标签：#AI

With this acquisition, DeepL is opening an office in San Francisco to expand its U.S. business.

Let's be real: There's no shortage of Zelda-style games to play right now. That's true even if you've exhausted all there is to see in Tears of the Kingdom or Echoes of Wisdom. Indie developers keep finding new ways to refresh the formula, whether it's fusing it with a Soulslike, making it cozier, or simply […]

On 14 April, the Trump administration quietly acknowledged the widespread use of AI to automate government processes. The office of management and budget (OMB) disclosed a staggering 3,611 active or planned use cases for AI across the federal government. The list has ballooned by 70% from the one published in the final year of the Biden administration, and includes many disturbing-seeming plans to hand over sensitive governmental functions to AI. Scanning this list, many readers may find many causes for alarm. It represents a transfer of decision processes from human to machine on a massive scale over matters of individual freedom, public health and well-being, nuclear reactor safety and more...

Aditya Kumarakrishnan explains how to move past the "amnesia phase" of AI. He shares a blueprint for engineering leaders to build modular agent frameworks using CoALA, leverage decades of process science for scalable workflows, and "terraform" legacy environments into robust, event-sourced artifacts capable of handling unpredictable, cross-functional agent demands. By Aditya Kumarakrishnan

Pinterest has launched 'Ask Pinterest,' an experimental AI-powered shopping app that lets users seek recommendations and inspiration through a conversational interface.

Hello, I'm Maneshwar. I'm building git-lrc, a Micro AI code reviewer that runs on every commit. It is...

Ask a language model to score a developer portfolio out of 100 and you get a confident number back. Hand it a near-empty page with a name and a broken avatar, and it will often still tell you something like 92. Nice layout. Strong personal branding. The model is being polite, not accurate. That was the first wall I hit building Leon, the reviewer inside getfolio. If the score is not trustworthy, nothing downstream matters: the critique, the suggestions, and the fix button all hang off a number the model invented to sound encouraging. This is the build log of how I stopped letting the model hold the pen. Short version: a deterministic rules engine owns the score, and the language model only owns the words around it. The failure mode: an LLM judge wants to be liked If you have shipped anything with an LLM evaluator you have probably seen this. You hand it a rubric, a JSON schema, even worked examples, and it still drifts upward. Empty inputs get encouraging scores. Weak inputs get the benefit of the doubt. Strong inputs land in the same band as the weak ones, just with longer praise. A few reasons, roughly in order of how much they hurt: Tuning rewards a helpful, encouraging tone. Harsh scoring reads as unhelpful, so the model softens it. The model has no ground truth for what a 70 versus an 85 looks like in your specific domain. It is scoring on vibes. Scoring and explaining are entangled. The model writes the kind explanation first, then picks a number to match the nice things it just said. Run it twice on the same input and you get two different numbers. There is no anchor. For a portfolio reviewer that real recruiters and developers would act on, that was a non-starter. If Leon says 64, an empty page should not be able to reach 64 by accident, and a strong portfolio should not get talked down to it either. The number has to mean something. The fix: rules engine owns the score, model owns the language The architecture splits responsibilities hard. A deterministic e

Originally published on the Sieve blog . Sieve is an open-source (Apache 2.0) context-reduction proxy — I work on it, and I've tried to keep this post about the problem rather than the tool. If you run an LLM agent for real work, the bill is the part nobody warned you about. It starts small, it grows with use, and the worst of it is invisible — most of what you pay for on any given turn is text the model has already seen, or text you never meant to send. There's a temptation to treat this as one problem with one fix. It isn't. An agent's token bill has two distinct causes, and they need two genuinely different kinds of tool. This post is about telling them apart — because once you can, the question stops being "which tool wins" and becomes "which of my two problems am I looking at right now." The bill is mostly things you didn't choose Start with where the tokens actually go, because it's rarely where people assume. When your agent calls a tool, the model doesn't just pay for your request — it pays for the machinery of asking. Anthropic's own pricing documentation spells this out: the tools parameter alone adds hundreds of tokens of schema to every request, the bash tool adds a fixed overhead, and a single web fetch pulls the fetched page straight into your context — "Average web page (10 kB): ~2,500 tokens... Research paper PDF (500 kB): ~125,000 tokens". A tool result you glance at once and never need again can cost more than the entire conversation around it. Now add the part that repeats. On every turn, a typical agent re-sends its system prompt, its full tool catalogue, its persona, and the conversation so far. The variable part of the request — what you actually typed — is often the smallest thing in the payload. The fixed overhead, multiplied across every turn of a long session, is the bill. So the cost has two shapes, and they're not the same shape: Verbose machine output — JSON tool results, logs, search dumps, fetched pages, code listings. Big, one-off, an

