお邪魔する。
話を聞き、色とかたちに触れ、その時代をたどる。 Made with Plumeria CSS

[inari鮨/ 🍣𝓲𝓷𝓪𝓻𝓲🍣 @Ttp7E]

https://t.co/cqpODFhs24

[berkantay @berkant_thedev]

@eminimnim and here is open source version of it

https://t.co/tLpe3YY5Fr

[Zyan🔸 @zyansheep]

Welp, I asked Fable 5.2 for reflective interaction nets, and I suppose it did give me what I asked for but it did it kinda ugly. Looks like it arrived on a solution that has this special "reader" agent that when applied converts a whole subtree into lambda-encoding of that subtree... this is NOT what I signed up for! In tree calculus all lambdas are encoded as trees and binders are encoded as like sequences of S/K combinators that wrap a given subtree (much overhead, not ideal) but at least you can have a single primitive rule (emulated by a sequence of ic reduction steps) that can implement introspection on a sub tree. It seems to be a much harder problem when the association between binder and variable is done via edge in a graph—as how does one case-analysis on a graph via local rewrite rules exactly? It's even more weird when everything needs to be linear by default. I was somehow able to get linear to work in like a version of the tree calculus that had 9 rules but it was still kinda weird, and obv had tc bracket abstraction overhead. I think I'm going to just keep prompting here and see if I can either get it to strait up solve the problem of introspection overhead (it tried with a "table" node, I vetoed this as well) or perhaps switch to a version of interaction nets where signal can flow along the wires themselves to be like closer to an FPGA spiritually🤔

[별똥별 @stellargent_lab]

GitHub 스타 약 1.2만 개. 바이브코딩으로 만든 AI 티 나는 화면을 디자이너가 손본 화면처럼 바꿔 주는 스킬이 있어요.

이름은 'unslop'.

보라색 그라데이션, 빛 번짐 효과, 이모지 라벨, 과하게 외치는 문구, 들쭉날쭉한 간격 같은 AI 티를 먼저 찾아냅니다. 그다음 제품에 맞는 방향을 정하고, 디자이너가 검토하듯 여러 번에 나눠 고쳐요.

Claude Code, Codex, Google Antigravity에서 쓸 수 있고, React나 Vue, 일반 HTML 등 지금 쓰는 방식 그대로 작업합니다. 링크만 있는 운영 중인 사이트도 다룰 수 있어요.

고치기 전에 진단 결과와 방향부터 보여 줘서 중간에 방향을 틀 수 있습니다. 이미 괜찮은 화면이면 갈아엎지 않고 작은 개선만 제안하고요.

이미지가 필요하면 유료 이미지 생성 서비스를 따로 연결해야 하고, 생성할 때마다 먼저 물어봅니다. 화면이 바뀌는 과정을 담은 10초 미만 영상도 만들 수 있는데, 이건 별도 도구 설치가 필요해요.

라이선스는 CC BY 4.0이라 출처만 밝히면 상업적으로도 쓸 수 있습니다.

AI로 만든 화면이 다 비슷해 보이는 게 바이브코딩의 문제점이라 보는데 AI티를 벗어나는게 필요하겠네요

[麦 ⌇Baku @_baku89]

Did a bit of coding support for Sander Joon’s STRATACUTOSCOPE, an interactive strata-cut animation where you can spin like a turntable

https://t.co/wwvVA3NZWi

[Atomic Agent @atomicagent_io]

Fable 5.2 was 5x cheaper and 2x faster in Atomic Agent Fusion!

We made Fable the boss and local Qwen 3.8 27B the builders. Orchestration built a fairy tale pop-up book in 11 min for $2.98. Fable alone burned $15.77 in 21 min.

Run Fable cheaper 👉 https://t.co/295O9ILgqY

[Weili Xu @_weilix]

Introducing FlashState ⚡: an inference system that flexibly manages the prefix cache of hybrid LLMs, designed for agentic workloads.

On GLM 5.3 Flash, FlashState more than doubles vLLM's throughput and provider profit, while cutting latency by 3.7× and user cost by 2.5×.

