AI 资讯

Here's the problem nobody talks about: the reason most AI outputs are mediocre isn't the model — it's that you asked for a final answer and got one. A model with no friction produces the path of least resistance. It pattern-matches to "good-enough" and stops. It doesn't know what your bar for quality is. It doesn't know what logic you'd push back on, what tone would make your audience tune out, or what structural flaw a sharp reader would catch in the first 30 seconds. It just fills the token space with the most statistically probable response and calls it a day. So the output hits your clipboard. You read it. You sigh. Then you spend 40 minutes editing something that should have come out right the first time. There's a better way — and it exploits the fact that AI critique is significantly sharper than AI generation. The Core Insight: Models Are Better Critics Than They Are Authors This sounds counterintuitive, so stay with me. When you ask an LLM to generate something from scratch, it operates in "produce plausible content" mode. The pressure is to fill the blank. But when you ask a model to critique an existing piece — especially if you hand it a specific evaluative persona — it switches into "find the gap between what is and what should be" mode. That's a fundamentally different cognitive task, and it's one where models consistently perform better. Research on iterative self-refinement in LLMs (Madaan et al., 2023) shows that when models are given their own output and asked to improve it with explicit feedback criteria, quality scores improve substantially across writing, code, and reasoning tasks. The key variable wasn't model size or prompt verbosity — it was the presence of a structured feedback loop. The mechanism is simple: the critique generates tokens that constrain and guide the rewrite. Those critique tokens become working context. The model rewrites against them. The output is necessarily better-fitted to the evaluation criteria than anything a single-

RQ (Redis Queue) is a delightfully simple way to run background jobs in Python. That simplicity is also why teams under-monitor it: it just works, until a downstream API gets slow or a bad deploy ships, and jobs start failing in bulk — quietly. Here's what to watch and how to get alerted before a customer tells you. RQ failures don't announce themselves When a job raises, RQ moves it to the FailedJobRegistry and moves on. The worker keeps running; nothing crashes. If you're not looking at that registry, the failure is invisible — the same trap BullMQ, Celery, and every robust queue share. So the job is to reach into the queue's state and turn it into a signal. The four signals that matter for RQ Failure count / rate — jobs landing in the FailedJobRegistry over a window. Backlog — how many jobs are queued vs. being worked; is the worker keeping up? Latency — how long jobs take, and how long they wait before a worker picks them up. Worker liveness — are your workers actually alive and heartbeating? Where to read them RQ exposes queue and registry state directly: from redis import Redis from rq import Queue from rq.registry import FailedJobRegistry , StartedJobRegistry redis = Redis () q = Queue ( " default " , connection = redis ) queued = len ( q ) # backlog failed = FailedJobRegistry ( queue = q ) # failures started = StartedJobRegistry ( queue = q ) # in-flight print ( " queued: " , queued ) print ( " failed: " , len ( failed )) print ( " started: " , len ( started )) Poll this on an interval and store the series — a single snapshot hides the trend , which is the part that matters. For failures specifically, walk the registry to get the actual exceptions: for job_id in failed . get_job_ids (): job = q . fetch_job ( job_id ) print ( job . id , job . exc_info . splitlines ()[ - 1 ] if job . exc_info else "" ) Two gotchas: Group by exception, not by job. A thousand jobs failing with the same traceback is one incident. Normalize the message (strip IDs, timestamps, host

"Exactly what that dialog looked like between the government and Anthropic and OpenAI is unclear."

A lot of Robot Framework projects still look like plain Selenium scripts with .robot file extensions. Someone imports webdriver , creates driver = webdriver.Chrome() , then calls find_element and send_keys in Python helpers. Robot Framework runs the suite, but readable keywords, shared libraries, and consistent waits never show up in the tests. If you already use Robot Framework with SeleniumLibrary , you do not need the raw WebDriver API. SeleniumLibrary gives you high-level keywords. The Page Object Model gives you structure. Together they keep tests short and UI changes localized. We published a small MIT template that shows the layout: rf-seleniumlibrary-pageobject-template . It targets Sauce Demo — clone it, run four tests, fork the folder structure. What breaks when you mix in raw WebDriver driver = webdriver . Chrome () driver . find_element ( By . ID , " user-name " ). send_keys ( " standard_user " ) driver . find_element ( By . ID , " password " ). send_keys ( " secret_sauce " ) driver . find_element ( By . ID , " login-button " ). click () Fine for a script. Painful in a growing suite. Locators spread across helpers and test files. Waits become time.sleep(2) in one place and missing in another. You end up maintaining SeleniumLibrary and a parallel WebDriver stack. CI fails on a Tuesday night and you are not sure which path opened the browser. Before and after Before After driver.find_element(...).send_keys(...) Login With Valid Credentials ${VALID_USER} ${VALID_PASSWORD} Locators in every file LoginLocators.USERNAME in one module Ad-hoc sleeps wait_until_element_is_visible in BasePage.click() Two browser stacks One SeleniumLibrary instance per suite Four layers Layer Job Example Locators Selectors per screen login_locators.py BasePage Shared waits and actions click() , enter_text() Page library Screen keywords LoginPage.login() Robot test Scenario only Inventory Should Be Visible Folder layout in the repo: resources/locators/ → selectors pages/ → Python pa

"Show a smile, hide the blade." — The 36 Stratagems, Conceal a Dagger in a Smile Previously on...

