AI News Today

DeepSeek's DSpark framework boosts AI response speeds by up to 85% using speculative decoding.

Batam is just 20 kilometers from Singapore and offers tax incentives in its Special Economic Zone.

Kunlunxin's P800 chip is gaining traction in data centers run by Chinese state-owned firms.

Article URL: https://decrypt.co/371877/ai-agent-nuclear-strike-civilization-vi-benchmark Comments URL: https://news.ycombinator.com/item?id=48712791 Points: 2 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>test: restore embedded agent fixture types</pre>

<p>I tracked every AI prompt I used for 30 days straight. Most were throwaway. But <strong>seven of them came back over and over</strong> — and together they save me at least 10 hours a week.</p> <p>These aren't "act as a senior developer" fluff. They're specific, structured prompts with slots you fill in. Copy them, paste them, edit the <code>[bracketed]</code> parts.</p> <p>Let's go.</p> <h2> 1. The "Explain This Code" Prompt (saves ~2 hrs/week) </h2> <p>We all inherit code we didn't write. In

Article URL: https://erikjohannes.no/posts/20260408-on-the-usefulness-of-ai-agents/index.html Comments URL: https://news.ycombinator.com/item?id=48712672 Points: 2 # Comments: 0

Article URL: https://www.cnn.com/2026/06/26/asia/ai-glasses-cheating-exams-intl-hnk Comments URL: https://news.ycombinator.com/item?id=48712649 Points: 4 # Comments: 0

Article URL: https://unmeshed.io/blog/bringing-ai-workflow-into-production-without-burning-tokens Comments URL: https://news.ycombinator.com/item?id=48712618 Points: 1 # Comments: 0

<p>Visa applications are stressful. You have dozens of documents to collect, country-specific requirements to meet, and strict deadlines to hit — all with zero visibility into where you actually stand. I built VisaTrack to fix that: an AI-powered document tracker that generates your required checklist automatically, lets you track progress, upload files, and get guidance from an AI advisor.</p> <p>I built it for the H0 — Hack the Zero Stack with Vercel v0 & AWS Databases hackathon. Here's how it

<h2> Beyond Regex: Building Detection Rules for AI Agent Vulnerabilities </h2> <p>When I started building <a href="https://github.com/dockfixlabs/agentguard" rel="noopener noreferrer">AgentGuard</a>, the first question was: how do you detect a prompt injection vulnerability in source code?</p> <p>Unlike traditional vulnerabilities (SQL injection, XSS), prompt injection doesn't have a single signature. It's a pattern of <strong>untrusted data flowing into LLM context</strong>. The vulnerability i

<h2> The Problem </h2> <p>AI agents are being deployed at scale — in customer support, code generation, data analysis, and autonomous workflows. But the code that powers these agents is rarely security-audited.</p> <p>Consider this pattern, common in production agent codebases:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight python"><code><span class="n">user_input</span> <span class="o">=</span> <span class="n">request</span><span class="p">.</span><span class="nf">json

Comments

<p>Using an agent feels like a dream. The limit is what wakes you up.</p> <p>Everyone's talking about loops, goals, agents, subagents, yada, yada, yada. But barely anyone mentions what to do when the hand of time slaps you in the face and the dream you were dreaming vanishes like an island cloud.</p> <h2> Then You Hit the Wall </h2> <p>It feels really nice to have that kind of power on your side — to command multiple entities working for you. One agent is fun. Five agents, all running at once? T

<pre style='white-space:pre-wrap;width:81ex'>Fix getSubagentDepth for legacy subagent session keys (AI-assisted) (#54593) The previous implementation split on ':subagent:' which requires colons on both sides. Legacy session keys like 'subagent:worker' (starting without a colon) were missed and returned depth 0. Fix: normalize key, match (^|:)subagent: pattern to correctly count nested subagent levels for both legacy and current key formats. - Add test cases for legacy key formats Co-authored-by:

We built Selixes, a self-hosted LLM failover gateway with atomic budget caps and PII redaction. Repo: https://github.com/selixes/gateway Comments URL: https://news.ycombinator.com/item?id=48712130 Points: 1 # Comments: 0

<blockquote cite="https://blog.jonudell.net/2026/06/28/doctor-it-hurts-when-agents-create-unreviewable-prs-dont-do-that/"><p><strong><del>Human</del> Agent in the loop</strong></p> <p>I dislike the phrase “human in the loop” because it cedes authority to the machines. Let’s flip the narrative. It’s our loop, we work the same way we always have, now we recruit agents to join the team. An agent-assisted process need not be a black box that takes in prompts and emits features. [...]</p> <p>Let’s do

Article URL: https://getdesign.md/ Comments URL: https://news.ycombinator.com/item?id=48712013 Points: 1 # Comments: 0

China's Zhipu AI (Z.ai) released its open-weight GLM-5.2, and some researchers have claimed that it matches Mythos in certain bug-finding and cybersecurity scenarios. While GLM lags behind models from Anthropic and OpenAI in other, more general tasks, it seems that China has dramatically reduced the gap in the capabilities between its models and those of […]

