✉️ Not subscribed yet? Subscribe to the Newsletter
Future of Coding Weekly 2023/07 Week 5
2023-07-30 23:55
🎵 Zing: Audio Integration Library 🧠 Hippocampus for LLMs 🏹 What are embeddings?
Our Work
🎵 Zing - A Simple Audio Integration Library via Chris Maughan
I recently made my audio components into a module; handles setting up audio, doing FFT, connecting to Ableton link, reading and playing midi. There's a video: github.com/rezonality/Zing
Devlog Together
📝 (Experimental) Converting Point-Form Notes into Text Summaries via Paul Tarvydas
Experiment: Converting Point-Form Notes from JSON Into Text Summaries using AI
JSON point-form notes to AI summary I created a page of "cards" in a Kinopio webpage (called a "space"). I exported this page to JSON. I uploaded the JSON to my GitHub Pages blog. I grabbed the n…
I’ve been re-building Subset.so to make it way easier to reuse templates directly in the editor. The idea is that you should be able to build beautiful spreadsheets without starting from scratch. For example if you want to build dashboard with a P&L and inputs, you can find both templates and drop them into the same editor. The concept is pretty common in design editors like Canva where you can look up and drop in design elements, but haven’t seen it done with spreadsheets before. Here’s a 7 second demo
💬 Marcel Weiher
I implemented support for property paths in my native compiler.
So more expressive than something like sinatra, but running at cool 140K requests/s.
The dichotomy between expressiveness/comfort and performance is a false one, a bit like the one we used to have between user-friendly but crashy client-operating systems and solid but unfriendly server operating systems. Now my phone and my watch both run Unix.
Thinking Together
💬 Martin Shaw
Hi all, for those of us who have an interest in learning about how to build a more wonderful future and who wish to be inspired by those rare few actually building something different, I recommend that you visit and enjoy the Norman Foster exhibit at the Centre Pompidou which ends on the 7th of August.
Those who are unfamiliar may not see the relevance of a building architect's work to programmers and software / hardware designers, but I assure you that his long history of work is exceptionally relevant.
His designs are a deeply-considered human-scale 'high-tech' alternative to traditionalist buildings and to postmodern buildings. With tonnes of innovative features which had never been implemented before and which seemed obvious and common-sense after the fact.
This work sounds a lot like what we are working on in the software, hardware and otherwise industrial realms.
If you are in Paris or are able to get to Paris, entry is 17 euros.
I'll be going on the 7th of August if anyone is around for drinks, lunch and/or a chat 🙂
A major figure of world architecture who is often seen as a leader of the "high tech" trend, British architect Norman Foster has designed many iconic creations throughout the world.
🧠 Experience CozoDB: The Hybrid Relational-Graph-Vector Database - The Hippocampus for LLMs via Peter Saxton
I've found this article on using CozoDB (A relational/graph database) as the hypocampus for AI. It's an interesting read though it's probably best to leave it explained in their own words docs.cozodb.org/en/latest/releases/v0.6.html
What do people think, is this approach plausible? Will it make a notcable different to what AI's can do? The author has also written their own knowledge management tool so interesting all round.
Content
🏹 An interesting paper on embeddings via Eli Mellen
An interesting paper on embeddings; relatively accessible to a lay audience, but lengthy.
🧰 explainshell.com via Beni Cherniavsky-Paskin
explainshell.com and showthedocs.com are a thing of beauty. Why don't we have something like that integrated in our IDEs/REPLs/terminal?
Both are interactive — go on, take a non-trivial shell / SQL command from your history and plug it there.
I'm not sure I've seen that level of doc integration in any language env, yet an interpreter/compiler doing the actual parsing and name resolution should be well positioned to offer it, no?