A look at the challenges some AI developers face in building models to extract trillions of high-quality tokens from PDFs, which are hard to parse, for training

Last November, the House Oversight Committee had just released 20,000 pages of documents from the estate of Jeffrey Epstein …

The Verge 2026-02-24 Josh Dzieza

@dorialexander Alexander Doria on x
Ah ah. Another part of my timeline that has not moved in a year. [image]
@prietschka Paul Rietschka on bluesky
Uh, just going to point out that there's been zero progress on AI building “complex software,” and AI cannot, and will not, solve “advanced physics problems.” — These claims are lies, plain and simple. [embedded post]
@theverge.com @theverge.com on bluesky
Despite rapid progress in AI's ability to build complex software and solve advanced physics problems, the ubiquitous format of PDF remains something of a grand challenge. — Read more from @joshdzieza.bsky.social: www.theverge.com/ai-artificia... [image]
@joshdzieza Josh Dzieza on bluesky
I wrote about the unconquerable frontier of AI capabilities: reading a pdf
@alex.bsky.team Alex on bluesky
oh man I was working on ML for parsing PDFs over a decade ago and you could never get me to touch that space again www.theverge.com/ai-artificia...
@tcarmody Tim Carmody on bluesky
The bill for storing so much information online in formats designed for printing rather than in HTML was always going to come due, although I confess this wasn't how I imagined it [embedded post]

Chronicles