The next humanoid robot might not have a head. It might not have legs. It might even sit on a wheeled base and fold down like a deck chair. But, as Genesis AI puts it, "humanoid robots don't need to look human." That explains the look of Eno, the new robot from the French startup […]

GitHub has introduced the GitHub Copilot app, a desktop control centre for agent-native development that aims to keep engineers in charge while AI agents handle more coding work. Mario Rodriguez writes on the GitHub blog that the recent wave of coding agents has brought faster delivery but also "disjointed workflows, more context switching, and too much time spent reviewing agent-generated code". By Matt Saunders

มีคำถามที่น่าสนใจเกิดขึ้นระหว่างใช้งาน AI — "มันรู้ได้อย่างไรว่าต้อง return อะไร?" คำอธิบายที่ AI ให้มักฟังดูซับซ้อนและน่าประทับใจ แต่คำตอบที่ตรงไปตรงมากว่านั้นคือ: มันเห็น pattern นี้มาหลายล้านครั้งแล้ว LLM คิดแบบมนุษย์จริง ๆ หรือไม่? คำตอบคือไม่ — แต่มันทำบางอย่างที่ให้ผลลัพธ์คล้ายกับการคิดได้อย่างน่าทึ่ง ลองนึกภาพคนที่ได้อ่านโค้ดทุกบรรทัดที่เคยถูกเขียนบน GitHub, Stack Overflow, เอกสาร library ทุกตัว รวมถึงบทความด้าน programming จากทั่วโลก แล้วจดจำ pattern ทั้งหมดนั้นไว้ LLM คือสิ่งนั้น เพียงแต่ทำในระดับที่มนุษย์ไม่สามารถทำได้ Tokenization: AI มองโค้ดอย่างไร? เมื่อส่งโค้ดให้ AI ประมวลผล มันไม่ได้อ่านทีละตัวอักษร แต่แบ่งข้อความออกเป็น token ซึ่งเป็นชิ้นส่วนที่มีความหมาย pythondef greet(name): return f"Hello, {name}!" โค้ดนี้อาจถูกแบ่งเป็น token ประมาณนี้: def / greet / (name / ): / \n return / f"Hello / , / {name} / !" แต่ละ token ถูกแปลงเป็นตัวเลข (vector) แล้ว model จึงประมวลผลตัวเลขเหล่านั้น Attention Mechanism: ทำไม AI ถึง "เข้าใจ" Context ได้ ส่วนที่น่าสนใจที่สุดของ LLM คือ attention mechanism — กลไกที่ทำให้ model รู้ว่าเมื่อจะ predict token ถัดไป ควรให้ความสำคัญกับส่วนไหนของ input ที่ผ่านมา ตัวอย่างเช่น เมื่อ model กำลังจะเขียน error handling ใน function มันจะวิเคราะห์: ชนิด exception ที่ function อาจ throw pattern ของ error handling ที่ปรากฏในโค้ดใกล้เคียง library ที่ใช้อยู่และวิธีที่มักจัดการ error ทำไม AI จึง Hallucinate บางครั้ง? เพราะ LLM ไม่ได้ "รัน" โค้ดในกระบวนการคิดจริง ๆ มันแค่ทำนาย token ถัดไปจาก pattern ที่เคยเห็น เปรียบได้กับคนที่ศึกษาโจทย์คณิตศาสตร์มาอย่างมากมาย พอเห็นโจทย์ใหม่ก็เขียนวิธีแก้ออกมาดูสมเหตุสมผล แต่ถ้าโจทย์นั้น novel และไม่เคยเห็น pattern ที่คล้ายกันมาก่อน ก็อาจให้คำตอบที่ผิดได้ นั่นจึงเป็นเหตุผลสำคัญว่าทำไมต้อง test โค้ดที่ AI เขียนทุกครั้ง สรุป LLM เขียนโค้ดได้ดีเพราะสามเหตุผลหลัก: เห็น pattern มาในปริมาณมหาศาล, มี attention mechanism ที่ช่วยเชื่อมโยง context, และถูก fine-tune ให้ output มีประโยชน์จริง การเข้าใจกลไกเหล่านี้ช่วยให้ใช้งาน AI ได้ฉลาดขึ้น — รู้ว่าเมื่อไหรควรเชื่อผลลัพธ์ และเมื่อไหรควรตรวจสอบเพิ่มเติม ด้วยความสามารถของ

