• 2 Posts
  • 588 Comments
Joined 1 jaar geleden
cake
Cake day: 24 juni 2023

help-circle






  • Omg, this is amazing. Ok hear me out, you think bribes are expensive, but at local or even state levels they can be very reasonable, so here’s what we do is GoFundMe for legislators, where regular people bid money into escrow for a certain rep’s vote and if they do it, they get the cash. Of course people of wrong opinion can also bid so it’s not guaranteed to get you what you want, but you can collectively at least make it more expensive for them! Normal bribery requires all kinds of relationship development and professional lobbyists to make sure it’s works right, so it’s available only to the well funded few. Post-vote bribery is open to all and cuts the smarmy middlemen.

    I know you’re appalled, but consider how out of touch most reps are from their constituents. They are told all day long by these lobbyists that you don’t really hate genocide or that you really count on them to keep drilling for oil. Here you can tell your real values to their face with money, the same arbiter of truth those billionaires are using. Not for a campaign, but for a specific vote!












  • We automatically decompose a long generated answer into factoids. For each factoid, an LLM generates questions to which that factoid might have been the answer. The original LLM then samples M possible answers to these questions. Finally, we compute the semantic entropy over the answers to each specific question, including the original factoid. Confabulations are indicated by high average semantic entropy for questions associated with that factoid.

    It sounds like they verify one LLM’s answer by getting a second one to ask the same question over and over again in slightly different ways and see if it’s answers stay the same, interesting over each piece of the answer so that essentially the original prompt is exploded and then montecarloed.