About
Who writes PARALLAX, and how its articles and model cards are sourced.
Updated
01 Jacopo Parretti
Jacopo Parretti works on machine learning, with a focus on model architecture and inference. He trained in artificial intelligence at the University of Verona.
His projects include attention ablations in vision transformers and meta-learning for cold-start recommendation. He runs open-weight models on his own hardware, to measure what quantization and speculative decoding change in practice.
PARALLAX reads the frontier labs the same way, from their primary sources and their numbers.
02 Sources
Every article starts from what the labs publish themselves: launch posts, documentation, papers, system cards and pricing pages.
Every number carries a footnote. The footnote names the page, its publisher and the day it was read, with the link.
Claims that come from the press are attributed to the outlet by name. A figure that shows something schematic says so inside the figure.
03 Models
The Models section shows the current lineup of eleven AI labs, one card per model. Each card is checked against the lab's own pages every week.
When a price or a specification changes, the card shows the day it was updated.
04 Corrections
A mistake, or a source that says otherwise, can be reported to info@parallax.ninja. Corrections are made in the article and dated at the top.