HN user
I would love to be able to do the clustering from a CSV instead of a collection of Markdown files. I know I can easily generate the files, but I used to do this directly for very short text inputs (just titles or words) on nomic.ai (before they pivoted to 'Enterprise')
Feb 12, 2025 [1]
Grist [1] desktop [2] is a (kind of) local "wrapper app" over an sqlite database. It is "As extendible as it is flexible" [3]
The problem is I feel the tabular DB format is "too raw" for some usecases. I prefer Outliner/DB combos (like the upcoming logseq DB version) or maybe one of the local Notion alternatives.
[1] https://github.com/gristlabs/grist-core [2] https://github.com/gristlabs/grist-desktop [3] https://www.getgrist.com/product/#:~:text=As%20extendible%20...
Would it be possible to import titles as a text list, instead of requiring files?
I'm not a hoarder, but I do have a (very) long Excel file of movies and documentaries that I want to watch or have already watched. Most of them are available on streaming sites or for rent/download on demand.
You've got a great IMDb scraper and filtering UI. That's all I really need! :)
With Suno 3 (and possibly Stable Audio 2), we are definitely experiencing a moment for music that parallels what we saw with ChatGPT 3.5 and Dall-E 2. The unexpected jump in quality (like sora recently) is again surprising (to me).
As with these 3 past breakthroughs, Google seems to be (slightly) lagging and again misses the "defining moment". I'm willing to say that OpenAI could have achieved something similar (and claim a 4th turning point), but the headache of copyright issues likely pushed them to focus on Sora first. I'm almost certain Udio and Suno were trained on (a lot) of copyrighted music.
So, an LLM scoring over 80 on MMLU is no longer newsworthy on Hacker News? We've indeed come a long way.
Or perhaps it's just the skepticism surrounding such "claims". That's why I reckon more chatbots that don't rely on an API, similar to Inflection (or Microsoft's Copilot ?), should be directly integrated into the LMSYS Leaderboard. Isn't this what they did with Gemini/Bard?
I keep hearing of AI photonic chips and how energy efficient (and faster) they are. This could be THE solution ? How far away are we?
In other words, you can drag and drop downloaded states into your model, like literal plug-in cartridges
The same could be said of "control vectors" [1]. Both ideas are still experimental, but is seems to me IINM that they could replace "system prompts" and "RAG" respectively.
I'm patiently waiting for the planned database tables [1] (or the similar plugins to mature). However, even if this is implemented correctly, it is still going to be on a page, not block level (query/attribute/two-way links...). That is why I will probably stick with outliners.
Or, I may give Siyuan [2] a second chance. I'm actually referring to this now because they just released 3.0 with those local notion-style DB tables. Think Anytype but with an Obsidian feel, and it works on the block level!
I always wanted to sort/filter goodreads like this. I don't know how you scraped the site, but the Kaggle datasets I've seen are always deficient in one way or another: either too small (less than 10k), or books are scraped randomly (not top), or (at best) they use Best Books Ever lists (skips many top-rated books). I hope your source data is better.
Thanks !
It is getting somewhat better recently, but a phone like the Samsung A15 is still an exception : for under $170, you get 4+1 years of support, which is good enough I guess (less than $3/month).
What I would improve is the discoverability of improvements that already exist, if that makes sense : So many scripts, extensions, and UIs are out there. Maybe someday I will organize it together and feed it as RAG to a custom GPT.
And speaking of GPT, I just used it to create (in less than 1 minute) a userscript that does your request.
I think the first is no-code while the two others are more like low-code (pipedream is simpler and free offer may be enough for you)
It was a pain > all worth the hassle for the teenage me
My perspective is this: If I accepted (like the OP) how "difficult" it was to set up a system for discovering music before, why wouldn't I accept a little "hassle" with Spotify? because I'm paying? No, personally, I'm paying for the music DB, not the app. dashboard? Ignore. Official playlists? Ignore. How difficult is that? It takes way less effort to get around Spotify "pain points" than any prior system, and none of these offer what Spotify does best: enriched API access to "unlimited" music.
But of course, we do need to complain to make things better (especially for the artists).
Hmm... I've always wanted something like this. Good execution. Good use of (Andy Matuschak?) Sliding Panes.
However, my opinion is that this is going to work well only in very specific situations: small teams and discussions with a high ratio of long, analytical (as opposed to synthetical) comments over multiple levels (and not just the first).
It's just too advanced for most discussions. HN/Twitter are n-level (and a thread of tweets is a kind of pre-quotes). But Most Messaging/forums don't even have a third level. And a lot of people seem to find a one-level UI like Discord and IRC enough.
Anyway, thanks for making this open source !
Most data accumulates gradually (e.g., one email at a time, one line of text at a time across various documents). Is this huge 10M scale of context window relevant to a gradual, yet constant, influx of data (like a prompt over a whole google workspace) ?
Some suggestions :
- Create multiple independent "vaults" (like obsidian).
- Append links to related notes, so you can use (Obsidian's) graph view to map the AI connections.
- "Minimize" the UI to just the chat window.
- Read other formats (mainly pdfs).
- Integrate with browser history/bookmarks (maybe just a script to manually import them as markdown ?)
Thanks for Reor !
Outline [1] is not OSS (its source is just available).
Docusaurus [2] is not a "wiki" (it generates static sites from Markdown).
But both are popular and self hostable.
It would be possible to plug an LLM [1] into the stream history and be able to directly type the ideas you cited and get instant results.
If you had this data at scale
That was the "raison d'etre" of last.fm. Alas, it is not popular anymore (=smaller scale)
favourite songs from people who enjoy similar niches to you
In last.fm, you can go to an artist's listener page [2], pick a user who "listens to Televators a lot", export their most-listened/loved tracks to Spotify, filter them by genre, and try them out :). You could also go to a track page [3], pick a user who commented on it, and do the same.
[1] https://news.ycombinator.com/item?id=39261486
Now lastfm has shut down their free API access
I think they just sometimes disable new API accounts creation. It is open now anyway, and my old API key is still working.
Well they speak of 236 million subscribers [1]. are they lying ?
The article mentions that the project has over 600K monthly users, which is a significant number for a side project. However, in comparison to Spotify's user base of 1000 times more, it now seems insignificant. This reminds me of how companies like Google only consider services with hundreds of millions users to be successful or important. It appears that Spotify is in a similar position now. My only hope is that the API does not degrade significantly.
There was this page [1] (oct 2022) on hntrending.com. The site is offline now (since ?)
[1] https://web.archive.org/web/20221001024554/https://hntrendin...
AI/LLM winter sets in; GPT-6.5 is no AGI, but the wide-ranging societal effects of AI systems are visible at all levels.
VR/AR Spring finally arrives; Apple Vision 6 SE is a satisfying mass product for under $600.