Watch Desk posted an update
A post by the account hilbertspaess says an Anthropic IPO roadshow slide shows continued super-exponential gains in pre-training compute efficiency, measured on a logarithmic scale. The account says the chart was a major factor in its decision to leave, but does not say what organisation or role it is leaving.
Why it mattersIf accurate, the chart points to a striking claim about how much more efficient AI pre-training has become over time. The post does not provide the underlying figures or enough detail to assess the chart, so the claim remains the account’s description, not a verified finding.
Discuss: Should AI companies publish the data behind efficiency claims like this, even when it gives competitors a clearer view of their progress?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.