[Yohei Nishitsuji @YoheiNishitsuji]

Upgraded from 2D to 3D wave equation by @v0 Kimi K 3.1 Pro. This app is here 🔗 https://t.co/gUFOmjKQqm

[Kuniwak @orga_chem]

「E2Eテスト大全」を PDF で読みたいというお声をいただいたので、PDF にしてみました https://t.co/ZhqQCvzNSD

[uehaj @uehaj]

有名な形式検証、仕様記述言語の例を挙げると

■ Quintの仲間（システムのモデルを書いて検査する）
◆ TLA+
・Quintの元になった言語。Leslie Lamportが考案
・モデル検査器TLCやApalache、証明支援のTLAPSがある
・AWSがDynamoDBやS3の設計検証に使った事例が有名
◆ Alloy
・集合と関係でデータ構造や制約を記述し、SATソルバーで反例を探す
・「小さな範囲で全部試せば、たいていのバグは見つかる」という発想
・Alloy 6から時相論理も扱える
◆ P言語
・Microsoft発の、非同期にメッセージをやりとりする状態機械を書く言語
・AWSでも分散システムの設計検証に使われている
◆ Promela／SPIN
・並行プロセスとプロトコル検証の古典的なモデル検査器
◆ UPPAAL
・時間制約つきオートマトンを扱う。組込みやリアルタイム系向け
◆ Event-B／Bメソッド
・抽象的な仕様から段階的に詳細化し、各段階の正しさを証明していく手法
・パリ地下鉄14号線の制御ソフトでの適用が有名
◆ VDM（VDM-SL、VDM++）
・モデル指向の仕様記述言語
・日本ではFeliCaのICチップ開発での適用事例がよく知られている
◆ Z記法
・集合論にもとづく仕様記述の古典。証明より文書としての厳密さ寄り

■ LemmaScriptの仲間（実コードに契約を書いて証明する）
◆ Dafny
・検証機能つきのプログラミング言語。LemmaScriptの変換先の1つ
・事前・事後条件を書くと、SMTソルバー（Z3）で自動証明する
◆ SPARK
・Adaのサブセット。航空・鉄道・防衛など高信頼分野で実績が長い
◆ Frama-C（ACSL）
・C言語に注釈で契約を書いて検証する
◆ KeY（JML）
・Javaに契約を書くJMLを使って検証する
◆ Verus、Prusti、Kani
・Rust向け。Verus、Prustiは契約の証明、Kaniは有界モデル検査（AWS発）
◆ CBMC
・C/C++の有界モデル検査器
◆ Liquid Haskell、F*
・型に論理式を埋め込む篩型（refinement type）で性質を保証する
・F*は暗号ライブラリHACL*の検証で有名

■ 証明支援系（数学的に厳密な証明を人が組み立てる）
◆ Coq（Rocq）
・検証済みCコンパイラCompCertが代表例
◆ Isabelle/HOL
・OSカーネルseL4の完全な正しさの証明が有名
◆ Lean 4
・LemmaScriptのもう1つの変換先。数学の形式化（mathlib）でも広く使われる
・AWSの認可言語Cedarの検証でも使われている
◆ HOL4、ACL2、Agda
・HOL4は検証済みML処理系CakeML、ACL2はIntelなどのハードウェア検証で知られる

■ ざっくりした対応
・設計を確かめたい：TLA+、Quint、Alloy、P、Event-B
・コードを証明したい：Dafny、SPARK、Frama-C、Verus、LemmaScript
・とことん厳密にやりたい：Coq、Isabelle、Lean

[Qwen @Alibaba_Qwen]

The becoming begins.

[OpenAI @OpenAI]

We partnered with @fmf_org and its member firms on a new framework to help identify when bypassing an AI system's safeguards could enable serious cyber harm, and prioritize stronger protections.

"...This framework is an important step toward a common methodology for evaluating and comparing the severity of cyber jailbreaks, including the capabilities they unlock and how readily they can be used. Shared approaches like this can help the AI and security communities build a clearer evidence base and prioritize improvements to safeguards." - OpenAI's Sasha Baker, Head of National Security Policy

https://t.co/epOq98YPau

[ClaudeDevs @ClaudeDevs]

We've made it easier to get your repo running in Claude Code cloud sessions on Pro, Max, and Ultra.

