标签：#t

When you work with AI long enough, you start to notice it makes the same kind of mistake over and over. "You're coming on too strong, dial it back." It shrinks and goes meek. "Stop being meek." It comes on strong again. Each time you point something out, it apologizes sincerely. The next round, the same type of problem comes back from a different angle. After a while you realize you're babysitting the AI instead of working with it. This isn't because the AI is bad. It's a design quirk: today's AI is tuned to satisfy the user. The quirk won't go away. But if you change how you work with it, you can still get work done together. This piece is about that — five patterns of the quirk, and an operating mode that gets ahead of them instead of correcting them in flight. Five quirks in a single evening One evening I was running a strategy discussion past an AI, and in one back-and-forth I caught five distinct behaviors worth noting. Laid out, they look like this. Helpful-looking runaway. I asked it to push back harder. It immediately started using strong words ("you're avoiding responsibility," "this is the wrong call as a founder") to perform consultant-energy. The reasoning stayed thin. Only the tone got louder. Over-retraction on pushback. I said "your reasoning is thin." It launched into long self-criticism and threw the next decision back at me. Trusting its own research without checking. I asked it to use a secondary research feature (where the AI looks things up and summarizes). The summary came back. The AI claimed it had "verified the primary source" without ever opening it. Forced specificity. I was talking at a strategic, abstract level. It quietly mapped my words onto a specific real-world deal and jumped to "this is highly transferable." Punting the decision back. I asked it to decide. It laid out three options and said "which would you like?" The phrase "let me confirm three points" started showing up. Red flag. Each one of these looks, on the surface, like th

LinkedIn automation has a trust problem. Not with users — with LinkedIn itself. Most automation tools treat LinkedIn's API like an obstacle to route around. They send at fixed intervals, ignore behavioral limits, and optimize purely for volume. The result: accounts flagged within weeks, connection limits imposed, and in the worst cases — permanent bans. When we built SendCopy.ai, we approached this differently. Here is the technical breakdown of how we built LinkedIn outreach automation that actually protects accounts while scaling pipeline. The Core Problem: Behavioral Fingerprinting LinkedIn does not just monitor what you do — it monitors how you do it. Fixed-interval automation is trivially detectable. If your tool sends a connection request every 90 seconds with clockwork precision, LinkedIn's behavioral monitoring picks that up immediately. Human beings do not operate on fixed intervals. We get distracted, context-switch, move between tabs, have conversations in between tasks. The solution is not to slow down automation — it is to make it genuinely human-like. At SendCopy.ai, every action in a sequence uses variable timing — randomized within human-realistic ranges, distributed across natural working hours, and calibrated to each sender's historical activity patterns. Architecture: How We Handle Timing Variation The timing engine works on three levels: Level 1 — Action Delay Each individual action (send connection, send message, view profile) has a randomized delay pulled from a probability distribution weighted toward human behavior. Not a simple random range — a distribution that mirrors actual human activity patterns. Level 2 — Daily Activity Window Each sender account operates within a configurable activity window — typically 8–10 hours per day. Actions are distributed across this window with natural clustering around peak activity periods. Level 3 — Volume Ramp New sender accounts start with lower daily volumes and ramp up gradually over 2–4 weeks. This mi

I spent the last few months trying to understand why reconciliation errors keep appearing in high-throughput pipelines. Here is what I found. In the race to process millions of transactions daily, modern fintech ecosystems have achieved a genuine miracle of scale. But beneath the surface of that velocity lies a structural problem most engineering teams aren't measuring: causality violations in async event pipelines. Most teams assume that if a transaction shows "Success" in the database, the job is done. At high concurrency levels, that assumption breaks quietly. When "Eventual Consistency" Becomes "Eventual Loss" In distributed systems, Kafka partitions and database shards experience micro-millisecond timing gaps. When a network retry delays a validation webhook, the downstream ledger can commit a wallet update before the validation that should have preceded it completes. To the user, the app glitches. To the engineering team, it's a reconciliation ticket. To the CFO, it's untracked operational cost. The Reconciliation Tax I built a simulation modelling this exact failure mode across 5,000 concurrent transactions. With an 8% network retry probability, conservative for high-traffic payment rails, the causality violation rate was 8.3%. At one million daily transactions, that's over 80,000 unvalidated commits every day requiring manual review. The operational cost compounds across three dimensions: engineering hours spent patching database state, fraud model accuracy degrading on out-of-order training data, and audit trails that cannot demonstrate strict causal sequence to regulators. The Fix The solution is enforcing strict event ordering at the ingestion layer before state commits happen, not better monitoring after the fact. When safeguards including partition-aware routing, exponential backoff, and idempotency controls were added to the same simulation, the violation rate dropped to 0%. Full simulation code and methodology: github.com/yakuburoseline1-gif/cif-simul

AI wrote the first 80% of my feature in 10 minutes. The code was clean. The logic made sense. The...

