An anonymous writing test initiated by Vitalik Buterin, a co-founder of the Taifab, has still not been decrypted by AI. On 5 July, he updated the progress of the experiment, stating that 13 days after the launch of the challenge, neither the researcher nor the automated script had found the article that he had published anonymously.

After 13 days, no results.

This test was first published on June 22. Buterin stated at that time that he had published several papers on behalf of others during the current decade in connection with the ITA and invited the outside world to use the writing style to identify whether the texts were from himself.

The latest developments indicate that the existing searches remain untargeted. He gave a new hint on the social platform that many of the AI scripts omitted the category of documents “should have been included” at the time of the search, and therefore the search needs to be further expanded.

Script Missing Part Document Source

According to Buterin, the problem is not entirely in the model capacity itself, but in the narrowness of the search. Many automated processes scan only common sources, such as official blogs, technical regulations and so forth, without covering other possible outlets.

This means that AI remains vulnerable to pre-set templates in the face of open search tasks. As long as sample ranges are incomplete, critical materials may be missed at the initial screening stage even if the model has some text analysis capability.

  • Buterin estimates that about 200 to 2000 copies of relevant texts can be compared
  • The target text is described as an official document related to the ETA ecology
  • No specific document type and posting location has been confirmed to date

Test points to the AI capability boundary.

The test was not only focused on Buterin, but also because it touched the actual boundaries of AI on deanonymization and style recognition. It was often premised on “no privacy on the Internet”, but the experiment at least showed that identification was not easy in a cross-source, non-standard sample environment.

The report also mentions that this result echoes Buterin ' s previous cautious approach to AGI progress. For the time being, AI has shown stability in dealing with a well-structured and well-defined mandate, but there may still be a clear gap in the scenes requiring a broader search of borders to complement the context.