Tech Tips, Tricks & Trivia

This Week I Learned -  * Moonshot’s Kimi K3’s weights were released on July 27 under a Modified MIT licence, making the 2.8-trillion-parameter model available for anyone to download, fine-tune and self-host. Giving weights away is a smart distribution strategy as it could potentially turn developers, cloud providers and enterprise IT units into resellers, and make it hard for anyone to under-price you as the base model itself is free. * Jot down annoying things . AI may have a solution. * findphone is a nifty app to locate a nearby Bluetooth device by signal strength, from the macOS command line — for when Find My Phone isn't available due to MDM restrictions * ProHance is a software platform that provides real-time visibility into workforce activities, helping organizations measure productivity, optimize costs, and improve operational efficiency. It tracks employee engagement through metrics such as time spent on applications, mouse and keyboard activity, and task durations, whil...

This Week I Learned -  * " Models learn from "exhaust," the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know-how.  In consuming intelligence, you are creating intelligence. If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself. In the cloud era, enterprises accumulated data. In the AI era, they accumulate learning. A company should be able to use a model without giving up the knowledge that makes it unique. " - Satya Nadella * Wire services such as PTI, ANI, Reuters, AP, and AFP provide verified facts to other outlets, which then adapt the content to their own style. Their business model depends on reader trust in the accuracy of the wire’s reporting and on receiving compensation for the verification work. While search engines direct readers t...

This Week I Learned -  * The OpenRouter Pareto Router is a dynamic model selection tool that allows users to select a coding model based on a minimum coding score preference. It maintains a curated shortlist of strong coding models ranked by Artificial Analysis coding percentiles. Users can set a min_coding_score between 0 and 1 to control the strength of the coder they need. The router automatically selects the cheapest model within the chosen tier or the fastest with the Nitro variant. This approach ensures that users receive a strong coder without overspending, as the exact shortlist evolves over time with new models and benchmarks. The OpenRouter API is compatible with most SDKs, allowing for easy integration into applications. * openrouter/auto-beta is a task-aware router. It classifies each request, routes it to the model teams are actually spending on for that kind of task, and filters that choice by the cost-quality tradeoff you set. The response is priced at the routed mo...

This Week I Learned -  * Microsoft's Phi-4 Mini is widely regarded as one of the most capable models under 4 billion parameters. It punches significantly above its weight class, outperforming older 7B and even 70B models in specific reasoning tasks. It runs at a smooth 13–18 tokens per second on flagship devices with 8 GB+ RAM (like the iPhone 16 Pro/17 Pro or Galaxy S25/S26 series). * From The Batch : - PrismML, a Caltech spinout backed by Khosla Ventures, publicly released compressed versions of Alibaba’s Qwen model on Tuesday, shrinking it from 54 GB to under 4 GB so all 27 billion parameters run on an iPhone (15 or newer). PrismML achieves the compression by reducing the models to ternary or binary quantization, reducing each value from 16 bits to just two or three possible values.   - Two people asking the same model for business feedback, one in Hindi and one in Russian, may receive substantively different advice not because they’re using different models, but because th...

This Week I Learned -  *  arXiv offers HTML papers alongside the existing PDF to improve accessibility . * " Students need to learn how to work with AI, recognize weak output, think through choices, test results and decide what is worth doing. I described my own approach as “delegate maximally”. Instead of keeping a fixed list of AI-proof skills, keep giving AI everything it might be able to do. Whatever remains is the human role for now. " - S Anand * " The reason I dislike this emphasis on term “LLM Visualization” or “Claude Viz” or even “Agentic Viz” is the focus on the tool. No one goes out and says “Excel Viz” // “PowerPoint Viz” // “Figma Viz”. The tool is there to augment the human and has never been the important part.  The craft and skill of the  human using the tool is what we should acknowledge and celebrate. " - Amit Kapoor * Draw with Data by Schubert De Abreu   * Bharat Atlas is a lightweight open-source visualiser for India administrative boundar...