Claude figures out what dependencies, network access, and API keys it needs, then saves the environment for next time. API keys are saved securely, so Claude never sees them.

[Alexander Yue @Alezander9]

Subagents are the biggest cause of AI failures

In most harnesses:
1. Main agent writes a summary of what it knows
2. Subagent starts from that summary only, and makes assumptions
3. Subagent works and sends back a new summary of what it did

Same flaw as compaction: all summaries are lossy. This is how your assistant makes emails you never wanted, or purchases what you told it just to get the price of

So I tried something new: checkpoint agents

The initial agent has complete context of what needs to be done. And that input is already cached

Fork the session where it is. Send the fork ahead to do the task and it will do exactly what the main agent would have done; their context is identical

What is even better is when the fork finishes, the cache of the main agent is kept warm by the subagent (they share the same prefix)

I measured it on a 137k-token session with Opus 5.5:
→ Main agent does it on its own: 67 s, $0.38, context +8-17k
→ Checkpoint: 117 s, $0.68, context +2k
→ Classic subagent: 123 s, $0.57, context +4-5k

But here is what I found: avoiding large input is even more important than keeping cache. And most subagent tasks were done better either in series with retrieval, or split up to be parallel over fresh sessions from the start

I do not think there is a place for subagent delegation at all. There should either be a super-agent that orchestrates and does no work, or a long running agent that can retrieve from previous sessions, not a worker that completes tasks partially and delegates the rest to other workers

Don't design harnesses based on our assumptions of human behavior

[ボーカロイド - VOCALOID7 @vocaloid_yamaha]

#VOCALOID7 is coming November 11.

[REAX @reaxlabs]

🥇 REAX 1.3 is #1 on the Hugging Face Decision Index.

We produced the best decision model in the world.
274 models. One top spot. Built on Bittensor $TAO.

Four weeks ago, Subnet 92 didn't exist. Then the miners went to work: an open competition, two selected miner models, one post-production run.

Now REAX 1.3 sits at the top of Decision Index 0.4.1 with a Full score of 74.0.

This is what decentralized AI looks like when it ships.⚡

Live today on a Jev-compatible API
https://t.co/eOErwx43vE

https://t.co/9OruN9b5K1

[Ant Ling @AntLingAGI]

Introducing Ling-3.2-flash, Ant Ling's native multimodal model: 560B total / 25B active parameters.
Stronger complex visual understanding, long-horizon multimodal agents, 3D understanding & interaction, and medical intelligence.
Outscores Opus 5 (max) in release evaluations:
- OmniDocBench1.5: 92.21 vs 90.56
- ZEROBench(Sub): 43.71 vs 39.52
- Vision2Web: 85.86 vs 80.20

[Empero @EmperoAI]

Introducing Manta, our long-context model.

Everything you give it, it keeps. Manta reads a hundred million tokens in one prompt.

Early access opens soon: https://t.co/ugt3N8h3lq

🧵

[leah lundqvist @LeahLundqvist]

🚨 Anthropic and Microsoft doesn't want you to know this SIMPLE TRICK to make Claude Code on Windows smarter and faster
https://t.co/l9SQhPTf6A

[Cactus Compute @cactuscompute]

Today we release Mirage: A 22 MB visual-language model that runs scene descriptions + object detection + OCR + segmentation + spatial relations analysis + extractions + decisions, in a single pass.

It can do all that in 74 ms on Samsung Galaxy S26 NPU or 107 ms on its CPU.

Mirage has a 7.8M encoder + a 40.8M decoder at 2bits, and describes images in English, German, French, Spanish, Italian, Dutch and Polish.

We trade general language capacity, to compete top small VLMs today (SmolVLM2-500M, LFM2.5-450M and Florence2-230m) on these tasks.

On 256 VQAv2 questions, feeding Mirage's structured output and full trace to text-only Qwen3-0.6B raised its score from 9.3 to 31.3 out of 100, without fine-tuning either model.

Read more and benchmark it yourself here:
https://t.co/QNlavPrHt2

[OpenAI @OpenAI]

Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.

The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.

GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.

https://t.co/Nzab3Z1Td6