I’ve been thinking about a small image-processing problem lately: how to reduce an image to a limited palette without making it look muddy. This comes up in a lot of places: pixel art tools printable pattern generators low-color previews LED matrix displays icons and small thumbnails craft or grid-based workflows The easy version is: pick the nearest color for every pixel. The hard version is: keep the important shapes readable after the palette gets much smaller. Nearest color is only the baseline A simple nearest-color pass usually works like this: Take each pixel. Compare it with every color in the target palette. Pick the closest one. Replace the pixel. That gives you a valid output, but not always a good one. The problem is that closest is local. It does not know whether the whole image still reads well. A face can lose warm midtones. A shadow can turn into a flat dark blob. A small highlight can disappear. Skin, fur, fabric, and background colors can collapse into the same bucket. So palette reduction is not just a color problem. It is also a structure problem. RGB distance can be misleading A common first attempt is Euclidean distance in RGB: function rgbDistance(a, b) { return Math.sqrt( (a.r - b.r) ** 2 + (a.g - b.g) ** 2 + (a.b - b.b) ** 2 ); } This is easy to implement, but it does not match human perception very well. Two colors can be numerically close in RGB and still feel different. Other colors can be farther apart numerically but visually acceptable. A better approach is to compare colors in a more perceptual color space, such as Lab or OKLab. You still have to be careful, but the distance metric starts closer to what the eye notices. Dithering helps, but it changes the style Error diffusion, like Floyd-Steinberg dithering, can preserve gradients and perceived detail with fewer colors. That is useful when the output is meant to look like a low-color image. But dithering is not always desirable. In grid-based outputs, it can create scattered single-p

The Smartest AI Workflow I Have Ever Seen Ran on Three Pages of Prompt Project managers are quietly building their own AI chief of staff. The duct tape is the interesting part. A few weeks ago I was talking with a project manager who runs large industrial projects. Real ones, with safety officers and subcontractors and go-live dates that cost serious money when they slip. Somewhere in the conversation he mentioned, almost apologetically, a side project of his. Every week, he feeds an AI model his project charter, the project plan, the risk register, the action tracker, and the last six weeks of status reports. Then he adds the current week's meeting notes and any relevant emails. On top of all that sits a prompt he has iterated on for months. It covers three A4 pages in font size 10. Out the other end comes a list of specific open topics he needs to chase down before writing his end-of-week status report. He has a second prompt that helps him prepare sharp questions for the weekly team meeting. A third one, about 200 lines, assembles everything and drafts the status report itself. He even runs scenario checks: the safety officer found discrepancies during vehicle inspections, the subcontractor says compliance takes two extra weeks, does this move the critical path and the go-live date? He called it manual and clunky. I think it is one of the most sophisticated AI workflows I have ever seen a working professional build, in any field. And I have been building software for a long time. But he was right about the clunky part. And the reason it is clunky tells you almost everything about where AI in project work is actually stuck. The analysis was never the hard part Here is the thing he said that stuck with me, close to verbatim: The AI is good at analysing lots of text sources. The challenge is to obtain all the information, and the effort to write it down comprehensively. Read that again. The intelligence is not the bottleneck. The bottleneck is assembly. Every single

A fintech RAG pipeline poisoned its own vector store and the LLM-as-a-judge validator approved every hallucination. The fix: gate writes with code.

Muse Image allows users to generate AI images using photos from public Instagram accounts. As long as a person's profile is public, another user can tag that account and use their images as part of an AI-generated creation.

I created Pylon to make it easier to move from hobby projects to full production apps. When I work on hobby projects, I usually use React or Next.js because they are quick to set up and easy to deploy on Vercel. For production apps, I separate the frontend and backend, then deploy the backend on AWS. But setting up a full backend on AWS can be complex and costly, especially for simple apps. Pylon is a full-stack, real-time framework that includes server-rendered React, TypeScript functions, enti

Runtime authorization for Claude, Cursor, Codex and OpenClaw Discussion | Link

I run a small AI lab and playground and got super excited about Anthropics paper "Verbalizable Representations Form a Global Workspace in Language Models" ( https://transformer-circuits.pub/2026/workspace/index.html ) It talks about how they use a tool they call a Jacobian Lens to view inside the middle layers of LLM while it's working before it commits to a word (token). I wanted to see if I could get a version of this running on the open models and to my surprise it worked! I ran some experime

The company is taking a modular approach to designing these chips, anticipating that their needs will change as AI evolves rapidly by the time the chips are in production.

Partner for your most ambitious work Discussion | Link

https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf https://developers.openai.com/api/docs/guides/latest-model

Windows 11 updates could soon include fixes for more security issues at once. Microsoft said in a blog post on Thursday that it's now using AI to "identify potential issues earlier," which means "customers will see a higher volume of security updates included in each security release." Hackers, even amateurs, have increasingly been using AI […]

About two weeks after OpenAI's GPT-5.6 was caught up in regulatory drama - rolled out only to government-approved organizations during a "limited preview" period - the company has received the Trump administration's greenlight for a public rollout of the model. OpenAI CEO Sam Altman called it "the best model we have ever produced." To celebrate, […]

Devthropology is a passion project built on top of GitHub pull data. The name is a play on developer anthropology. Pull request data can be cut a lot of ways. The functionality has been built out of curiosity as I want to see different insights into codebases that I work on. Some of the data is typical and other parts I haven't seen elsewhere. I think of this as an improved GitHub Insights page, with faster performance, more detail, and a focus on how work moves through a codebase. The main enti