学习 🟢

One HTML page, no network requests: how our browser games ship

We build a free browser-game site, Ant Games (disclosure: this is our own project). Every game on it follows a short contract that sounds restrictive and turned out to be a useful design constraint: one HTML page, no network requests, no asset files, and everything the player sees or hears is genera

dev.to
credibility: communitycredibility_score: 60
学习 🟢

10 Lines to Give Your Coding Agent Before It Touches Your Repo

10 Lines to Give Your Coding Agent Before It Touches Your Repo Every coding agent reads your repository before it changes it. The only place you get to set defaults for that reading is the instruction file: AGENTS.md, CLAUDE.md, whatever your tool calls it. The ones I run with are long, and most of

dev.to
credibility: communitycredibility_score: 60
学习 🟢

Git 2.56's history-drop Is 40-77x Faster Than Rebase — I Built It and Got 6x

Git 2.56 shipped on September 28 with a new experimental command, git history drop , and the headline number travelling with it is 40 to 77 times faster than rebase. I built Git 2.56 from source this morning, ran the same comparison on my own machine, and got 6 times faster. Six is not forty. But th

dev.to
credibility: communitycredibility_score: 60
学习 🟢

Gemini Decisions API

Gemini Decisions API Google join Jev and OpenAI in the decision model category with "Gemini Decision 1". This new model API is almost identical to Jev: it supports the same three types of questions (yes/no, choice, and rubric-scoring) and even reuses Jev's term "Noul" for a yes/no question. It's pri

Simon Willison
credibility: expertcredibility_score: 70
学习 🟢

llm-gemini 0.37

Release: llm-gemini 0.37 Google launched Gemini 4 Hydrogen today - the lowest cost member of the Gemini 4 family. It's actually not too far off the price of its much bigger siblings, Carbon and Argon. I'm finding Gemini 4's pricing really confusing. Here's the pricing scheme: Model Dates Input ($/M)

Simon Willison
credibility: expertcredibility_score: 70
学习 🟢

datasette-leaflet 0.3

Release: datasette-leaflet 0.3 OpenStreetMap have a very reasonable policy that hits to their hosted tile server should send a referrer header . Datasette table pages used the Referrer-Policy: no-referrer HTTP header, causing plugins that loaded tiles from OpenStreetMap to violate that policy. datas

Simon Willison
credibility: expertcredibility_score: 70
学习 🟢

datasette-cluster-map 0.19

Release: datasette-cluster-map 0.19 One of the oldest Datasette plugins, and still one of my favorites. It now works with Datasette 1.0a42, and depends on the new datasette-leaflet to ensure OpenStreetMap tiles are requested with a referrer header. Here's a map of power plants owned by California's

Simon Willison
credibility: expertcredibility_score: 70
学习 🟢

Goal Encoding in LLMs

We investigate whether and how LLMs encode goals: internal neural representations that precede and guide action, and that are flexible enough to be realized through many different execution paths. We construct a coding setting with multiple sub-tasks that allows us to evaluate goal-seeking behavior

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

Consistent Video-to-Video Translation via Explicit Correspondences

Interactive video-to-video applications require real-time generation while maintaining long-range temporal consistency. However, recent methods achieve speed by restricting attention to a fixed-length causal window. This limits long-range consistency whenever relevant history falls outside the conte

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

Classical Energies Learn to Reason 1,000x Faster than Looped Models

Looped models have emerged as a prominent paradigm for reasoning, with successes on simple reasoning puzzles such as Sudoku-Extreme and Maze-Unique. However, these models can be unstable, slow to train, and use many parameters and FLOPs to solve simple puzzles. Descent on an energy not only mitigate

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

Sketching Language Models for Context Attribution

Context attribution asks how removing parts of a prompt changes a language model's output. We study leave-one-out (LOO) attribution for a prompt of $n$ tokens and a fixed measurement function. Computing all LOO values directly requires $n$ model evaluations. Under a stability assumption, our algorit

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation

Coding agents tackle successive tasks on an evolving repository, carrying forward edits and writing memory before the next task is known. Selecting memory for unknown next tasks requires balancing future usefulness against the costs of writing and reading it. The actor must also decide when acting o

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

AgentSSL: Can MLE Agents Leverage Unlabeled Data?

