Want to revolutionize your business with AI? Check SSW's Artificial Intelligence and Machine Learning consulting page.
In ChatGPT prompt engineering, various elements play a critical role in shaping the user's prompt and obtaining the desired output. These elements are role, result, intent, context, and constraint.
When you're building a custom AI application using a GPT API you'll probably want the model to respond in a way that fits your application or company. You can achieve this using the system prompt.
You can give ChatGPT examples of the type of output you want, including the format. Each example you feed it is known as a “shot”.
Understanding when to use 0-shot, 1-shot, or multi-shot prompts can help you get the most out of ChatGPT. These approaches allow the model to learn from different contexts, making it more likely to generate the desired output.
GPT models utilize "tokens" as their smallest units of text. These tokens can be as small as a single character, like "a," or as large as a word, such as "apple." Interestingly, tokens can also represent parts of a word.
For instance, in GPT's perspective, the word "unhappiness" could be tokenized into three parts: "un," "happi," and "ness." Each of these parts is considered a separate token. This flexible understanding of tokens is what allows GPT models to efficiently break down text data for analysis.
AI agents are autonomous entities powered by AI that can perform tasks, make decisions, and collaborate with other agents. Unlike traditional single-prompt LLM interactions, agents act as specialized workers with distinct roles, tools, and objectives.
Repetitive tasks like updating spreadsheets, sending reminders, and syncing data between services are time-consuming and distract your team from higher-value work. Businesses that fail to automate these tasks fall behind.
The goal is to move from humans doing and approving the work, to automation doing and humans approving the work.
You start a new AI feature in .NET, search for "Semantic Kernel", and copy the first sample you find. Six months later the SDK only gets security fixes, and every new capability you need lives somewhere else.
Semantic Kernel and AutoGen are in maintenance mode. Microsoft Agent Framework is their successor, built by the same teams.
When using Azure AI services, you often choose between Small Language Models (SLMs) and powerful cloud-based Large Language Models (LLMs), like Azure OpenAI. While Azure OpenAI offer significant capabilities, they can also be expensive. In many cases, SLMs like Phi-3, can perform just as well for certain tasks, making them a more cost-effective solution. Evaluating the performance of SLMs against Azure OpenAI services is essential for balancing cost and performance.
When building an AI-powered solution, developers will inevitably need to choose which Large Language Model (LLM) to use. Many powerful models exist (GPT, Claude, Gemini, Llama, Mistral, Grok, DeepSeek, Qwen, etc.), and they are always changing and subject to varying levels of news and hype.
When choosing one for a project, it can be hard to know which to pick, and if you're making the right choice - being wrong could cost valuable performance and UX points.
Because different LLMs are good at different things, it's essential to test them on your specific use case to find which is the best.
AI models don't stay available forever. On Azure, a model version is retired about 18 months after its launch. On that day the deployment either switches to a different model by itself or stops working. Either way, the app changes without a release.
That's why every project should know when each of its models retires and move to the replacement on its own schedule.
When you need AI assistance for development but find yourself offline (whether you're on a flight, gone camping, or facing the inevitable zombie apocalypse) you'll appreciate having local LLMs ready in your workflow!
Everyone wants to integrate AI into their application and workflows, but often times you're using the wrong tool for the task. Most AI work in your app isn't writing, it's deciding.
System 1 models are built for that, you ask typed questions and you get typed answers with probabilities in blazing fast times. Generative System 2 models like Claude and ChatGPT shine at jobs that need reasoning or words. Don't guess which one you need - test them on your own data and let cost, accuracy and speed decide.
You tweak a system prompt to fix one complaint, or switch to a newer model because it's cheaper. The app still compiles, the unit tests still pass, and a week later users report that answers got worse. LLMs are non-deterministic, so a classic test that expects an exact string either fails randomly or checks too little to catch the regression.
Evals fix this. An eval runs your AI feature against a set of real examples and scores each output against criteria you define, so you can see whether a change made things better or worse before you ship it.
Azure OpenAI models come with a content filter that blocks harmful prompts and responses. It is on by default, so most teams never look at it. Then a real customer question gets blocked, someone changes a setting by hand in the portal, and nobody can say what the app filters any more.
That's why you should choose the filter settings for your app on purpose, and keep them in the repo.
Embedding a user interface (UI) into an AI chat can significantly enhance user interaction, making the chat experience more dynamic and user-friendly. By incorporating UI elements like buttons, forms, and multimedia, you can streamline the conversation flow and improve user engagement.
Comparing and classifying text can be a very time-consuming process, especially when dealing with large volumes of data. However, did you know that you can streamline this process using embeddings?
By leveraging embeddings, you can efficiently compare, categorize, and even cluster text based on their underlying meanings, making your text analysis not only faster but also more accurate and insightful. Whether you're working with simple keyword matching or complex natural language processing tasks, embeddings can revolutionize the way you handle textual data.
“Your loan is approved under Section 42 of the Banking Act 2025.” One problem: there is no Section 42.
That single hallucination triggered a regulator investigation and a six-figure penalty. In high-stakes domains like finance, healthcare, legal and compliance zero-error tolerance is the rule. Your assistant must always ground its answers in real, verifiable evidence.
AI is a powerful tool, however, sometimes it simply makes things up, aka hallucinates. AI hallucinations can sometimes be humorous, but it is very bad for business!
AI hallucinations are inevitable, but with the right techniques, you can minimize their occurrence and impact. Learn how SSW tackles this challenge using proven methods like clean data tagging, multi-step prompting, and validation workflows.
AI tools are powerful, but they introduce attack surfaces that didn't exist before. It has never been easier to write malicious code, trick systems into leaking data, or poison software supply chains - and AI is at the centre of all of it. The threat landscape is evolving fast, with real breaches happening now.
You should understand the specific risks and have concrete defences in place.
Most AI apps start with an API key, because it is the quickest way to get a first answer from a model. But a key gives full access to the resource and can't be limited to one user or one action. Nobody can tell from the logs who used it.
That's why every app and every person should get only the role they need on your AI resources, with API keys turned off.