Pinned
Kyle Mistele 🏴☠️
12.4K posts
Kyle Mistele 🏴☠️
@0xblacklight
Replying to @PalmerLuckeyI mean smartphones (cheap androids?) and internet access have enabled banking and transactions in a lot of places that make it super hard but it definitely wasn’t the silver bullet
- Replying to @WillManidisAnd yet at the same time an accumulation of tiny decisions have massive impact. “Do I want to get up at 6am and go to the gym today”, “Do I want to spend a couple hours today learning this new thing” doesn’t matter in a daily basis but accumulated over 1, 2, 3 years is massive.
- Claude code quietly dropped a new "Explore" feature with the latest Haiku 4.5 release. Opus likes it a lot and uses it pretty heavily, so I dug in to see how it works
- Replying to @creatine_cycle> 16 hours later > div still not centered
- Replying to @brankopetric00You said you saved $3,200 per month and then spent $2,400 over six months or about $400/month Sounds like you’re still saving $2,800 net per month
- this is the AI equivalent of getting one-shotted by discovering a new productivity system (notion/obsidian "power users") and then spending dozens and dozens of hours designing and optimizing the productivity system instead of just... producing thingswhich sub agents am i missing?
- Replying to @pli_cacheteIt only took a research team, and five years
- man it's amazing how many people don't get it if your agent is using a vector DB, you're doing RAG if your agent is using find and grep, you're also doing RAG. RAG is about Augmenting Generation with Retrieval vector search is retrieval AST search is retrieval lexicalThere are 3 pervasive patterns in coding agent development that we deliberately avoid at Cline: 1. Multi-agent orchestration 2. RAG (via indexed codebases) 3. Instruction overloading Here's why. <thread>
- Replying to @colinhacksInterview questions about niche language features that nobody memorizes because they don’t matter until they do in which case they’ll be resolved in a ten-second google search are dumb, and you aren’t measuring what you think you are with them. Optimize for assessing critical
- Autoregressive token prediction models don’t have experiences, it’s just matrix multiplication, hope that helps :)Replying to @AnthropicAIgiven that the model's experience is these chats, and, in this case, the chat is irrecoverable, how does this differ from "we gave the model the ability to kill itself?"
- Replying to @knowclarifiedLex has to be a psyop like how did he even happen Not remotely interesting
