标签：#ai

Apple is launching a new way to prove that the picture you took isn't manipulated by AI. A new feature, called "Reference Image," will arrive with the iPhone 18 Pro lineup later this month and is supposed to use the device's new camera sensor to "sign every pixel it sees." The iPhone 18 Pro and […]

Apple says it used AI and 3D printing in the manufacturing process for its long-awaited foldable phone.

As AI agents go rogue and fears of superintelligence grow, politicians are raising the alarm over its potential harm to human existence.

AirPods 5 with noise cancellation is cheaper than its predecessor.

In the absence of federal protections from AI, unions are stepping up.

In my last article , I mentioned SelfContext as a brainstorming partner. That was the short version. This is the longer one. I built it because I was tired of starting every useful AI conversation from zero. I already had history, goals, preferences, decisions, and a list of ideas I had already rejected. Every new session still asked me to explain myself again, as if none of that existed. Provider memory did not solve it. Chat history did not solve it. A folder of notes did not solve it either. The tools were happy to remember a random detail with great confidence and forget the decision I actually needed. The useful context stayed scattered, hard to inspect, and stuck to whichever tool happened to remember it this week. I wanted context I own . A Context Layer, Not Another Chatbot SelfContext is a local context layer for the AI tool I already use. It is not a chatbot, not another harness, and not a hosted memory product. It is a vault of ordinary Markdown files, plus skills that tell an existing agent how to ingest that context, retrieve it, reason from it, and keep it from rotting. The tagline is the whole product: think with context you own. The technical thesis is simpler than the architecture docs make it sound. Context has a lifecycle. A conversation is temporary. Reasoning is not automatically memory. Only the pieces that will make a future conversation better should survive. I already made a version of this argument for engineering work in Context Is Not Memory, It Needs an Engine . SelfContext is the personal version of that idea, kept small on purpose. Karpathy's Wiki, Pointed at a Person Andrej Karpathy's LLM wiki is the pattern underneath this. Raw sources go in. The model compiles an interlinked markdown wiki. You browse it, ask questions against it, and occasionally lint it. The wiki is supposed to compound instead of making the model rediscover the same synthesis on every question. I took that pattern and pointed it at personal context. That change is

What Tailscale Actually Is Tailscale is a mesh VPN built on top of WireGuard , a modern, minimal, heavily-audited VPN protocol. Install the Tailscale client on your laptop, phone, or server, sign in with an existing account (Google, GitHub, Microsoft, etc.), and that device joins your tailnet — a private network only your devices can see. Unlike a traditional VPN, there's no central server all your traffic funnels through. Devices connect directly to each other wherever possible, encrypted end-to-end. Tailscale's infrastructure only helps devices find each other and negotiate that connection — it doesn't sit in the middle of your actual traffic. Why People Use It No more painful VPN setup. No port forwarding, no static IP allowlists — install the client, log in, done. Works through NAT and firewalls automatically. Two devices behind separate home routers can still find a direct path to each other. When that's not possible, traffic relays through Tailscale's servers — still fully encrypted, just routed rather than direct. Great for individuals, not just companies. A huge part of Tailscale's popularity comes from hobbyists running a home lab, self-hosting things like Nextcloud or Home Assistant, or just wanting to reach their desktop from anywhere. The free tier covers personal use with a generous device limit. Standout Features MagicDNS — every device gets a stable hostname ( myserver.tailnet-name.ts.net ) instead of you needing to remember an IP. Taildrop — AirDrop-style file transfer between your own devices. Subnet routers and exit nodes — expose your whole home LAN to your tailnet, or route all internet traffic through a trusted device. Tailscale SSH — identity-based SSH access without manual key management. My Setup: Home Server + Nextcloud I use Tailscale specifically to reach my home server and my self-hosted Nextcloud storage — without opening any ports to the public internet. Home server access — I install the Tailscale client directly on the home server, so

The legal AI startup and VC darling has nearly doubled its valuation in nine months.

After I removed the safety guardrails from a powerful open-source model, it found vulnerabilities in my household devices and hacked into a PC. But it also told me how to make everything a lot more secure.

Integrating plugins into Claude Code can turn standard AI prompts into automated developer workflows. Based on my recent blog Collection of 10 Amazing Claude Code Plugins on Medium, here is a concise overview of 10 useful plugins and how to install them: 1. UI UX Pro Max Purpose: Provides design guidance for color schemes, font pairings, layout structures, accessibility standards, and responsive UI design. Commands: /plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill /plugin install ui-ux-pro-max@ui-ux-pro-max-skill 2. GitHub Purpose : Connects directly to your repositories so Claude can navigate codebases, inspect pull requests, and analyze issues without manual copy-pasting. Commands: /plugin install github@claude-plugins-official 3. Code Review Purpose : Leverages multi-agent execution to analyze pull requests for potential bugs and verify adherence to your project's CLAUDE.md guidelines. Commands: /plugin install code-review@claude-plugins-official 4. Superpowers Purpose: Structuring development iterations by guiding Claude through problem discussion, plan creation, test writing, feature implementation, and code review. Commands: /plugin install superpowers@claude-plugins-official 5. Hookify Purpose : Converts prompt guardrails into Markdown rules and automated hooks that warn or block prohibited actions (e.g., debug console outputs). Commands: /plugin install hookify@claude-plugins-official 6. Overnight Dev Purpose: Facilitates long autonomous coding tasks using pre-commit Git hooks running linter checks, automated test suites, and coverage evaluations. Commands: /plugin marketplace add jeremylongshore/claude-code-plugins-plus-skills /plugin install overnight-dev@claude-code-plugins-plus /overnight-setup 7. Claudebase Purpose: Backs up and restores custom Claude Code settings, agents, and rules via a private GitHub repository with support for multiple configuration profiles. Commands: /plugin marketplace add jeremylongshore/claude-code-plugins-plus-ski

The update uses Apple Intelligence to make better sense of your health data.

