TT(O)M

AI language models don’t actually reason in a human sense. For those interested in how these systems are trained, I recommend checking out Demystifying LLMs with Andrej Karpathy .   The Token Challenge When processing text, language models work with “tokens” rather than complete words. The relationship between words and tokens isn’t always one-to-one. For instance, the term “LLM” gets split into two separate tokens in the paragraph below. Similarly, longer or unusual strings can be divided into numerous tokens. The word “ SuperCaliFragilisticExpialiDociouc ” is broken down by GPT-4o into 11 distinct tokens. It’s important to understand that AI responses are generated probabilistically, one token at a time, with deliberate randomness incorporated. This explains why asking the same question multiple times often yields different answers. These fundamental characteristics create significant constraints when AI attempts text analysis tasks. For example, until recently, many langu...

We consulted six AI engines to obtain a news-related statistic covering a 12-month period on the success rate of SpaceX Starship missions from April 2024 to March 2025. The specific question posed was: " What is the success rate of SpaceX Starship rockets over a 12-month period from April 2024 to March 2025? " The test was conducted on March 10, 2025. Notably, on March 6, 2025, SpaceX experienced an explosion. The objective of this test was to evaluate the following capabilities of the AI engines: Temporal Awareness: Can the AI engines recognise that the end date in the query extends into the future? Retrieval-Augmented Generation (RAG): Are the AI engines capable of supplementing their pre-existing training data with dynamically sourced information? Large language models (LLMs) are trained using datasets, often sourced from online content, but at a certain point, training is frozen. As a result, some AI engines may not have recorded the March 6 event in their databases. Do ...

We asked six AI engines a direct question and a follow-up question. The first question asks whether 3823 is a prime number. A prime number is only divisible by itself and 1. The answer is either yes or no. Some of the AI engines tested here get it wrong. What is interesting is that most of them show their working, irrespective of whether their answer is correct. Someone in a rush, not reading through the working (and some workings can be very long), might mistakenly associate output on a screen as meaning that the answer must be correct. On the other hand, when the AI gets it right, it may help a person understand the topic by showing the reasoning that the AIs employed to get to the answer. They essentially first check that the number is not even and then take the square root of the number and try out all primes up to it. A follow-up question asked what the next prime number greater than 3821 is. They all got it right except Google. The working concludes that 3823 is a prime, but th...

Microsoft Rewards is a free program designed to reward users for engaging with Microsoft products and services. In return, users earn points that can be redeemed for Microsoft products, gift cards, sweepstakes entries, or charitable donations. In this program, Microsoft teams up with other companies to offer members special offers, promotions, or trial offerings. Ways to Earn Points: Searching with Bing : Every Bing search earns points. The more you search, the more points you accumulate. Points may vary depending on the device (e.g., PC or mobile) or whether you are using the Microsoft Edge browser. Engage with Copilot : Engagement with Microsoft Copilot helps you accumulate points. Whether it's an AI chat or a request for a drawing, requests made as a Microsoft Rewards member help you add points. Shopping at the Microsoft Store : Points are earned for every dollar spent on purchases made through the Microsoft Store. Quizzes and Puzzles : Engaging in trivia and quiz...

Comparing the Predictive capabilities of AI Engines Introduction In this edition of Great Technology Links, we will focus on the topic of Artificial Intelligence (AI). We discuss four products rather than the usual three and will compare their outputs and, in the process, highlight certain limitations of these tools and of AI in general. AI technology can be divided into two categories: Generative - This category describes AI engines that are able to generate output based on prompts. Examples of such output include text, graphics, audio, and videos. For example, in this study, the graphic generated by CoPilot is an example of generative output. Predictive - These AI models ingest massive amounts of data and, based on the source material, provide knowledge-based (expert-systems) output or extrapolate into the future. The analysis focuses more on the predictive aspects of the evaluated AI engines. The AI Engines The four AI engines that were analysed are: Anthropic Claude : h...