The claim as announced

KAIST's official press release headlines the technology as "an AI that gives answers without going through a server," adding that it "reuses knowledge obtained once." As presented, the system operates independently of server communication and draws on knowledge it has already acquired rather than obtaining the same information anew each time.

The release was published on KAIST's official news page with three accompanying images, all of which were confirmed accessible during the review.

What the review found

Two qualifications accompany the finding. First, the ruling carries a "does not" condition — a limitation on the technology's operation that narrows the scope of the claim as headlined, and which readers should treat as part of the claim itself. Second, the review confirmed a discrepancy in the reduction rate figure: the number cited in the press release differs from the figure stated in the paper's abstract. Because the two documents do not state the same improvement rate, the magnitude of the reported gain should be read with caution until reconciled against the full paper.

Transparency notes

• The review drew on 10 primary sources.
• The three images accompanying the announcement are hosted on KAIST's official press release page; the third is explicitly identified within the release as an AI-generated image.
• A Naver search-results page relevant to the verification could not be retrieved during the review and requires direct capture before it can be cited.

Weighing the verified core of the announcement — an AI that answers without server communication and reuses knowledge obtained once — against the attached "does not" condition and the press release-versus-paper discrepancy in the reduction rate figure, our desk's judgment on the claim is: Mostly True.