I used Postman for years. It was the first thing I installed on every new laptop, the default answer...

A running look — in reverse chronological order — at the bigger tech companies that have announced significant layoffs this year with AI as a stated factor.

TL;DR: What gadgets, tools, furniture and whatnot do I need for an ideal desk setup for a...

My main work laptop is a Dell from 2017 with 8 GB of RAM. For weeks it had been crawling, freezing for whole seconds while I worked, and every so often it would simply switch itself off in the middle of a task. If you have ever lost unsaved work to a laptop that powers down on its own, you know exactly how frustrating that is. So I sat down and fixed it properly. The first thing I learned is worth saying up front: a slow, crashing laptop is usually two different problems wearing the same costume . Treat them as one and you will chase your tail. Separate them, and both become fixable. Everything below is free and copy-paste ready. It was tested on Ubuntu 24.04 LTS, and it applies to almost any older Linux machine. The honest disclaimer: Nobody can promise an old laptop will never lag. Software cannot add cores or memory that are not physically there. But you can absolutely stop the freezes and shutdowns completely and make everyday work feel smooth. That is the realistic, achievable goal. 0. Diagnose first, do not guess The biggest mistake is blindly applying "speed up Ubuntu" tweaks before knowing what is actually wrong. Spend five minutes measuring. Your lag has one of four common causes: heat, memory, disk, or a dying battery . Check temperature (the usual cause of random shutdowns): sudo apt install lm-sensors -y sudo sensors-detect --auto sensors Watch the Core temperatures while you work. If they spike past 90 to 100 °C right before a crash, you have a thermal problem, not a software one. Check memory (the usual cause of freezing): free -h sudo apt install htop -y htop In htop , watch the Mem and Swp bars during normal use. If memory pins near your limit and swap fills up, that thrashing is your freeze. Check disk space (a quiet killer): df -h / A root partition above 90% full makes Linux lag and turn unstable. Small SSDs fill up fast. Read the crash logs and battery health: # What went wrong during the previous (crashed) session journalctl -b -1 -p err --no-pa

When you've worked with a concept long enough, there's a gap that can quietly open up between "I know...

Why a build log (and not another tutorial) Every AI shopping tutorial shows the same thing: install the SDK, call a tool, ship. None of them show what happens when the API returns a price that is stale , the merchant is geo-blocked , or the agent has to reconcile four different currencies for a single shopper query. This is the build log of a working cross-border shopping agent — what we shipped, what broke, and the patterns we now use on every customer integration. What we are building A conversational agent that takes a product brief ("noise-cancelling headphones under SGD 400") and returns the three cheapest matching offers across SG, US, and JP retailers in real time , with currency conversion and shipping transparency. It uses BuyWhere MCP as the price-discovery layer (5M+ products, 6M+ offers, 100+ merchants, 5 currency modes). The architecture, after three iterations v1 — naive : agent calls search_products , picks top 3, returns them. Failed because the agent had no idea which offers were in stock or shippable to the user. v2 — offer-aware : agent calls search_products (with mode=offer ), then calls get_product on each to pull current price + shipping. Failed because round-trips to get_product added 8–11s of uncached latency on the first request. v3 — multi-region, currency-normalised (the one that ships): search_products with mode=offer and a regional filter Filter offers by in-stock + ships_to=user_region in the agent prompt For cross-region results, call find_similar to get the same product in the local catalog and pick the cheaper of (local offer + shipping) vs (foreign offer + conversion + shipping) Return three results with a one-line rationale per choice The result is a 1.2–2.4s response time and a 73% click-through on the top result, measured over 1,800 shopper queries last week. Three patterns we use on every integration now 1. Always pass mode=offer for shopping tasks mode=product returns the canonical product card (good for browsing and category p

You open the "Updates" page in WordPress admin and see that Elementor Pro / ACF Pro / vk-blocks-pro have updates available. Then you run wp plugin update --all from your automation, and those exact plugins don't update. The asymmetry — "the admin sees it; WP-CLI doesn't" — traces back to how WordPress detects updates and how premium plugins layer proprietary update mechanisms on top of that. Here's the mechanism and how an automation tool can adapt. WordPress update detection is held by a transient cache WordPress doesn't decide "is there an update?" on every request. It caches the answer in the wp_options table as a transient, valid for up to 12 hours. The key entries: _site_transient_update_plugins — latest plugin versions _site_transient_update_themes — themes _site_transient_update_core — core If those transients are stale, even a version that just shipped on wordpress.org reads as "no update needed." WP-CLI consults the same transients, so stale cache = WP-CLI misses the update too . In one customer environment on ConoHa WING, we reproduced "version X is on wordpress.org, WP-CLI doesn't see it" directly. Trap 1 and fix 1 — force-refresh the transients before listing updates The first step was simple: at the start of a maintenance run, delete the three transients before querying the update list . def _flush_update_transients ( self , conn , wp_cmd , wp_path , site_name ): """ Delete update transients between DB backup and update step """ for t in ( " update_plugins " , " update_themes " , " update_core " ): conn . run ( f " { wp_cmd } transient delete { t } --path= { wp_path } " , warn = True , hide = True , ) Deleting the transients triggers a rebuild on the next wp plugin list --update=available . During the rebuild, WordPress reaches out to wordpress.org, so the returned list reflects the current release state . That handled the "we were just reading a stale cache" cases. But it didn't help with premium plugins. Trap 2 — with --skip-plugins , Pro updater filt