json.loads(response) fails at a certain point. You told the model "return JSON only," but it added a ```json markdown code fence around everything. A quick regex strips it — until that regex hits an edge case, and that edge case blows up in production. Since Ollama 0.3.0, passing a JSON schema to the format parameter eliminates this problem at the root. The model's inference itself is constrained by the schema, so no code fences, no explanatory text, no mid-thought artifacts. Just parseable JSON. I ran these tests locally with Gemma4 and Ollama 0.30.7 to see how well it holds up in practice. Why LLM Response Parsing Is Tricky The most common problem when running Ollama locally — without a cloud LLM API — is JSON parsing. Two reasons. First, text generation models are trained toward "natural text." Even if you ask for JSON only, they'll often wrap it in json ... blocks or prepend "Of course! Here is the JSON you requested:" style text. Here's what I reproduced directly: json Input: 'Give me 3 Python tips as JSON with keys: tips (array), difficulty (1-5)' Model output (no format parameter): ```json { "tips": [ "Master the fundamentals first...", ... ] } JSON parse: FAILED Python ' s `json.loads()` can ' t handle the markdown wrapper . The " JSON only " instruction is unreliable in production . Second , speed . I measured the same query both ways : 32 seconds without structured output , 5 seconds with it . More on why below . ## How the Ollama format Parameter Works Ollama ' s `/api/generate` endpoint has a `format` field. Pass a JSON schema object and Ollama applies **constrained decoding** during inference. python import json import urllib.request def ollama_structured(prompt, schema, model="gemma4:e4b"): payload = { "model": model, "prompt": prompt, "format": schema, # ← pass JSON schema object directly "stream": False, "options": {"temperature": 0} } data = json.dumps(payload).encode() req = urllib.request.Request( " http://localhost:11434/api/generate ", data=data

The Disinformation Supply Chain: How Coordinated Influence Campaigns Are Built Before They Go Viral Article from Digital HUMINT Series, For better understanding read the full report Right now, somewhere on X/forum people are fighting about a post that feels real raw, emotional, perfectly worded to hit a nerve. It has the right language, the right anger, the right timing. It sounds like someone who thinks exactly the way you do, or exactly the way you hate. It wasn't written there. It wasn't written today. And the person who wrote it doesn't care about the issue at all. That post was created two or three days earlier, on a hidden forum or a private chat group, following a set of instructions that described who to target, what emotions to trigger, which platform to use, and how much the job pays. By the time you see it, the operation has already worked. You engaging with it for or against is the whole point. I've spent almost two decades watching these hidden spaces where online manipulation is planned. What I've learned isn't that fake content exists everyone knows that by now. What most people don't realize is that it works like a factory. There's a production line. There are workers, managers, and paychecks. And just like any factory, if you know where to look, you can see the product being assembled before it ever reaches the shelf. It Works Like Any Other Business We talk about "disinformation campaigns" as if they're political movements. Some are. But more and more, what you're actually looking at is a business with four steps, each handled by different people, often in different countries. Step 1 — Someone writes the plan. A person with a goal and a budget writes a document that says: push this story, target these kinds of people, make them feel this emotion, use this language, post it on these platforms. These plans used to appear on hidden internet forums. Many have moved to private Telegram groups, but the structure hasn't changed since I first saw it in 201