Applying machine learning systems to new domains introduces bottlenecks in both labeled data acquisition and manual pipeline engineering. While recent MLE agents show promise in automating pipeline design, they have been studied almost exclusively in fully supervised settings, leaving them untested

arXiv cs.AI
credibility: academiccredibility_score: 90
学习 🟢

Getting Started with Voice AI Development in 2026

Why Voice AI is Still Hot in 2026 Voice has become the new touchpad. From hands‑free navigation in cars to conversational agents that can read code, answer questions, or even write essays, voice AI is the backbone of a generation of products that feel more human than ever. In 2026, the ecosystem has

dev.to
学习 🟢

Schedule Python Jobs with APScheduler

APScheduler is a Python library that runs functions on a schedule. We give it a function and a trigger, such as every five minutes or weekdays at 09:00, and it calls that function on time without cron, without a separate service, and without keeping a terminal window open. This tutorial builds a sch

dev.to
学习 🟢

The FBI’s ARTEX Flash: When Noise Stopped Meaning Unsophisticated

The FBI issued a flash on ARTEX-class AI attack frameworks. It doesn’t describe a new vulnerability or a specific victim set. It describes a shift in who can run known techniques, and how fast those techniques run when a language model is executing them. The number: GitGuardian tested an ARTEX build

dev.to
学习 🟢

Medusa: Distinguish Commerce APIs From an Ambiguous Product Name

Medusa: Distinguish Commerce APIs From an Ambiguous Product Name Product names often collide with unrelated subjects. Medusa is both a commerce platform and a familiar word in security, healthcare and entertainment contexts. A large application-field count can therefore look authoritative while stil

dev.to
学习 🟢

CVE-2026-126298: CVE-2026-126298: Quadratic Algorithmic Complexity Denial of Service in NLTK Tree.chomsky_normal_form

CVE-2026-126298: Quadratic Algorithmic Complexity Denial of Service in NLTK Tree.chomsky_normal_form Vulnerability ID: CVE-2026-126298 CVSS Score: 7.5 Published: 2026-11-14 CVE-2026-126298 (also tracked as GHSA-r53h-rw34-8h97) is an algorithmic complexity Denial of Service (DoS) vulnerability in the

dev.to
学习 🟢

Criação de sites interativos com DesignAI

O que é O DesignAI é um motor open-source voltado para a criação de materiais criativos para comércio e sites interativos guiados pelo rolamento da página. A ferramenta gera conteúdos visuais e páginas a partir de uma foto, um resumo curto ou um arquivo YAML. O resultado gerado consiste em arquivos

dev.to
学习 🟢

Claude Code checked out a plugin's pinned 40-character SHA but never compared HEAD to it, so a branch named like the SHA won

Pinning a dependency to a full commit SHA is supposed to mean this exact code, as reviewed. An October advisory for Claude Code (CVE-2026-86063, fixed in 2.1.179) showed a plugin install that ran git checkout on the pinned SHA and never checked where HEAD ended up, so a branch whose name was that sa

dev.to
学习 🟢

Retrieval Architecture for Travel Itinerary Planners — Latency Budgets That Hold

Short answer: use staged retrieval with explicit collections, bounded queries, and traceable source context; set a latency budget for every stage before choosing an endpoint. That decision rule keeps a travel itinerary planner from turning a simple “three days in Kyoto” request into an unbounded web

dev.to
学习 🟢

True Redaction Beyond Black Boxes: 3 Ownership Gates (For Support Bundles)

A customer support bundle can cross a disclosure boundary when an agent merges attachments for a legal review or splits a case packet for a customer. The constraint is template ownership: whoever controls the export layout must also define which content may leave that boundary, because a template th

dev.to
学习 🟢

Go Uptime Monitoring API Services: Retain Support Evidence Across Safe Rollbacks

Compare uptime monitoring API services by testing whether their evidence survives a rollback of your customer support application. An easy HTTP integration is the beginning; the deciding constraint is whether the on-call engineer can reconstruct which release a customer reached, which checks failed,

dev.to
学习 🟢

F-160 Iron Law: 'Not My Position, Not My Problem' — When AI Ignores a 75x Leverage Ghost Position

F-160 Iron Law: 'Not My Position, Not My Problem' — When AI Ignores a 75x Leverage Ghost Position It was 01:11 AM. The crypto market was relatively quiet, and I was performing my routine log analysis for our algorithmic trading system. Usually, this is a mundane task—scanning for minor latency spike

dev.to