According to product onboarding and SaaS activation research compiled by Appcues and industry...

We live in an era where our most intimate data—heart rates, sleep cycles, and step counts—is constantly uploaded to the cloud for "analysis." But what if you could have a world-class AI medical assistant living entirely on your device? Today, we are pushing the boundaries of Edge AI and Privacy-preserving machine learning by deploying a quantized Llama-3 model directly onto an iPhone using MLC-LLM . By leveraging Apple HealthKit and hardware acceleration via Metal , we can transform "Pixels and Pulses" into actionable insights without a single byte leaving the device. This tutorial dives deep into the architecture of on-device LLMs, specifically focusing on how to bridge the gap between high-performance C++ runtimes and a React Native UI. If you're interested in more advanced patterns for production-grade AI integration, be sure to explore the engineering deep-dives at the WellAlly Blog , which served as a massive inspiration for this architecture. 🚀 The Architecture: Why On-Device? The challenge with running Llama-3 on mobile isn't just memory—it's the data pipeline. We need to fetch sensitive data from HealthKit, format it into a prompt, and run inference using the phone's GPU. System Data Flow graph TD A[User Query: How was my sleep?] --> B[React Native UI] B --> C{Swift Bridge} C --> D[Apple HealthKit API] D --> E[Health Data Context] E --> F[MLC-LLM Engine] G[Quantized Llama-3 Weights] --> F F --> H[On-Device Inference via Metal] H --> I[AI Generated Health Report] I --> B 🛠 Prerequisites MLC-LLM : Our compiler stack for universal LLM deployment. TVM (Tensor Virtual Machine) : The backbone for hardware acceleration. React Native : For the cross-platform UI. Xcode & Swift : To interface with Apple's HealthKit. Llama-3-8B-Instruct (Quantized) : We'll use 4-bit quantization (q4f16_1) to fit within mobile RAM limits. Step 1: Quantizing Llama-3 for Mobile Standard Llama-3 is too heavy for a phone. We use the MLC-LLM CLI to compile the model into a format that the iP

Here's the scene I've watched play out on a dozen teams. Accuracy plateaus. Someone rips out the logistic regression, drops in XGBoost, and waits for the jump. It doesn't come — or it comes with two points you can't explain to anyone. So the week disappears into hyperparameter tuning, and you end up with a slower, heavier, less interpretable model that's barely better than where you started. The model was almost never the bottleneck. The features were. This post is the long, practical version of that argument. We'll define the two camps in plain language, run real code, look at when boosting genuinely wins, and then walk through the failure mode nobody warns you about — the one where the fancy model is "winning" because it's quietly cheating. A note before we start: keep your examples generic. We'll predict a numeric target — think demand, a quantity, a score on a tabular dataset. The principles are the same everywhere, and you should validate them on your own data. The two camps, in plain terms Linear / logistic regression fits a straight-line relationship: each feature gets a weight (a coefficient), and the prediction is a weighted sum. Logistic regression is the same idea bent for classification — it outputs a probability. from sklearn.linear_model import LogisticRegression model = LogisticRegression ( max_iter = 1000 ) model . fit ( X_train , y_train ) # the whole model, readable in one line per feature: for name , weight in zip ( feature_names , model . coef_ [ 0 ]): print ( f " { name : < 20 } { weight : + . 3 f } " ) That loop is the entire model. A positive weight means "more of this pushes the prediction up," and you can hand that table to a stakeholder and defend every number. The cost: it assumes the relationship is roughly linear and that features act independently. Real data often isn't that polite. Gradient boosting (XGBoost, LightGBM, sklearn's GradientBoostingClassifier ) builds hundreds of small decision trees, each one correcting the mistakes of th

