Mitch Downey :pci:

Imagine listening to Joe Rogan Experience and using your own personal in-app llm integration to skip over all discussion of elk meat, feral boars, and brazilian jujitsu 🫠

I am behind on my podcasting 2.0 listening. Does the new MCP server open up the possibility for apps to add an AI option to their search pages?

Was wondering why Cursor Grok 4.7 kept refusing to work on my plan file, telling me to "try a less sensitive prompt."

Apparently, it's because my plan investigates "candidates" in my code that could explain a UI slowdown, and it doesn't like the phrase "Read and execute candidate..." 😂

Sneak peak video of the Podverse NextGen mobile player screen

The player screen should be for the most part the same as the previous gen, with the exception of these tabs at the bottom of the screen, and the ability to scroll down.

This is an experiment and not sure what % of people will like it. The upside of this is that the features that are available for the episode/music track will be instantly visible, instead of hidden behind a carousel in the middle of the player.

listening to last week's P2.0...the nextgen Podverse schema (and ogen) relies heavily on the Podcast Index IDs as basically the authority on feed uniqueness. It sounds like we may need/want to do a substantial rewrite to account for how Podcast Index API nextgen will handle ids?

I don't fully understand what we'll need to do differently yet. Just mentioning it for awareness.

cc

Podverse NextGen Update 🛠️

Been working on audio + video playback enhancements. Here's a demo of what we call the "alt enclosure" system in podcasting 2.0 RSS feeds.

Listeners can set if they want audio or video to load by default. In this demo the user has video on by default.

The "wave" icon indicates that an episode or music track has both an audio and video version.

and the Doerfels have an alt enclosure feed that really showcases what it can do

https://alpha.podcastdj.com/album/L93EkHGxGT

I notice in our PV alpha that we are getting 500 errors when we request "get a list of feeds that have updates over the last X amount of time".

We make a series of paginated requests until we get the full list. It seems like the process works fine for most of those requests, but eventually we get a 500 error response from one of the pages. Is this a known issue?

I think I'm just going to add some retry handling for when this happens and see if that resolves it.

Does anyone know of an RSS feed that has music videos in them?

one for the "what does LLM mean by that" list. Do I want LLM to "scrub" this directory?

"If you want .llm scrubbed in the metaboost monorepo too (not only metaboost.cc), say so and we can add that as an explicit todo; right now it’s scoped to metaboost.cc .llm/ plus “don’t contradict” guidance elsewhere."

have you setup an "entity expansion limit" for RSS parsing before?

LLM advises me to set an entity expansion limit to avoid "XML bombs." I guess it means a feed is trying to waste your CPU resources.

It tells me 5,000 - 10,000 characters is sensible, but I saw what appears to be a valid (I think?) Arabic language feed from PI that exceeds our current 50,000 limit.

Any thoughts? I'm leaning towards keeping 50,000 as a limit and worrying about this if people complain about a feed missing.

Podverse Alpha update 🛠️

EXPERIMENTAL: USE AT OWN RISK

Bitcoin Lightning V4V transactions are live in Podverse alpha!

It supports the traditional "node" send approach with metadata for sending messages, and it also supports lnaddress, but does not yet support sending messages with lnaddress yet (it will just send an amount for now).

I tried sending a 1234 sats boost to

and Podcasting 2.0 with mixed results...unsure if these failures are due to node issues, or buggy PV code.

Can anyone point me to an RSS feed that has a proper podcast:value lnaddress implementation?

Podverse Alpha Update 🛠️

NEW: Add By RSS feeds on PV web 🎉

"Add by RSS" are feeds that are NOT in the Podcast Index, and are not supposed to be.

For example, private, premium content feeds that you get a link to when supporting a creator on Patreon.

Podverse legacy supported these feeds in the mobile app, but it did not let you add content from those feeds to your queue or playlists.

Podverse nextgen does support Add by RSS queue and playlist functionality. (cont.)

https://alpha.podverse.fm/add-by-rss/add

This is a test thread for podcast:socialInteract tag testing.

(I need an activity pub post url for us with test RSS feeds.)

I think we discussed this before, but wanted to confirm.

The PI docs said a podcast:guid should be v5, but you also said that shouldn't be a strict requirement.

I'd like to implement at least some validation on that column in our database. Do you think v4 and v5 would be enough? Or just validate that it follows any of the uuid schemas? (apparently there are 8 versions)

Podverse Alpha update 🛠️

We are nearly ready to implement V4V in Podverse Alpha web...and I am completely out-of-date in terms of what the current practices are.

Can anyone direct me to resources I should follow for a web-based Bitcoin V4V implementation? Is the standard still keysend? Are there multiple types of implementations we should support by now?

#podverse #pv2

LLM programming lesson learned...I have been using Opus 4.5 for large, complex tasks where I want the highest quality model available for the work, but Sonnet 4.5 (also by Anthropic) is also very good, and costs ~67% less $ than Opus 4.5 to use.

My latest approach is, when working on a large and complex task, have Sonnet 4.5 plan all the work upfront, and split the work into many plan files, and save them locally (target under 300 lines per plan, to keep them reasonably sized), then...

As the Podverse infrastructure has grown to 10+ repos, and given how LLM seems to perform better with all its context in one place, and it would be easier to save llm prompt history in one place...I'm wondering if we should make a major switch to a monorepo 😣

I've never done a monorepo before. If it were just LLM, then we may not need it (we could create yet another repo...podverse-llm or something) (cont)

I’m finding trying to keep up with LLM advances for coding pretty disorienting…on one hand there are some real and magnificent breakthroughs, and on the other there is a lot of useless slop masquerading as breakthroughs.

Improved LLM models are already accelerating my productivity by several times, but I’m still seeing enough errors and anti-patterns, that I have a hard time believing the “you can do it all with prompts” and “you don’t need to review every line” camp isn’t being naive.

I don't know if there is anyone left who is skeptical...but the LLM updates over the past couple months for programming are a giant leap forward, and it seems like we've crossed an inflection point for the industry.

Using Opus 4.5, I was able to get what would have been 5 days of work for me finished in 1 day.

One stage of my work included writing 8 paragraphs of instructions for Opus, and it wrote hundreds of lines of code across 7 repos, and was ~99% accurate and finished after a few minutes

how can we avoid Cloudflare breaking our connection to PI API?

I was able to make 3 successful requests to PI API from Alpha server, and I didn't change or redeploy anything...and since then, my 4 attempts to make a request to PI API have failed. The error message is 429, and it is an HTML respnse that includes:

Access denied | api.podcastindex.org used Cloudflare to restrict access

cc

Podverse Alpha Web update 🛠️

Added the Artists and Artist page.

This is our first implementation of Podcasting 2.0 "publisher" feeds.

A "publisher" feed is an RSS feed that contains references to other RSS feeds.

In general, individual Music RSS feeds represent an Album or Track of music.

Publisher feeds allow artists to group all of their Album RSS feeds under their artist name.

https://alpha.podverse.fm/artists

https://alpha.podverse.fm/artist/FTTwdF9Ral

#podverse #pv2