I think it’s great that Plex is attempting to use AI to give us a better music listening experience via Sonic Sage, but I feel like it completely falls flat compared to entering the exact same query into ChatGPT itself at chat.openai.com
I won’t go through an exhaustive comparison, because I think it’s obvious if you try a query like “Make a 100 song playlist that’s a mix of 80s synthpop, 90s college radio, and 2010s alternative dance. Include about equal parts Top 40, b-sides, deep tracks, and fan favorites.”
The response between Sonic Sage and ChatGPT is like night and day. ChatGPT spits out exactly 100 songs that perfectly fit what I’m looking for, while Sonic Sage gives up after 10 songs and includes mostly super popular songs, and not the mix of Top 40/b-sides/deep tracks/fan favorites that I asked for. Additionally, Sonic Sage isn’t interactive, so I can’t say something like “That looks good, but there are too many popular tracks. Can you include more obscure tracks?” which ChatGPT handles easily.
So what’s the end goal of Sonic Sage? What does Plex hope it will do? Why is it so limited and feels broken? I don’t understand how it’s supposedly using ChatGPT on the backend but generates such subpar results.
And YES I DO have Tidal AND I have 80,000+ songs on my local library, so there should be no reason that Sonic Sage is crapping out after giving me 10 tracks. I have thousands that fit the criteria that I asked for. For the playlist that ChatGPT generated, only about 15% of the songs weren’t in my local library and I had to grab the Tidal version.
One difference is that ChatGPT is just spitting out titles of things, Plexamp has to match them into existing tracks in your library or TIDAL. Plexamp logs would explain what it’s doing.
It used to work to request more tracks than the default, but for some reason (maybe a ChatGPT upgrade) the prompt isn’t working anymore with that additional suggestion, we’ll look into it.
While you are here @elan it would be great if there was a way to try this feature even on a limited basis without signing up for OpenAI.
The pricing information on Plex’s Sonic Sage page is also incomplete. It says you get $5 in free credits for your first 3 months, but how much does a Sonic Sage query actually cost? I have no idea if it’s a penny or a dollar.
I can generate a list of songs using ChatGPT directly, cut and paste the list into my playlists.txt file, run the script, and have a playlist appear in my Plex account faster than Sonic Sage. The bonus with ChatGPT directly is that it’s interactive, so if you don’t like what it generates you can tell it something like “That looks good, but there are too many popular tracks. Can you add more b-sides and deep tracks?” And it does exactly what you ask.
I tried Sonic Sage from the very beginning, with paid ChatGPT 4 integration, and I could never get it to generate anything nearly as interesting and on-point as using ChatGPT directly. I don’t doubt that as @elan says there may be some current issues messing Sonic Sage up, but I honestly never had a great experience with it at any point in the past 5 months – for me, it always skewed heavily towards the most popular and predictable tracks for every artist that it spit out, even if I said “Don’t include popular songs” or “Include only b-sides and deep tracks”. I don’t know why that is, since I never had the same issue with ChatGPT directly.
