Artificial Intelligence on AI
in one paragraph define artificial intelligence and give an example
Claude
Artificial intelligence (AI) refers to computer systems designed to perform tasks that would normally require human intelligence — things like understanding language, recognizing patterns, making decisions, or learning from experience — by processing data and identifying statistical patterns rather than following a fixed set of explicit rules for every scenario. A practical example is a large language model like Claude or GPT, which is trained on vast amounts of text to predict and generate human-like language, allowing it to answer questions, write code, or hold a conversation despite never being explicitly programmed with rules for every possible sentence it might produce.- (440 billion parameter, approx 300GB)
Gemma3 1B
Artificial Intelligence refers to the ability of computer programs to mimic human cognitive functions, like learning, problem-solving، reasoning， and creativity - essentially enabling them to perform tasks that typically require intellect. It achieves this by creating algorithms and statistical models designed to analyze vast amounts of data, allowing machines to learn patterns and make decisions without explicit programming for every individual situation. A key component involves training AI with large datasets – the more data, often the better – to improve performance over time through feedback loops. An example of current widespread use is recommendation systems like those used on Netflix or Amazon suggesting content based on user preferences; another significant area includes self-driving cars utilizing machine learning for navigation and obstacle recognition (1 billion parameters, approx 800MB)
AI Ethics and the Case for Controls
Artificial intelligence presents society with a genuine, unresolved trade-off rather than a simple story of progress or peril. On the benefit side, AI has demonstrated real capacity to accelerate scientific research, improve medical diagnosis, expand access to education and expertise for people who couldn’t otherwise afford it, and automate tedious or dangerous work freeing human effort for higher-value tasks. Tools like large language models already lower barriers to coding, writing, and analysis for people without specialist training, a genuine democratising effect.
Against this sit real, documented risks. Widespread automation threatens significant labour market disruption, and the distribution of AI’s economic gains is far from guaranteed to be equitable ; the benefits may concentrate among those who already own or control the technology. AI systems can also encode and amplify existing biases present in their training data, producing discriminatory outcomes at scale in areas like hiring, lending, or policing. Misinformation and synthetic media, including deepfakes, threaten public trust and democratic processes. At a more systemic level, increasingly capable and autonomous AI systems raise open questions about alignment, accountability, and control ; ensuring systems reliably do what’s intended, and that responsibility for harm remains traceable to a human or institution.
This is precisely why controls matter, not as an obstacle to progress but as the condition that makes broadly beneficial progress possible at all. Sensible controls, transparency requirements, safety testing before deployment, human oversight for high-stakes decisions, and clear accountability structures, aim to preserve AI’s genuine benefits while limiting its capacity for harm. Without them, the risks aren’t hypothetical constraints on some future system; they’re already-observable failure modes waiting to scale further as the technology becomes more capable and more widely deployed. The question isn’t whether to develop AI, but whether development proceeds with enough foresight and governance to keep its benefits from being outweighed by the harms it could otherwise cause.
Broad areas within the field
- Machine learning- systems that improve at a task by learning from data rather than being explicitly programmed for every case. It encompasses a huge range of mathematical and statistical models that learn patterns from data and use those patterns to make predictions or decisions. Examples include models based on Bayes Theorum
- Large language models (LLMs)- models trained on huge volumes of text, underlying tools like Claude, GPT, and the open-weight models run locally via Ollama or LM Studio
- Computer vision- systems that interpret and process visual information from images or video
- Retrieval-augmented generation-combining a language model with a search step over a specific set of documents or notes, covered in more detail under the retrieval-augmented generation of this garden.
- Agentic AI - systems that can take multi-step actions, use tools, or operate somewhat autonomously toward a goal, rather than just responding to a single prompt
Limitations of AI
AI systems, particularly language models, are not infallible sources of fact. One of the most important limitations to understand is hallucination; where a model generates information that sounds plausible and confident but is factually incorrect or entirely fabricated. This is covered in detail in its own note: Hallucination.
Position in this vault
This structural note anchors the AI-related notes in this vault, connecting practical, hands-on work Local AI, Ollama & LM Studio. Of particular interest to me is retrieval-augmented generation or RAG whch utilises AI models, frequently smaller sized models or models with specialised reasoning capabilities to work with fixed data/ information sets that are known sources of truth or reference.