I built a free, open-source (MIT) macOS menu-bar app that runs Whisper models locally to assist with dictation. There is also the option to use an LLM (BYOK). You hold a global hotkey, speak, and the text gets pasted at your cursor similar to other similar/popular apps. Some details: - Transcription runs locally via WhisperKit (CoreML), so it works offline once the model is downloaded. BYO model — tiny through large-v3, auto-downloaded and cached. The base model has been working surprisingly wel

Article URL: https://github.com/rileyq7/drift Comments URL: https://news.ycombinator.com/item?id=48711629 Points: 2 # Comments: 0

I'm a developer. I have a basic working knowledge of classic neural networks. I'd like to learn about genetic algorithms. What books should I be reading? I prefer dead-tree, but pdf suggestions too please. Comments URL: https://news.ycombinator.com/item?id=48711408 Points: 11 # Comments: 2

Suno has ambitions to be more than just a toy to churn out AI slop, it also wants to be a streaming destination and to break new artists. Spark is their new incubator program for independent artists that provides grants, mentorship, and marketing support. To apply, artists need to be an unsigned singer, songwriter, or […]

Article URL: https://chriskiehl.com/article/evolving-thoughts-on-ai-2026 Comments URL: https://news.ycombinator.com/item?id=48711187 Points: 2 # Comments: 1

Lately I've been planning on running my first 100 miler and been struggling with organizing everything around the race. Been back and forth between google docs, notes, spreadsheets and AI agents. So in the middle of that, and a long run, though about creating a kind of journaling app for taking notes about the race and everything about it and it got interesting. So here is the first kind of working version. I would love some brutally honest feedback: - What do you think? - What is the clunkiest

Article URL: https://github.com/Abloatai/ablo Comments URL: https://news.ycombinator.com/item?id=48711106 Points: 2 # Comments: 0

Can we revitalize mitochondria? ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Article URL: https://techcrunch.com/2026/06/28/ford-rehires-gray-beard-engineers-after-ai-falls-short/ Comments URL: https://news.ycombinator.com/item?id=48710749 Points: 2 # Comments: 0

Article URL: https://github.com/gi-dellav/use-zerostack/ Comments URL: https://news.ycombinator.com/item?id=48710717 Points: 3 # Comments: 0

There was a recent thread about a tool that "proves" a human actually wrote something via an animated edit history ( https://news.ycombinator.com/item?id=48699969 ). I commented that I can recognize AI writing regardless of such tools, and received some pushback...so, I wanted to be sure I wasn't lying to myself, and made a little game that shows samples or writing in a variety of styles, and you pick whether it's human or AI. It's harder than I expected, especially with current frontier models,

Article URL: https://netbird.ai/ Comments URL: https://news.ycombinator.com/item?id=48710420 Points: 1 # Comments: 0

Article URL: https://github.com/qataruts/monlite Comments URL: https://news.ycombinator.com/item?id=48710233 Points: 1 # Comments: 0

Article URL: https://hedonicescalator.substack.com/p/the-llm-shoggoth-meme-is-weirder Comments URL: https://news.ycombinator.com/item?id=48709988 Points: 3 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>fix(gateway): scope tools.effective global agent lookup Pass the requested agent into tools.effective global-session loading while preserving non-global session ownership guards, with handler and gateway regression coverage.</pre>

Article URL: https://malerchen.com/high-schoolers-and-ai/ Comments URL: https://news.ycombinator.com/item?id=48709893 Points: 1 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): truncate console text on code-point boundaries (#96296) sanitizeForConsole filtered control characters code-point-aware but then truncated with sanitized.slice(0, maxChars), which cuts on UTF-16 code units. When the cap landed between the two code units of an astral character (emoji, CJK extension, etc.) the output ended in a lone high surrogate before the ellipsis. Cap on code points instead. Co-authored-by: ly-wang19 <ly-wang19@users.no

<pre style='white-space:pre-wrap;width:81ex'>fix(matrix): truncate reply context on code-point boundaries (#97471) truncateReplyBody used a raw value.slice(0, MAX_REPLY_BODY_LENGTH - 3) to shorten long reply bodies. When the cut index fell between the two UTF-16 code units of a surrogate pair (e.g. an emoji), the slice left a lone high surrogate before the ellipsis, which renders as a broken glyph in the reply context shown to the agent. Replace the raw slice with sliceUtf16Safe from the plugin

<pre style='white-space:pre-wrap;width:81ex'>[codex] Update Claude CLI billing docs (#96848) * Update Claude CLI billing docs * Sync Claude live testing billing note</pre>

<p>Apple's newest on-device model carries about 20 billion parameters, and on any given request it fires maybe one to four billion of them. That gap — 20B stored, roughly 3B running — is the whole story of 2026. The model that now ships inside the latest iPhone is no longer a shrunken, lobotomized cousin of the cloud model. It's a different kind of object: large in flash, small in motion, and it never phones home.</p> <p>For three years the on-device pitch was mostly aspirational. Demos ran, lat

<blockquote> <p><strong>TL;DR —</strong> Coding agents stopped being a checkbox in your IDE and turned into a four-way platform war in the first half of 2026. Anthropic is winning the model-and-product fight, OpenAI is winning distribution, and Cognition is winning the enterprise. The real moats are model cadence, install base, and price — not features. Pick your agent like you'd pick a vendor you might have to leave, because one of these companies is going to whipsaw your workflow before the ye