Tool for Thought (2005) | Hacker News Reader

Tool for Thought (2005)

stevenberlinjohnson.com

55 pointswalterbell13 comments

Steven Johnson is one of my favorite people. I created a playlist of 150 of his videos from YouTube. https://youtube.com/playlist?list=PLZ4_Rj_Aw2YlwhpEHE4SRIbRD...

This itself was a challenge because he has such a common name. I had to do a lot of it by hand.

Also, I wonder if James Burke's Knowledge Web is relevant here. He is the creator of BBC Connections from the '70s and another one of my favorite people.

https://k-web.org

https://www.nytimes.com/2005/01/30/books/review/tool-for-tho...

> There's a fundamental difference between searching a universe of documents created by strangers and searching your own personal library. When you're freewheeling through ideas that you yourself have collated -- particularly when you'd long ago forgotten about them -- there's something about the experience that seems uncannily like freewheeling through the corridors of your own memory. It feels like thinking.

Has anyone applied a local-only LLM to their personal library?

I’ve not tried it but this Obsidian plugin does exactly this to your notes: https://github.com/exoascension/vault-chat/tree/main

I think it’s just a matter of time to scale it to something larger.

Thanks for sharing, but bear in mind that this plugin sends data to Open AI.

It certainly does — and I expect anything in this space that gets successfully commercialized will likewise siphon all of your personal data into a large corporation. But then I am a pessimist about these matters.

As a positive precedent, OpenAI's Whisper speech recognition is open-source and can be run locally with zero data exposure, https://www.newyorker.com/tech/annals-of-technology/whispers...

Some progress with local LLMs: https://arstechnica.com/information-technology/2023/03/you-c...

> Things are moving at lightning speed in AI Land. On Friday, a software developer named Georgi Gerganov created a tool called "llama.cpp" that can run Meta's new GPT-3-class AI large language model, LLaMA, locally on a Mac laptop. Soon thereafter, people worked out how to run LLaMA on Windows as well. Then someone showed it running on a Pixel 6 phone, and next came a Raspberry Pi (albeit running very slowly).

I think a lot of people are thinking about that right now. At least I read that idea a lot here on HN and on Twitter.

I personally cannot wait to try it. But for now, I wait until there are better options for local LLMs. Given the current development speed of all things AI, probably that won't take too long.

With the accelerating force of information generation, I keep thinking about how to cope with this much entropy coming from everywhere. I currently have a pretty strict "information diet" as no to drown in the sea of words and words and words.

I'm also currently working on a "Tool for Thought" but as a browser. Fully customizable(plugins, themes, functions, HTML as UI etc)

https://github.com/ilse-langnar/notebook

The Ghost Map, by the author, is one of the best popular science books I've ever read.

https://en.wikipedia.org/wiki/The_Ghost_Map

It was quite satisfying to see someone solve an epidemic, rather than make things worse than if we had done nothing at all.

> The software I use now is called DevonThink.. it is only available for Mac OS X.

A subset of functionality is available on iOS as "DevonThink To Go", with optional sync to the Mac desktop version via WebDAV or cloud services, https://devontechnologies.com/apps/devonthinktogo

I completely forgot about Google Desktop. RIP

If DevonThink find a smart way to integrate an LLM it would be such a great level up.

Apple Silicon includes an under-utilized Neural Engine, hopefully we'll learn more about their local-first, privacy oriented AI plans at WWDC, https://machinelearning.apple.com/research/neural-engine-tra...

Reply on news.ycombinator.com