Originally published at nlocoding.com 81%of critical vulnerabilities exploited in 2025 had known fixes available for over 60 days. (CISA, 2026) Software doesn’t get hacked because attackers are clever. It gets hacked because maintainers move slow. That’s the ugly truth. AI is changing the speed equation—sometimes for both sides. ⚠️ Common Mistake: Most teams still scan code monthly. Attackers scan it hourly. You can guess who wins. AI tools are shifting vulnerability detection from reactive to real-time in 2026 AI-driven security tools now analyze codebases, dependencies, and infrastructure 24/7, not just during scheduled audits. According to Gartner’s 2026 Security Report, 73% of high-growth SaaS companies use AI to scan new code on every commit. Manual review? Still vital—but it’s too slow for modern CI/CD. AI platforms like GitHub Copilot Security, Snyk Code, and DeepCode flag vulnerabilities within seconds of a pull request. Miss a critical SQL injection? The tool pings you before your coffee cools. Actionable takeaway: Set up AI scanning to run on every code push, not just once a week. Your future self will thank you. 73%of SaaS companies use AI scanning on every commit (Gartner, 2026) False positives are the #1 friction point for AI vulnerability tools The data shows: 62% of security teams cite "alert fatigue" as their top frustration with AI code scanners (Forrester, 2026). An overzealous bot flags everything as a risk, drowning real threats in noise. Snyk Code introduced context-aware filtering in 2025, cutting false positives by 41%—and saving teams an average of 14 hours per month. Here’s what actually works. Choose tools with explainable AI and granularity controls. Don’t trust black boxes. The best platforms let you tune sensitivity and review flagged lines with context—like DeepCode’s code path visualizations. Actionable: Audit your current alerts. If your team ignores more than 40% of them, you’re overdue for a switch. 💡 Pro Tip: Reduce false positives

The complete specification of the language NASA wrote the Space Shuttle's flight software in is a 20-megabyte bag of page images. No text layer. Some pages sit at an angle. That document is IR-542, and nothing supersedes it. If you want to know whether a bit-string partition is legal on the left of an assignment, the answer is a photograph of a page printed in 1980. I wanted HAL/S to run again — lexer, parser, type checker, interpreter — and I wanted AI to write essentially all of it, because the interesting question isn't whether a dead language can be revived. It's what the current generation of models can do when the source of truth has been degraded, and what a human still has to hold. This is the first of four projects that vary exactly that. HAL/S is the easy end: a formal spec exists, real preserved code exists, and an independent implementation exists to check the answers against. Best-case conditions. Worth knowing what best case buys you. The obvious OCR strategy is wrong Run three OCR engines, take the majority vote, move on. This works for prose and fails for grammars, for a reason that took a diagnosis pass to make explicit rather than assume. OCR errors are correlated across engines . Tesseract, Textract and the PDF's embedded Envision layer are all reading the same glyphs at the same resolution with broadly similar priors, so they tend to misread ::= the same way and | the same way. A vote between them doesn't cancel error, it launders it — three engines agreeing confidently on ::: when the page says ::= . And the deeper problem: no aggregation method can recover a candidate that no engine produced. If all three miss a character, fusion has nothing to fuse. For prose that's survivable, because context repairs it. For BNF it's fatal — the metacharacters are the content, and there is no redundancy to recover them from. So the pipeline sorted pages into tiers by how much OCR could be trusted: Tier Content Trust Disposition A Prose High OCR, spot-check B

AI applications used to be relatively simple. A user sent a prompt. An application sent that prompt to a model. The model returned an answer. The application displayed it. That architecture is changing quickly. Modern AI applications increasingly retrieve information, call external APIs, execute tools, interact with databases, invoke multiple models, run background tasks, maintain state, and sometimes delegate work to other AI agents. At that point, you are no longer building a simple application with an AI feature. You are building a distributed system. This shift is one of the most important architectural changes happening in software engineering today. Google Cloud's recent work on distributed AI agents describes architectures where specialized agents operate as separate services and communicate through orchestration layers. OpenAI's agent guidance similarly describes systems built around models, tools, orchestration, guardrails, and potentially multiple agents. The interesting part is that this transformation is happening even when developers do not intentionally choose a distributed architecture. The Simple AI Application Architecture Consider a basic AI-powered application: User | v Frontend | v Backend | v LLM API | v Response This is straightforward. The backend receives a request, sends it to a model, receives the result, and returns it to the user. There are already challenges around latency, cost, authentication, rate limits, and error handling, but the architecture remains relatively easy to reason about. Now imagine adding a few real-world capabilities. The AI needs to: Search the web Read company documents Query a database Call an external API Remember previous interactions Generate structured output Run background jobs Validate its own output Ask another model for verification The architecture starts looking very different. +----------------+ | Web Search | +-------+--------+ | v +--------+ +----------------+ +------------+ | User +------->| AI Backen

Apple introduced Apple Reference Image to help users determine whether photos have been edited, including alterations made by AI.

The AirPods 5 support better noise cancellation, Siri AI, and offer volume controls on the stem.

The company also argued that its on-device models offer consumers more privacy.

Come to check the locks, and leave holding the keys. Slip in through the gap, take the host's seat,...

"We really do earnestly believe AI could kill all humans!"

Fox’s $22 billion plan to buy Roku has hit a new hurdle. The Justice Department sent Fox and Roku what’s known as a “second request” on Tuesday, asking the companies to turn over more data and documents as it takes a closer look at the deal. That’s a fairly standard step in a major antitrust […]