Tech Tips, Tricks & Trivia

This Week I Learned -  * OpenAI's open weight models gpt-oss 20B and gpt-oss 120B support advanced reasoning and tool use. Open models are more easily customizable to build AI that can reason over your enterprise data and domains, providing a powerful option alongside proprietary models. gpt-oss can be used alone or alongside models like GPT-4o, Claude, or Llama. Built with a Mixture of Experts architecture gpt-oss delivers low-latency performance for use cases like search, chat, and real-time decisioning. It features a 131k context for long documents and RAG and it is provided under an Apache 2.0 license. * A typical software-application startup that’s not involved in training foundation models might spend 70-80% of its dollars on salaries, 5-10% on rent, and 10-25% on other operating expenses (cloud hosting, software licenses, marketing, legal/accounting, etc.).   Many of Meta’s properties rely on user-generated content (UGC) to attract attention, which is th...

Notes from the short course " Reasoning with o1 " on DeepLearning.AI, presented by Colin Jarvis of OpenAI - o1 is a reasoning model for complex tasks that require broad general knowledge, including function calling and image input. It can reason through complex tasks in domains like mathematics, coding, science, strategy, and logistics. o1 is different from other models because it thinks before it speaks. o1 requires less context in prompting to produce very effective results. o1 uses large-scale reinforcement learning to generate a chain of thought before answering. o1's chain of thought (CoT) is longer and higher quality than what you can typically attain by a prompt alone. CoT  contains behavior like: - Error correction - Trying multiple strategies - Breaking down problems into smaller steps o1 performs well at understanding images out-of-the-box. It can be used to extract a detailed JSON that describes the image and what's going on in it. o1 follows a test-and...

Summary of a recommendation from GitHub blog  (April 2025) Balance between cost and performance : Go with GPT-4.1, GPT-4o, or Claude 3.5 Sonnet. Fast, lightweight tasks : o4-mini or Claude 3.5 Sonnet Deep reasoning or complex debugging : Claude 3.7 Sonnet, o3, or GPT 4.5. Multimodal inputs (like images) :  Gemini 2.0 Flash or GPT-4o. Currently you can switch between 5 AI Models within the Copilot Chat interface. The OpenAI Models are hosted on Azure tenants. GitHub Copilot uses Gemini 2.0 Flash hosted on Google Cloud Platform (GCP). Claude 3.5 Sonnet is hosted on Amazon Web Services and so when you're using Claude 3.5 Sonnet, prompts and metadata are sent to Amazon's Bedrock service. When faced with options, I like asking all my AI assistants what they think. I liked the answer from Gemini. I added additional points from my own reading of the GitHub Copilot documentation & the answers from other AI assistants to come up with this analysis breakdown: 1. GPT 4o (...

This Week I Learned -  * Officially announced by Mozilla in 2010, Rust is a systems programming language that delivers C++-level performance while eliminating entire classes of bugs. Rust is well-designed for performance, safety, and concurrency. Rust is a top choice for Wasm modules, enabling fast, secure, browser-based applications. * Microsoft ran a hackathon " Accelerate App Development with GitHub Copilot " on the DevPost platform for developers to build apps using GitHub Copilot using Azure services.  The submissions that are available for review provide proof of what's possible. * "AI tools can be unreliable—AI is often only as good as the data it is trained on and the manner it has been trained. But humans are often inadvertently overawed by advancements in technology. This may result in some of us trusting AI chatbots blindly...it’s imperative we enable and build systems with trust at the core if AI has to reach its true transformational potential." - B...