One of the most persistent traps in modern software development is the tendency to turn architectural styles into absolute dogmas. We see it constantly on social media and inside engineering rooms: teams arguing over cloud native versus cloud agnostic as if they are choosing a lifelong political alignment. A recent perspective from the engineering team at GeekyAnts titled "Cloud-Native and Cloud-Agnostic Are Not Ideologies; They Are Business-Stage Decisions" cuts through this industry noise. Looking critically at their argument, it becomes clear that many organizations are suffering from premature architectural complexity. Engineering leaders frequently romanticize absolute portability long before their business has the operational maturity or the market validation to justify it. The core takeaway is simple yet profound: your architectural choice should be a reflection of your business stage, not a philosophical stance. The Go To Market Trap In the earliest stages of a business, the primary goal is not infinite scalability. The primary goal is survival. A startup needs to discover product market fit before running out of capital. This requires maximum release velocity, rapid experimentation, and minimum operational overhead. For an early stage company, leveraging a cloud native approach is entirely rational. Relying on managed databases, serverless functions, provider native identity management, and integrated monitoring allows a tiny engineering team to focus entirely on product features. The critical flaw in many early architecture reviews is treating this cloud dependency as a failure. It is actually a deliberate speed asset. At this stage, worrying about vendor lock in is a distraction because if you do not find customers quickly, there will be no vendor left to be locked into. Changing Priorities as the Business Matures The architecture that helps a company launch is rarely the one that sustains its long term growth. As a software product gains traction, the op

An open-source security guide, compliance checklist, and LLM-based virtual CISO persona for startups -- with specialized coverage for foreign companies entering the Korean market. The Problem Startups are vulnerable. Limited resources, no dedicated CISO, and security always deferred to "later." But customer data and intellectual property accumulate from day one -- and legal obligations apply regardless of company size. Three incidents from Korea in the first half of 2026 demonstrate that one misconfiguration can cascade into existential damage: Tving Data Breach (2026.06): Mass exposure of CI (Connecting Information, Korea's digital identity key) and refund bank account numbers. Classified as a "major breach" by the Personal Information Protection Commission. The leaked CI enables cross-service identity correlation, multiplying the damage. (CTI-2026-0604-TVING) CU Convenience Store Delivery Hack (2026.06): A simple web vulnerability led to the exfiltration of CI, addresses, phone numbers, and 9+ other data fields. The leaked data was linked to illegal private investigator inquiries and secondary crimes. (CTI-2026-0604-CU_BREACH) FastCampus / DayOne Company GitHub Master Key Theft (2026.06): A single GitHub master key was exfiltrated, granting attackers 30 days of undetected access to internal systems. Over 700,000 user records were exposed. The company took approximately 30 days to detect the breach, and customer notification was delayed beyond 72 hours. (CTI-2026-0611-FASTCAMPUS_DAYONECOMPANY) The common thread: All three began with a single misconfiguration or a single overlooked vulnerability. None required sophisticated zero-days. The damage was inversely proportional to organizational maturity. What startups need is not a $100K security suite. It is knowing what to do first , and a system to check it regularly . The Core Hypothesis An LLM can serve as a startup's first CISO. As of mid-2026, Claude 4, GPT-4o, and DeepSeek V3 -- alongside locally-run models via O

