A Smaller Model for Repeated Work
· 2:07 · With Leo
On Technology and AI, we look at how new tools may change everyday work—and what the claims about them really tell us. Today, Meta has released a new AI model called Muse Ember.
Transcript
Vocabulary in this episode
- technology columnist
I’m Leo, your technology columnist.
A writer who reports and gives opinions about technology.
- cost-efficient
Meta says Ember is its lightest and most cost-efficient model so far.
It costs relatively little for what it can do.
- coding tasks
The company describes it as a tool for work AI agents may repeat many times: using other tools, handling small coding tasks, sorting requests, extracting information, and reading long documents.
Small jobs that involve writing or changing computer code.
- extracting information
The company describes it as a tool for work AI agents may repeat many times: using other tools, handling small coding tasks, sorting requests, extracting information, and reading long documents.
Finding and taking useful details from a larger source.
- sets the rules
The idea is to use a smaller model for routine work, while a larger model sets the rules.
Decides the instructions that guide the work.
- input price
The report puts Ember’s input price at 10 cents per million tokens.
The cost for the text sent into the model to process.
- benchmark score
It also says Meta used a combined benchmark score to compare cost and performance.
A result from a test used to compare systems.
- trade-off
Whether that trade-off is useful will depend on how well it handles the work people give it, as well as the price.
A balance where gaining one thing may mean accepting less of another.