I primarily externalize my cognition by taking digital notes in Markdown, using Obsidian. I would also include digital timers (especially in personal devices like smartphones) and whiteboards or sticky notes. # The Extended Mind [[The Extended Mind (book)]] is a great place to start on extended cognition ![[The Extended Mind (book)#Parts]] The book came out at the end of 2021 and the *Building a Second Brain* book came out around six months later and is essentially part (4) above - "**thinking with apps**" (where someone else or most likely a profit-driven business decides *your* system constraints). ## Forecasting In 2022 I had an earlier incarnation of this garden, written prior to ChatGPT's release. In it, I wrote: > What comes after thinking with apps? I would say (5) is "thinking with code" and (6) would be "thinking with AI" \[...\] What I think is magical about thinking with code though, compared to (4) and (6), is that ==the process of writing the code involves *encoding* -- unambiguously -- your intention== for that code. # Part 5 - Thinking with Code ![[Thinking with Code]] # Brief thoughts on (6) I had previously written: > I think that (5) is kind of bleeding-edge right now, in the sense that it is time-consuming and I will likely explore many routes that are wastes of time (partly due to lack of previous exploration, partly due to my not knowing about prior work). I think it is worth it, and I'll have a detailed record of the entire process. > > That means that (6) is probably not within reach right now. Just like futuristic medicine, there are surely use-cases today that are legitimate under particular circumstances. My biggest concerns are the difficulty in tracing ML "reasoning" as well as [potential bias](https://www.theverge.com/2018/1/12/16882408/google-racist-gorillas-photo-recognition-algorithm-ai). I don't want to pay a corporation money to use their blackbox that is going to think for me. I need to be able to trust my system. As of late 2024-04, I still don't see anyone doing anything paradigm-breaking with generative AI although Llama 3 looks extremely promising. In my personal experience with GPT-4, it has gotten pretty consistently worse over time. I'm happy to be building [[My Tinker Cast]] instead of investing in LLMs, for now. That said, the actor model makes LLMs a pretty natural fit once the time is right. I need to learn more about the ethics around Llama 3 in particular.