"An AI agent just built a production landing page, with GDPR audit logs and encryption baked in. I wasn't even at my desk." That is not a lucky one-shot. It is a repeatable workflow. Piotr Karwatka recorded a full tutorial showing how to go from idea to a production-ready app on Open Mercato - the AI-Engineering Foundation Framework for CRM/ERP - with no babysitting and no ping-pong prompting. This is the technical version: what the loop actually looks like, why it doesn't fall apart, and which patterns you can lift into your own stack. The problem with conversational coding The default AI coding loop is single-threaded and human-bound: prompt -> generate -> you spot a bug -> correct -> re-prompt -> repeat It holds for snippets. It collapses the moment the task touches real architecture - multi-tenancy, RBAC, event flow, encryption, audit logging. Corrections pile up in the context window, the agent loses the thread, and you are back to typing. You are the bottleneck, sitting in the inner loop. The workflow in the tutorial moves you to the outer loop : you review a finished, tested PR instead of every keystroke. goal -> agent: branch + implement + test + open PR -> you: review PR The reason this is even possible on Open Mercato is that the hard architectural decisions are already encoded as conventions, specs and agent-readable skills ( AGENTS.md , task routing, spec skills). The agent is not inventing how RBAC or GDPR logging should work - it reads the foundation and follows it. 1. Fire-and-forget: the autonomous PR loop The execution agent owns the full unit of work: 1. git checkout -b feat/lead-capture-landing 2. implement against framework conventions 3. run the test suite (Playwright integration tests included) 4. open a structured PR: what changed, why, how it was verified You are no longer correcting tokens. The deliverable is a reviewable artifact. In the tutorial the output is concrete: a live site capturing leads straight into the Open Mercato CRM, with GD

In this blog post, we will see how I use Qwen Code's slash commands and workflow strategies to build Achu my screenshot beautifier app without burning through tokens or losing context mid-session. If you haven't heard of Achu , it's a desktop app built with Electron + React + TypeScript. It does screenshot beautification, Privacy Guard (offline OCR redaction), Auto-Vibe (palette-extracted backgrounds), and an AI Bug Agent with GitHub integration. It's a side project I'm genuinely proud of, and Qwen Code has become my go-to agentic coding CLI for it. A developer shares their day-to-day workflow for using Qwen Code, an open-source agentic coding CLI, to build Achu, a desktop screenshot beautification app built with Electron, React, and TypeScript. The post covers how slash commands like /init, /plan, /compress, /remember, and /btw are used to manage context, reduce token costs, and maintain consistent output across sessions. The core approach centers on spec-driven planning through iterative /plan sessions before any code is written, combined with parallel subagents for independent tasks and strict context hygiene using /compress and /clear. Additional practices include pointing the model at library source code instead of documentation and using /remember to persist architectural decisions across sessions. This isn't a tutorial about what Qwen Code is. It's about how I actually use it day-to-day, the slash command tricks I rely on, and the discipline it takes to get real work done with an LLM in a terminal. It all started with Google Antigravity, but the 5 hours reset and weekly limits is killing my productivity and thinking flow. I had to switch to more affordable and open source model where I chose Qwen. Why Qwen Code? I've tried Claude Code, Gemini CLI, and a bunch of others. Qwen Code is open source, has excellent subagent support, a rich slash command system, and Qwen Max is genuinely strong at reasoning through complex TypeScript and Electron internals. My go-to

🚀 Technical Briefing: This tutorial is part of our deep-dive series on Agentic Workflows at Gate of AI . For the full technical breakdown, interactive code sandbox, and the native Arabic translation, visit the original article here . <span>Tutorial</span> <span>Advanced</span> <span>⏱ 45 min read</span> <span>© Gate of AI 2026-06-16</span> Learn how to fine-tune large language models (LLMs) to enhance communication capabilities in specialized domains, such as homeless shelters, using modern AI tools and techniques like LoRA. Prerequisites Python 3.10+ OpenAI API key (latest version) Familiarity with machine learning concepts What We're Building In this tutorial, we will embark on a journey to fine-tune a large language model (LLM) to cater to the specific communication needs of homeless shelters. By leveraging a bespoke dataset compiled from the Youth Spirit Artworks (YSA) Tiny House Empowerment Village website, we aim to create a model that can effectively assist in the nuances of communication required in such environments. The finished project will result in a model capable of generating contextually relevant and empathetic responses to inquiries typical within the homeless shelter community. This involves structuring data into a standardized question-and-answer format to enhance the training process, ensuring the model's outputs are aligned with the communication style and needs of the target audience. Setup and Installation To begin, we need to set up our development environment with the necessary tools and libraries for model fine-tuning. We'll be using Python along with the OpenAI library to interact with the LLMs. pip install openai pandas numpy Additionally, you'll need to configure environment variables to securely store your API keys. This ensures that sensitive information is not hardcoded into your scripts. .env file OPENAI_API_KEY=your_openai_api_key Step 1: Data Collection and Preparation The first step in fine-tuning our model involves collecting and

