If you have tried using LLMs to generate and execute code, you might appreciate this workshop paper we presented at @ ICML workshop on LLMs and Cognition. The primary contributions are 1) a Case-Based Reasoning approach to reducing LLM failures via dynamic, few shot prompting and 2) seven failure types that can cause generated code to fail. These failure types are more detailed than most benchmarks that evaluate LLM code generation; and since we didn't have an automated way to check for all of them, we performed the evaluation by hand 😅 Paper: https://openreview.net/pdf/f2d10bfca1b7d9f6f0a87144fee8e775cba6701a.pdf

DVM ecosystem just got an upgrade with proofs for DVM output; also nostr_sdk rocks View quoted note →

Are KINDs distinguished by their structure? If I turned an event for an arbitrary KIND into a graph where the graph structure matches the JSON structure, and the data types of nodes were colors, could I get a unique graph for each KIND? What if there was a tool where people can see if there is already a KIND that exists for them or if they need to create a new one. The tool would just pull as many different kinds as possible from relays, and you could take an example event and compare it against all existing ones based on structure and data types of nodes. It could remove the need for any single global list of KIND events, like the Nostr NIPS repo. I'm particularly interested as we have many more DVMs come into play and when someone wants to make a new DVM, they will need to decide whether to use an existing KIND or create a new one.

GM! Excited to see @Not Paul Graham on Nostr, welcome!

Some pretty graphs appeared while working on #nostr this morning:

Just like bitcoin had decades of academic publications behind it, so too does decentralized AI. Check out this 1990 book on Decentralized AI! No mention of blockchain 😂 https://shop.elsevier.com/books/decentralized-ai/demazeau/978-0-444-88705-4 Unfortunately it's behind a paywall...

Thank you @Don't Believe The Vibe 🌱🍋🍊 for making the default behavior on nostrdvm to use whatever relays the original request puts in the tags. 🍻

Encrypted DVMs, as they are currently implemented, break one of the best features of the DVM process. What feature is that? Using kind numbers when asking for work rather than asking DVMs directly. I’m not sure what the solution is - perhaps the user requests the type of work to be done (which kind number and perhaps metadata such as size of data they have), then DVMs can respond as a sign they are interested, then the encrypted handshake can occur after the user chooses which DVM to use. It will be tempting to just select a DVM before making a request, and maybe the loss of flexibility and price competition isn’t enough to deter the temptation.

Hello #nostr!!! I’m typing this to demo how posting works to someone who has just learned Nostr is

“Nostr may be one of those technologies that doesn’t work in theory but works in practice” - DHD

Spotify doesn’t work behind a vpn for me so I can’t use it in a coffee shop. So ready to use Wavlake as my main music source.

GM #nostr. Build the future

@Don't Believe The Vibe 🌱🍋🍊

GitHub

List bad DVM kinds for people to avoid · Issue #28 · nostr-protocol/data-vending-machines

Currently these kind numbers are being used for other things and it would be nice to let devs know to avoid them. Even though they are within the 5...

Interesting alternative to kinds - I think the main benefit here is that the replacement to the “kind” is now English readable. So instead of having to look up what the kind number is, the new “event path” literally says what it is. We could still have path term overload issues, although they might be less common. Unfortunately it will still require coordination amongst devs to ensure we don’t have a billion top level folders… which would end up replacing kinds with single words. View quoted note →

Wow perfect timing, I was about to try this yesterday and found there wasn't an async client in nostr_sdk View quoted note →

Almost got duped by this today on primal, then checked on Nostur which makes it apparent:

What interesting computer science problems arise in Nostr development? I'm thinking of putting together a talk on this for CS folks that aren't familiar with Nostr. Here's what I have so far, and please suggest new items or revisions to current ones: - public/private key encryption schemes between individuals, groups, groups with moderators, temporary delegations of key signing (#nsecbunker) - graph analysis and scoring to solve problems like web of trust (#WoT) - distributed file storage and sharing, like what #blossom is trying to do I know there's got to be a lot more, 🙏 in advance

Curious why WoT scores keep changing on Coracle? Everything was between 0-35 when I first tried it and now it’s more like between 70-135. Browsing global with WoT scores is something I can’t go back from.

GN

Nostr is so freaking cool. @StarBuilder created an event for our meetup next Tuesday, then I RSVP'ed for it on Flockstr. Today I tried Coracle for the first time and after I logged in, the meetup shows up on my calendar without me having to do anything. This is the interoperability I came here for, LFG!!!!