nytimes[.]com/2024/08/30/technology/ai-chatbot-chatgpt-manipulation.html?unlocked_article_code=1.HE4.l3JH.qqmYozkR7ICi
github[.]com/aounon/llm-rank-optimizer
When you do an online search, many if not all of the 1st page or two of results may be garbage, depending on what you searched for. That's because of SEO [Search Engine Optimization] -- basically figuring out how to game the search engine algorithms used for page ranking. You have content farms -- wikipedia[.]org/wiki/Content_farm -- vying to be the top results for the most commonly searched terms or phrases, making money from ads.
Still in its early stages, there's a kind of SEO for AI chatbots too. People are figuring out how to influence what chatbots say, and more importantly, recommend. The NYT article, by an author who wanted to counter chatbots disliking him, talks about a few ways that were recommended by AI experts, including a couple of professors who came up with a script for generating STS [Strategic Text Sequences] that: "optimizes the probability of the target product's rank being 1". And it's already a biz:
My first call was to James Cadwallader and Dylan Babbs, the co-founders of a New York-based start-up called Profound.Profound does what it calls “A.I.O.” — A.I. optimization — on behalf of Fortune 500 companies and other large businesses. The company tests A.I. models on millions of different prompts, analyzing their responses when asked about specific products or topics. Its clients then use this information to improve how they appear in chatbots’ answers.
Mr. Cadwallader and Mr. Babbs believe that A.I.O. is the natural successor to S.E.O. — search engine optimization, the $68 billion industry that grew up around businesses and publishers trying to boost their rankings on Google and other search engines.