Tristan Buckmaster on Artificial Intelligence

Participant in the calls, describing a personal collaboration

The reasons are below, and they involve our being pressured by outside factors.

Source and context

Original text

Tristan Buckmaster’s full account of research progress and discussions with OpenAI (opens in a new tab) - Tristan Buckmaster

About this source

Complete firsthand written account, including the September 6 conversations, proposed publication arrangements and the limits of his data-access knowledge. The PDF itself has no verified publication timestamp. No recording of the disputed calls has been located; this is Buckmaster's account, not an independently established transcript.

Original text

Buckmaster announces papers and links his statement (opens in a new tab)

About this source

Rendered page displays September 7 at 11:58 PM, edited September 8 at 12:08 AM. Date-only retained because the rendered timezone is not labeled.

Original text

Buckmaster urges attention to the implications of frontier models (opens in a new tab)

About this source

Follow-up asks labs to set aside posturing and focus on what model capabilities mean. Not a retraction of his account.

Before the quotation

His statement announces work with Alpöge and explains why the papers were released before the presentation was fully polished.

After the quotation

At publication he had not seen OpenAI's proof and expressly said he did not know whether their data was used. In a separate September 8 post, he urged labs to focus on the broader implications of frontier models. That follow-up did not withdraw his account.

Condemned
How this statement is classified

Criticizes the conduct of the negotiations; this does not establish that data was taken.

Recorded on
Published here
People and groups discussed
OpenAI, Sébastien Bubeck

More from this case

Andreas Thom

De-identification may remove a name; it does not remove the intellectual content of a mathematical idea.
Read statement
Read the case