Excellent article Simon. I do think this can be mitigated by future that a) understands time better and thus can implement forms of forgetting. This is inherent to some degree with RL systems, but could also be something learned by better time-tagged training data and meta-knowledge about when recency matters (i.e. not when discussing something historical!)
That's interesting. Like a learned decay function. But I still think cursed and blessed tokens will remain. Like, Kevin Roose will always be known for what happened in Sydney. This could be balanced out by other things he's done to help AIs more recently. But if he doesn't do those things, I don't think time alone will heal the wound :)
Another incredible insight Simon. Love the concept of latent reputation. There are several emerging startups that will do the audits for you. Some ways I have read about that folks can use to influence the weights: rich domain specific structured data, Q&A hubs, plugins for answer engines, the 👍🏽 / 👎🏽 button. These systems are becoming self learning and will need to get into the habit of providing feedback to every query. Always appreciate your thoughts leadership and passion. I always learn a lot!
Thanks, really appreciate it! This is a fast-evolving space, and also complicated by the fact that sometimes AI models are replying based on their own internal networks only, and sometimes from search results provided to them—which may be returned by a different search model entirely. It all gets pretty complicated, but it's clear that it's different than things like PR and SEO.
Excellent article Simon. I do think this can be mitigated by future that a) understands time better and thus can implement forms of forgetting. This is inherent to some degree with RL systems, but could also be something learned by better time-tagged training data and meta-knowledge about when recency matters (i.e. not when discussing something historical!)
That's interesting. Like a learned decay function. But I still think cursed and blessed tokens will remain. Like, Kevin Roose will always be known for what happened in Sydney. This could be balanced out by other things he's done to help AIs more recently. But if he doesn't do those things, I don't think time alone will heal the wound :)
Another incredible insight Simon. Love the concept of latent reputation. There are several emerging startups that will do the audits for you. Some ways I have read about that folks can use to influence the weights: rich domain specific structured data, Q&A hubs, plugins for answer engines, the 👍🏽 / 👎🏽 button. These systems are becoming self learning and will need to get into the habit of providing feedback to every query. Always appreciate your thoughts leadership and passion. I always learn a lot!
Thanks, really appreciate it! This is a fast-evolving space, and also complicated by the fact that sometimes AI models are replying based on their own internal networks only, and sometimes from search results provided to them—which may be returned by a different search model entirely. It all gets pretty complicated, but it's clear that it's different than things like PR and SEO.
Agreed. I am hearing AEO/GEO as the new incarnation (Answer Engine or Generative Engine)