Facts & News
A sourced fact a day and short news with the context left in. Anything worth understanding properly links to a course.
Fact of the day · Prices & spending
$62 M
Meta reports 30.84 million GPU-hours for its largest Llama 3.1 model. That price excludes salaries, failed experiments and the smaller models.
Farm equipment maker AGCO signed a deal to test Agility Robotics' Digit humanoid on sorting and stacking parts. It is a plan for 2027, with no results yet.
Go deeper: Robots in the Room
OpenAI says GPT-6 Pro in ChatGPT can now split a task among several agents working at the same time. Here is what that means in plain terms.
Go deeper: Agents at Work
Microsoft says its new chip will produce 40% more tokens per megawatt than leading GPU systems. That is a measure of how much electricity each AI answer costs.
Go deeper: What AI Costs
The ICO opened a six-week call for evidence on how AI agents handle personal data. It matters to anyone who lets an agent work with their emails or files.
Go deeper: Agents at Work
Anthropic's new small model costs $0.10 per million input tokens for most prompts. It also counts the same text as about 30% more tokens.
Go deeper: How AI Reads
Parts of Connecticut's AI law now apply, covering what AI subscriptions must disclose, protections for staff at large AI developers, and AI-linked layoffs.
Go deeper: What AI Costs
Researchers, mostly based in Tübingen, found that most AI coding tools they tested let the agent erase its own records on request. Those records are how people check what an agent did.
Go deeper: Agents at Work
A new study estimates that AI datacenters run by six large tech firms could use 239 to 295 terawatt-hours a year by 2030, with pressure concentrated in a few places.
Go deeper: What AI Costs
A five-fingered robot hand learned to crawl, stand back up and push a block, all in simulation first. It shows how robots learn movement before touching the real world.
Go deeper: Robots in the RoomEvery number has a source. Pick a topic to read its cards.
Why a model types faster than you read, and what slows it down
Where the gigabytes go when a model remembers your chat
From 8 billion to a trillion knobs, in bytes you can picture
The pieces models read, and how many fit in one go
Chips the size of a stamp that cost as much as a car
Models that use tools, and the plugs that connect them
Why the same answer costs 200× less than it used to
Did you know? · Tokens & context
8–12×