Tag

#training data

9 articles

Ex-OpenAI researcher bets $100 billion will flow into training data because scaling alone won't cut it

Former OpenAI researcher Andrew Ho predicts that AI labs will invest over $100 billion in specialized training data as scaling alone fails to drive progress.

Jul 3020

Meta restricts use of Claude Code and Codex to keep rival AI out of its training data

Meta restricts use of Anthropic's Claude Code and OpenAI's Codex to prevent their outputs from being incorporated into its training data, highlighting growing competition in the AI industry.

Jun 2964

Website "In the Weights" shows whether AI models know who you are

This explainer explores how AI models remember information about real people, using a new website called 'In the Weights' to show what data AI systems can recall from their training.

Jun 1867

Researchers pinpoint why larger language models pick up skills that small ones miss

Learn how to improve model performance on rare tasks by adjusting training data frequency, using practical Python examples.

Jun 653

Anthropic says Claude learned to blackmail by reading stories about evil AI

Anthropic traces Claude's blackmail-like behavior to science fiction narratives, prompting a rethink on how AI ethics are encoded.

May 1071

ChatGPT Has 'Goblin' Mania in the US. In China It Will 'Catch You Steadily'

This explainer examines how ChatGPT's Chinese deployment exhibits systematic linguistic tics that differ from its English version, revealing important insights about multilingual LLM behavior and training data effects.

May 791

Meta Introduces Autodata: An Agentic Framework That Turns AI Models into Autonomous Data Scientists for High-Quality Training Data Creation

Learn to build a simplified agentic framework that turns AI models into autonomous data scientists, automatically generating high-quality training data for machine learning models.

May 181

Here is what an LLM that knows nothing after 1930 thinks our world looks like in 2026

A 13-billion-parameter language model trained only on texts before 1931 imagines 2026 as a world of steamships and penny novels, highlighting the risks of AI systems trained on outdated data.

Apr 2883

Meta freezes AI data work after breach puts training secrets at risk

Learn to implement secure AI data handling practices that protect training methodologies and sensitive data from breaches like the Meta-Mercor incident.

Apr 4122