162 articles
Browse guides
Text-to-image systems pair a language understanding component with a diffusion generator, so a prompt steers the denoising toward matching pixels. Output quality depends heavily on prompt specificity — subject, style, composition, and lighting all need saying. Commercial use raises rights questions that vary by tool and jurisdiction. In practice: ‘Isometric illustration of a data centre, muted palette, soft rim light.’
Current text-to-speech produces natural prosody and can clone a voice from a short sample. That capability is the reason voice cloning consent and disclosure have become live legal and ethical issues rather than theoretical ones. In practice: A course narrated in a consistent voice across 40 lessons, without a studio.
Text-to-video extends diffusion across time, which adds the hard constraint that frames must stay consistent with each other. Clip length, physical plausibility, and character consistency are the current limits. The field moves fast enough that any specific capability claim ages within months. In practice: A ten-second product shot generated instead of filmed.
Throughput is the capacity question, distinct from latency’s speed question. Batching raises throughput while raising individual latency, which is exactly the trade you want for offline jobs and exactly wrong for chat. In practice: Batch a million classifications overnight; nobody is watching the clock.
Models do not see letters or words; they see tokens, which are chunks of characters produced by a tokenizer. English averages roughly four characters per token, so 750 words is about 1,000 tokens. Tokens are also the billing unit and the unit of the context window, which makes this the most commercially relevant term in the glossary. In practice: ‘unbelievable’ might split into ‘un’, ‘believ’, ‘able’ — three tokens, one word. ...
A tokenizer maps text to integers using a fixed vocabulary learned from data. It explains several famous quirks: models miscount letters because they never see letters, and non-English text often costs more tokens for the same meaning because vocabularies skew English. In practice: The same sentence in Greek can cost two to three times the tokens it costs in English.
Tool use is what lets a model do things it fundamentally cannot do alone: look up current data, run exact calculations, write to a system. The model chooses the tool and the arguments; your code executes it and returns the result. Tools are also the point where a model’s mistakes become real-world actions, so permissions matter. In practice: Rather than guessing today’s rate, the model calls a currency API.
Top-p keeps the most likely tokens until their cumulative probability reaches p, then samples from just those. Unlike temperature it adapts to how confident the model is at each step. Tuning both at once usually makes behaviour harder to reason about — pick one. In practice: Top-p 0.9 ignores the long tail of unlikely tokens without flattening the whole distribution.
TPUs are ASICs designed specifically for the tensor operations in neural networks, available through Google Cloud rather than as hardware you buy. They are one example of a broader move to purpose-built AI silicon as general-purpose GPUs stop being the only option. In practice: Large training runs on Google infrastructure typically use TPU pods.
Training data is the corpus used to fit a model’s parameters. Its coverage sets the model’s ceiling: gaps, errors, and skew in the data reappear as gaps, errors, and bias in the outputs. Under the EU AI Act, training data has a formal definition and, for high-risk systems, governance obligations attached. In practice: A hiring model trained mostly on past hires from one demographic will reproduce that pattern.