mattvr

You could also work around this by adding a scaling transformation that normalizes and centers (e.g. sklearn StandardScaler) in between the raw embeddings — based on some example data points from your data set. Might introduce some bias, but I’ve found this helpful in some cases with off the shelf embeddings.

How about showing all the money you spent/earned in the previous day, in one personalized receipt?

Hey, thanks for the critique (this is my product)! Yeah, LLM summarization definitely has its challenges. We've actually spent months optimizing for this:

1. Using latest models (gpt-4o & claude-3.5) which show major improvements in quality

2. Extensive prompt engineering focused on distilling surprising insights vs. pure summarization

3. Structured output format that's easier to parse

Re: transparency - you're right, we need to add more details. I'm new to this and still learning how to communicate this.

Btw, it's free to use 1x/day, and here's an example: https://nuggetize.com/n/claude-shannon-wikipedia-47ca5e5b

If you're interested in this, you might like https://nuggetize.com/ which works without installing anything and with any browser.

Looks much better!

I could definitely see using this, but don't think it solves anything for me right now.

I think if the diagrams didn't require horizontal panning, and/or supported interactive collapsing, it would be more effective. As it stands, the text view on the left side of the screen is actually easier to parse at a glance than the tree.

Generally, vertical scrolling is more natural for reading this kind of data, and also works better on mobile.

So a first guess of an improvement is something like: the tree diagram generally grows flat & vertical, with horizontal panning (or a manual "expand" click) reserved for deeply nested objects.

The example on the site is a very simple JSON object that doesn't demonstrate JSONtr.ee's potential at all.

To showcase it better, I'd like to see more complex objects on the initial reveal, so I don't have to go and find & format a JSON object just to see if the app/library is worth using yet!

Children of Men

Love posthog in many ways, but confused and a non-starter that they don’t support any data deletion yet.

i.e., if you want user analytics to be auto-deleted after 90 days, there’s literally no way to do that yet.

Its underlying library, OpenRelativity, does.

Wait. This is amazing. I think there are plenty of potential users. I've already found a bunch of cool stuff near me, and could see using this for any new area I go to.

The list of locations could have their thumbnail image as a low hanging fruit improvement.

Another consideration is not throwing in the towel after posting once. I've had the same link get 3 upvotes on subreddit A, and 300 on smaller subreddit B. The HN traction goes to show – it's context/timing/audience dependent. Take more shots on goal.

Have been using this constantly while developing, noticing a huge productivity boost.

I'm sure there's 1001 tools like this in existence or coming soon, but I was surprised to find nothing that felt as natural as I'd hoped.

For example, this has some nifty features like I/O piping, smooth character-by-character output, and flexible configuration.

Example use cases:

• Debug crashes: 𝚌𝚊𝚝 𝚖𝚢-𝚌𝚛𝚊𝚜𝚑.𝚕𝚘𝚐 | 𝚐𝚙𝚝 -

• Quickly look up & run a command: 𝚐𝚙𝚝 -𝚡 "𝙳𝚘𝚠𝚗𝚕𝚘𝚊𝚍 𝚊𝚕𝚕 𝚒𝚖𝚊𝚐𝚎𝚜 𝚏𝚛𝚘𝚖 𝙳𝚊 𝚅𝚒𝚗𝚌𝚒'𝚜 𝚠𝚒𝚔𝚒𝚙𝚎𝚍𝚒𝚊 𝚙𝚊𝚐𝚎"

• Generate files, docs, tests: 𝚌𝚊𝚝 𝚒𝚗𝚍𝚎𝚡.𝚝𝚜 | 𝚐𝚙𝚝 - "𝚐𝚎𝚗𝚎𝚛𝚊𝚝𝚎 𝚊 𝚁𝙴𝙰𝙳𝙼𝙴.𝚖𝚍" > 𝚁𝙴𝙰𝙳𝙼𝙴.𝚖𝚍

All open-source, xplat, powered by GPT-4 (or 3.5), and Deno JS.