This Week I Learned -  *  Intensive Vibe Coding Course With Google - self-paced Kaggle Learn guide * Forward Deployed Engineers (FDEs) bridge internal development and real-world deployment by embedding with enterprise clients to build solutions while guiding the product roadmap.  FDEs are the "Technical Special Ops" who bridge the gap (The Delta) between a core product and a client’s messy, real-world reality. * Google DeepMind CEO Demis Hassabis and researcher John Jumper won the Nobel Prize in chemistry for their work on the company’s AlphaFold model, and DeepMind has also made major contributions to meteorology, materials science, and a variety of other disciplines. John Jumper is leaving DeepMind for Anthropic.  * Like Claude Code, Claude Science can autonomously carry out meaningful work when given concise, high-level instructions, and it has access to tools that make it particularly useful for research in computational biology and drug development. C...

This Week I Learned -  * "Hallucination is a feature when you’re being creative." - S Anand * "AI doesn't have to be right to be useful. The goal is outputs you can iterate over time — not perfection on the first try." -  Agentic Thinking 101, Udacity * "...language models hallucinate because the training and evaluation procedures reward guessing over acknowledging uncertainty" -  Why Language Models Hallucinate * "Large language models (LLMs) do not work like databases, storing facts in neat rows and columns, and retrieving the right answer when asked. They are trained on enormous collections of text and learn statistical patterns in language. When prompted, they generate a response by predicting what will come next, one piece at a time. An LLM generates text one word at a time, each word a probability drawn from patterns it absorbed during training. A setting called the temperature governs how adventurous those draws are. If the temperature is l...

This Week I Learned -  * Vertex AI has been rebranded and evolved. Google officially replaced and expanded it under the name Gemini Enterprise Agent Platform.  This change marks a major structural shift for Google Cloud, moving away from a traditional MLOps platform (focused on training and deploying standalone models) toward an "Agentic AI" ecosystem.   * Turkey based HubX Studios used Gemma 4 to build BetterSpeak, an AI English tutoring platform that uses the Gemma 4 E2B model as the reasoning engine for its on-device pipeline — enabling private, low-latency tutoring without the need for an internet connection. HubX deployed the 4-bit quantized version of the model to handle tasks like grammar explanations and progress monitoring across languages. By using Gemma 4’s native audio input capabilities, HubX’s app is able to support direct speech-to-speech learning , while reducing costs and ensuring privacy. * Midjourney, the AI lab best known for image generation...

This Week I Learned -  * navigator.sendBeacon() method is used to quietly send analytics or telemetry data, often when a page is closing.  * P4RS3LT0NGV3 is a Universal Text Translator. Think of it as a universal translator for ALL alphabets and writing systems! It is a powerful web-based text transformation and steganography tool with 159 built-in text transforms spanning encodings, classical and modern ciphers, Unicode styles, formatting, and niche alphabets.  * When you see that Claude is "organizing its thoughts" during long conversations—it indicates automatic context management is working. * Fable 5 is now the most capable model of Claude and takes 2X the usage of Opus. Fable 5 is the same base model as Mythos but with cybersecurity guardrails. In the same week of its release, Anthropic has disabled Fable 5 and Mythos 5 for all users after a US directive restricting foreign national access, citing national security concerns. * Meta chief product officer Chris Cox o...

This Week I Learned -  * Anthropic and OpenAI have launched their own services companies, challenging traditional IT firms and service providers. Anthropic's joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman is valued at $1.5 billion, while OpenAI's venture, The Development Company, is valued at $4 billion.  * "Google which is cash surplus, just announced an additional capital raise of $80 bn. Google annual profit is $160 bn, last quarter $62 bn, and market cap $4.5 trillion. That is close to total profits and market cap of all Indian listed companies put together.  It’s a wake up call to all companies  to invest into the future, whatever the present maybe." - Uday Kotak * "One reason that the atrophy of coding skills is concerning is the “paradox of supervision” ... effectively using Claude requires supervision, and supervising Claude requires the very coding skills that may atrophy from AI overuse." - Anthropic * “People who go all i...