We gave Claude 25,000 customer feedback records as a CSV and via Enterpret's MCP server + Customer Context Graph.
Results:
💵 52% less tokens used.
🧠 4.5x tokens used on reasoning.
🏎️ 3x faster.
🥇 +48% better quality outputs
Cheaper. Smarter. Faster. Higher Quality.
@arscontexta@garrytan How are you thinking about building customer feedback into this? Currently thinking through reframing what we do as “customer context”.
@Kicom21@niccruzpatane It was actually fairly common for people to do stuff like this for crypto mining until miners got specialized hardware that a normal CPU/GPU couldn't compete with. For AI inference you need special hardware to do it efficiently and the cars would already have it for self driving.
@romangrachev@elonmusk Not sure if this still the case but last time I tried to run an ad campaign it kept telling me my audience was too small even when I was targeting 1M+ people.
You should let people target smaller audiences but make them pay a premium price to do so.
@Joeingram1 There is an interview with Sam Altman where he says younger people use chatgpt differently than Millennials. They don’t use it to do research and inform their decisions. They’re literally just asking it what to do. I don’t think he’s as far off as you think.
@ArthurMacwaters@therabbithole Not that they should be getting anything but I think this chart is potentially misleading. It’s saying that 59% of illegal immigrants are getting some form of welfare. That doesn’t mean 59% of welfare is spent on illegal immigrants. Would like to see that data myself.
@alexisohanian You can already see this effect in Chess. AIs have been godly for many years now but people don’t really watch AI chess matches.
I guess the only question mark is would people watch a fake version of live sports generated by AI.