Coding agents draft specs, architecture docs, changelogs, and README updates in seconds — but a human still has to judge the quality of all that output. The bottleneck shift A year ago, the typical workflow was: you write a spec, you get comments, you revise, then you implement and get code review. Humans did most of the writing and coding. Now, agents produce first drafts of design docs, API references, runbooks, and onboarding guides — and they do it in seconds. Code implementation and code review can now be handled by agents, so those are no longer the bottleneck. What surfaced instead is the step right before: document review. A human has to read 2,000 lines of generated markdown and decide what's wrong. The writing part got dramatically faster. LLMs can assist with document review too, but compared to code implementation and code review, the human judgment required is still larger. This asymmetry compounds fast. Every agent-assisted project now has a stack of "needs human review" documents growing in a shared folder. If you're running multiple agent loops in parallel — one for the spec, one for the implementation plan, one for the test strategy — review becomes a pipeline stall. GitHub PRs remain the right tool when you need third-party review. But the step before that — the fast local self-review loop where you and your agent iterate on a draft — doesn't belong in a PR. Branching, diffing, and assigning reviewers is a lot of process for a first draft the agent wrote in seconds. Why prose feedback is lossy The most common workaround today is to have the agent read the document and then fix things based on natural-language feedback: "The error handling in section 3.2 is too vague — be specific about what happens on timeout." This looks reasonable. The agent reads it, searches for something about error handling, and makes a change. But several things go wrong: Position is ambiguous. If section 3.2 has three paragraphs about error handling, which one did the revie

vibe citing: how KPMG used AI to write a report about AI and AI made them look like fools by t474-r0b07 There are companies that charge you to tell you how to use AI responsibly. KPMG is one of them. 250,000 employees. 138 countries. Decades advising governments and corporations on how to avoid costly mistakes. In October 2025 they published a report titled "Total Experience: Redefining Excellence in the Age of Agentic AI" . They wrote it with AI. The AI invented 88% of the sources. Nobody verified anything. They published it anyway. // what is "agentic AI" — because the title matters An agentic AI is not a chatbot. Not the assistant that answers your questions. It's a system that makes decisions and executes actions on its own, without a human approving each step. You give it an objective and it acts, corrects, moves forward. It's the product everyone in the tech sector was selling in 2025. KPMG was selling it too. That's why they needed a report proving their clients were already using it. Spoiler: they weren't. And the report invented it anyway. // the forensic analysis GPTZero — a company specialized in detecting AI-generated content — ran a full audit on the report. First: what is an AI hallucination, because the term is going to come up a lot. When a language model doesn't have the information you ask for, it doesn't say "I don't know." It generates a response that sounds correct. It invents with the same confidence it would use if it actually knew the truth. Perfect format. False content. No warning. That's a hallucination. Now the numbers from the KPMG report: TOTAL CITATIONS: 45 REAL CITATIONS: 5 INVENTED CITATIONS: 40 ACCURACY RATE: 11.1% 40 of 45 citations have invented titles, authors that don't exist, or sources that don't say what KPMG claimed they said. Half of the factual claims in the report are false or misattributed. A firm that charges for intellectual rigor published a document with 11% accuracy. // the organizations that read the report and sai