ทันเอไอ

Lucius AI proposes teaching bid-writing AI to 'refuse to lie' instead of fabricating empty content

Startup Lucius AI reveals its approach to designing AI for bid and tender documents: teaching the system to 'refuse to write' when there's no supporting evidence, rather than making things up

The biggest problem with using AI to help write tender/bid documents is that AI tends to 'hallucinate' or fabricate false information to make the content look credible. In bidding work, this is disastrous, as it could cost a company its credibility or get it disqualified from the tender altogether.

Davor Jerković, founder of Lucius AI, explained in the company's blog that his team chose to tackle this problem differently. Instead of trying to make the AI always write complete content, the system is designed to 'refuse' to write sections that lack supporting data.

The mechanism works by checking the bidder's actual capabilities against the document's requirements before writing begins. If it finds sections without sufficient supporting evidence, the system won't fabricate content. Instead, it flags them in a 'partner slot,' displaying a warning banner at the top of the document listing any unverified items — such as missing specific qualifications — so the team can find partners to fill the gaps.

Even more interestingly, the system will refuse to draft a document from the outset if it assesses that the bid has 'no chance of winning' or isn't worth submitting, preventing the creation of documents that look good but lack any basis in reality.

This concept aligns with analysis in the legal tech and procurement sectors, which suggests that for work requiring high accuracy, an AI's ability to 'not write what it doesn't know' is more important and necessary than trying to appear more capable than it really is. However, since this information comes solely from the developer's own blog, it should be viewed with caution as it may be part of the company's product marketing.