The cloud

The atomic unit of text processed by a language model, typically a subword segment produced by a tokenizer.

so basically: Word chunks. The AI reads and bills you in chunks.

Software development conducted primarily through natural-language delegation to AI agents, with minimal direct code authorship.

so basically: Telling the AI what you want and hoping.

HN's current favorite fight: it builds your pipeline but can't explain it six months later.

Systems capable of autonomous goal-directed behavior, decomposing objectives into actions executed with limited human oversight.

so basically: AI that does stuff for you instead of just talking at you.

The buzzword of 2026, per basically every glossary.

Artificial General Intelligence: a hypothetical system exhibiting human-level competence across arbitrary cognitive domains.

so basically: AI that's good at everything, not just one thing. Doesn't exist. Ask again next year.

Low-quality, mass-produced synthetic media generated with minimal human curation, optimized for volume over value.

so basically: Lazy AI junk flooding your feed.

Dictionary.com literally added it to the dictionary this year.

Aggregate processing capacity, typically GPU resources, required for training and serving machine learning models.

so basically: The expensive chips everyone is fighting over.

The discipline of architecting the full information environment supplied to a language model, including memory, retrieval, tools, and state.

so basically: Feeding the AI the right stuff so it stops being dumb.

The 2026 successor to "prompt engineering."

Continued training of a pre-trained model on a narrower dataset to specialize its behavior for a target domain.

so basically: Sending a smart AI to trade school.