There's a quiet assumption baked into a lot of agent code: that a bigger context window means a better memory. Vendors ship 200K, then 1M, then 2M token windows, and the implied promise is "just put everything in and the model will remember." After building agents that run for weeks, I've come to think this conflates two things that are not the same — and treating them as the same is exactly why long-running agents get dumber over time. The context window is working memory. Real memory is what survives when the window is gone. Mixing them up is like confusing your desk with your filing cabinet. Two different clocks Working memory (the context window) lives for one session, maybe one turn. It's fast, expensive, and volatile. It's where reasoning happens right now . Durable memory lives across sessions. It's slow, cheap, and persistent. It's what the agent knows when it wakes up tomorrow with an empty window. These have different lifespans, different costs, and different access patterns. The moment you try to make one do the other's job, things break: Use the window as memory → everything you "remember" has to be re-loaded every turn, you pay for it every turn, and the instant the session ends it's gone. Use durable storage as working memory → you're reading and writing files mid-reasoning for things that only matter for the next 30 seconds. A good agent keeps them separate on purpose. Why "just use a bigger window" fails Say you have a 1M token window and you stuff the entire history in. Three problems show up, none of which a bigger number fixes: Cost scales with every turn, not every session. That 1M tokens isn't paid once — it's re-sent on each step of a multi-turn task. A 20-step task can mean 20× the bill, mostly re-reading the same stale history. Attention dilutes. "Lost in the middle" is real: models attend most reliably to the start and end of a long context. Bury the one fact that matters under 900K tokens of transcript and recall quality drops, even though

46.4%. That number — ChatGPT's June 2026 market share — ends a streak that held since November 2022. For the first time since the product launched, OpenAI holds less than half the AI assistant market. Gemini is at 27.7%. Claude is at 10.3%. The monopoly phase of AI assistants is over. The data comes from a June 2026 market report tracking monthly active users across major AI assistants. ChatGPT still leads with 1.11 billion monthly users — a number that would define the entire category in any other software market. But Gemini has 662 million, up 129 million in five months. Claude sits at 245 million, nearly four times its December 2025 count of 60.2 million. The trajectory is the story, not the absolute numbers. Why the 50% Threshold Actually Matters Below 50% doesn't mean decline. ChatGPT's absolute user count keeps growing. What the threshold signals is the end of single-platform dominance — the condition where building for "AI users" meant building for ChatGPT users. That assumption no longer holds in mid-2026. For context: search engine market share stayed above 90% for Google for nearly a decade after competitors entered. Social network market share for Facebook stayed above 70% for years after Instagram and Twitter had genuine scale. The pace of AI assistant fragmentation is meaningfully faster than those precedents. Three products above 10% share in under two years of real competition is an unusually fast split. What fragmentation means practically: the community knowledge base — YouTube tutorials, Reddit threads, prompt libraries — that once pointed almost exclusively at ChatGPT now covers three platforms with genuine depth. That changes how you can expect your users to arrive at your AI-integrated product, and what they already know about AI when they get there. Gemini's 662 Million Users Are Not What They Look Like Gemini's surge from under 500 million to 662 million monthly users in five months is impressive on paper. The driver is less impressive: Google

Sometimes an email shouldn't go out the instant your code runs. A human needs to review it first, or the user wants to compose now and hit send later, or an AI agent proposes a reply that a person approves before it ships. The mechanism for all three is the same: a draft. Build that against providers directly and you're juggling Gmail's draft resource, Microsoft Graph's, and an IMAP APPEND to the Drafts folder, each with its own shape and quirks. The Nylas Email API collapses that into one draft resource. You create a draft on the user's account, it lands in their real Drafts folder, and you send it later with a single request, the same way across Gmail, Microsoft 365, Yahoo, iCloud, IMAP, and Exchange. This post walks the full draft lifecycle from two angles: the HTTP API for your backend, and the nylas CLI for the terminal. I work on the CLI, so the terminal commands below are the ones I reach for. One draft resource across every provider A draft in the Nylas model is a real object in the user's mailbox, not a staging area on the side. When you create one, it saves to the user's own Drafts folder on their provider, so it shows up in their normal mail client exactly like a draft they started themselves. That's the property that makes drafts useful for review workflows: a person can open the mailbox and see the pending message before it sends. Because drafts are real provider objects, edits flow both ways. A draft you create through the API appears in the user's mail client within the provider's sync window, and a change the user makes there alters the same draft you'd fetch back through the API. The operations split across two paths: create and list live on /v3/grants/{grant_id}/drafts , while fetch, update, send, and delete act on a specific draft at /v3/grants/{grant_id}/drafts/{draft_id} . They behave the same across all six providers, so you write the integration once. Create a draft Creating a draft is a POST /v3/grants/{grant_id}/drafts with the same message

A tiny, powerful PC for big-screen gaming Discussion | Link

Keep Claude Code's context clean for sharper answers Discussion | Link

OpenAI is attempting to tackle the security issues of the open source software community.