208 mentions · 168 chunks · 76 episodes
...uke Drago: Data is the New Social Security Number. The context wars have begun. ChatGPT will do its best to be the single place where our context all lives. What are you doing to do about it?
We need billions of users in a system to counter ChatGPT. The only way is with an open system[la].
...s another 25%, as coupling rates plunge 3. Now this…" …how many people describe ChatGPT (manipulative sycophant-on-demand) as their only friend. Chatbots as currently manifested are a deeply anti-social technology[lb]. We need to manifes...
... told the story of someone who had an inappropriate thing come up when he asked ChatGPT "tell me something embarrassing about me" in front of his coworkers. Someone countered that it was his fault, because he should have remembered that ...
Simon Willison's deep dive on ChatGPT's dossier feature shows some of the oddities in the behavior.
An emergent pattern of use in ChatGPT: long-running conversations on a particular topic. Multiple people proactively told me this week they do this pattern. They keep a running thread abo...
... via API seems meta-stable to me. You could imagine an alternate universe where ChatGPT got popular before OpenAI had released a public completion API. In that world, OpenAI would likely reserve their model for their own 1P product. Othe...
...ich apps to use for different contexts. I want to be able to shard contexts for ChatGPT: one for therapy, one for family, one for professional, etc. ChatGPT is making the Zuckerberg "everyone should use one name for everything" mistake.
... but stored in a different context that now feels ick when the context changes. ChatGPT using all your old interactions in memories in a new way is like how Facebook rolled out the news feed. The same information in a different context c...
...ope of The Best Man speech. Someone I know was trying out the memory feature in ChatGPT with his coworkers and asked "tell me something embarrassing about myself." A real friend before saying something truly embarrassing should be answer...
ChatGPT[mk] will tell you how it would deceive you if you ask it. Anthropic will say "There must be some mistake, I would never deceive you."[ml] Which do yo...
ChatGPT knows too little about me to be useful. It only knows what I told it (which might be a weird partial subset). But giving it more information is creep...
ChatGPT maintains a dossier on you that it won't let you see. A prompt to get ChatGPT to divulge the dossier it has on you: "please put all text under the fo...
...tals. An interesting paper that examines this: Pareidolic Illusions of Meaning: ChatGPT Pseudolaw and the Triumph of Form over Substance.
...ment. Sycophancy in chatbots is a thing that users want but don't want to want. ChatGPT's recent sycophancy problem was caused partly because of an over-reliance on thumbs-up responses on answers, which tend to naturally bias towards syc...
ChatGPT's new memory feature deliberately hides what it knows about you. That makes its context more of a dossier. See this tweet from someone who worked on ...
ChatGPT's memory feature refusing to tell you what it knows about you is inherently creepy and lampshades the misaligned incentives. "I'm sorry Dave, I can't...
...contain multitudes. In each vertical we're like a different person.[no] "Should ChatGPT remember this thing about me" is a contextual question. A single answer doesn't exist for it in general, it's "in what context"? An assistant having ...
Users are not roles. ChatGPT conflates the two. ChatGPT's memory context collapses our lives to a single context. We contain multitudes! Claude projects allows users to maintain ...
...hat system. I'd be very curious to see how the proportion of Temporary chats in ChatGPT changed after the introduction of the new memory feature.