Training
The process of showing a model examples and adjusting its parameters until its answers improve.
Training is the expensive part. The system makes a prediction, is told how far off it was, and nudges its parameters toward a better answer. Repeat that several trillion times, across thousands of processors running for weeks, and the result is a model. Nothing is memorized in the way a database stores a record, which is why a model can answer a question nobody asked it during training, and also why it can invent an answer with complete confidence.
Where it mattered in the record
-
Jensen Huang defends AI distillation as competition
On 28 September 2026 Nvidia CEO Jensen Huang told CNBC’s Squawk Box that training AI models on rivals’ outputs is ‘not robbery’: ‘That’s called competition.’ ‘You’re allowed to test somebody else’s products all you want,’ he said. Nvidia’s products get stripped ‘down to bones.’
-
OpenAI commits $1 billion to cyber defense for essential services
On 3 September 2026 OpenAI committed $1 billion in subsidized access, training and technical support for its Daybreak cyber defense program, aimed at water, power, local government and nonprofit defenders.
-
Washington Post describes Anthropic plan to scan and destroy books
The Washington Post reported on 27 January 2026 that unsealed court filings described an internal Anthropic program, called Project Panama, to buy, scan and destroy millions of physical books in order to build training data for Claude.
-
EU AI Act obligations for general-purpose models take effect
From 2 August 2025 the EU AI Act required providers of general-purpose AI models newly placed on the market to keep technical documentation, publish a summary of their training data, comply with EU copyright law, and share information with regulators and downstream developers.
-
Two federal judges rule AI training on books can be fair use
Judge William Alsup ruled on 23 June 2025 that training a language model on books was transformative fair use, while sending Anthropic’s acquisition of more than seven million pirated books to trial. Judge Vince Chhabria issued a narrower fair-use ruling for Meta two days later.
-
DeepSeek releases R1, an open-weight reasoning model
The Chinese laboratory DeepSeek released R1 on 20 January 2025, an open-weight model trained with reinforcement learning to work through problems step by step. It matched OpenAI’s o1 on several benchmarks at a small fraction of the reported training and inference cost.
-
Amazon doubles its Anthropic investment to $8 billion
Amazon said on 22 November 2024 that it would invest a further $4 billion in Anthropic, taking its total commitment to $8 billion while remaining a minority investor. The deal named AWS Trainium as Anthropic’s main training chip and expanded joint work on that chip.
-
Record labels sue Suno and Udio over training data
Sony Music Entertainment, UMG Recordings and Warner Records, coordinated by the RIAA, sued Suno in Massachusetts federal court and Udio’s Uncharted Labs in the Southern District of New York on 24 June 2024, alleging both trained on copyrighted recordings without a license.