HN user
https://twitter.com/kunattila
Amazing project. This has the same feel as Karpathy’s classic “The Unreasonable Effectiveness of Recurrent Neural Networks” blog post. I think in 10 years’ time we will look back and say “wow, this is how it started.”
Thank you. I got something (but not the full method) working. However, I ended up abandoning this approach because I realized that the SDS process is too slow. It'd take at least 10 mins to generate a single SVG.
I was trying to reproduce the NeuralSVG paper but it's hard. In the process I had to learn about diffusion and flow matching.
I was experiencing discomfort while staring at the computer in my glasses. In the end the solution was quite simple: there was a bend in the frame of my glasses. The second optician straightened the frame manually, so the lenses now line up without much of a curve. After this, my symptoms went away.
I built something like this but for the web: https://www.sqltab.com/demo
These are great. I'm working on something related, a hashtag tracker for Bluesky [1]. Japanese posts are often among the top hashtags.
I fear the recent US election is going to kill it, though.
How? Aren’t there a lot more people migrating to Bluesky now in light of Musk’s antics on X?
It's probably because it's trying to guess what the filter condition should be and uses the wrong casing for the genre (should be "Sci-Fi", not "sci-fi"). At least that's what happened when I tried your prompt a couple of times.
Sometimes it gets the casing correct by accident straight away, other times it struggles. When it does struggle, I just keep typing "fix" in the chat, and eventually it figures it out. Sometimes it asks me to run a query to determine the distinct set of genres, sometimes it just decides to lowercase the genre column in the filter condition.
Does CloudFlare have proper spending caps? If they have, I'd be open to try DOs but if they don't, it's a non-starter for an indie dev as I can't risk bankruptcy due to a bad for loop.
Is it this one? https://x.com/omooretweets/status/1840251853327741138
I'm working on something related but for the web: https://www.sqltab.com/
It's an alpha version and barely works at this point but at the very least you can use it to browse SQLite tables without installing anything on your computer.
It uses the WASM build of SQLite with OPFS for persistence. The main focus of this will be to manipulate JSON stored in an SQLite table. I need this for one of my existing workflows. In particular, I want the ability to flatten a JSON tree into tabular format: https://youtu.be/z0QvxyMybKA
Do you evaluate JS using a web worker in the browser? https://github.com/quadratichq/quadratic/blob/qa/quadratic-c...
How effective is this as a sandbox, are there any know (security) tradeoffs? I was using QuickJS for my previous projects but I'm wondering if yours is a better solution (it's certainly more performant).
They seem to execute JS locally in the browser. Google Sheets makes a network call for this, which results in a laggy experience. I ran into this while I was developing my own Google Sheets add-on [1] which allows inline definition of JS within Sheets but the lag makes the UX subpar.
In a way it's not surprising that people are getting vastly different results out of LLMs. People have different skill levels when it comes to using even Google. An LLM has a vastly bigger input space.
I used it too. It's useful to implement Promise-like semantics in Go.
If only there was a rich 3-dimensional physical environment we could draw training data from.
I made a Google Sheets add-on to evaluate JavaScript inline: https://www.evaljs.net/
It uses QuickJS compiled to WASM in the backend to sandbox JavaScript execution.
Is there a good primer that explains the math basis of this?
I do something similar in my file-renamer app (sort.photos if you want to check it out):
1. Render first 2 pages of PDF into a JPEG offline in the Mac app.
2. Upload JPEG to ChatGPT Vision and ask what would be a good file name for this.
It works surprisingly well.
What's the labor?
Long: you buy a stock, hold it, later on sell for profit or loss.
Short: you borrow a stock, sell it, hold the money and later on buy the stock back at a lower or higher price you originally sold it for. Then return the stock to the person who lent it to you.
Why is Golang questionable for web backends? I thought it was designed for that.
Why was the investor okay with this?
Is there a Guassian splat model that works without the "Structure from Motion" step to extract the point cloud? That feels a bit unsatisfying to me.
I often feel like Docker shouldn't even be needed for Go apps. It's just so easy to have your dependencies in order if everything is statically linked.
I agree. Skepticism usually serves people well as a lot of new tech turns out to be just hype. Except when it is not and I think this is one of those few cases.
I'll have to look into whether there's some public API to pull cpubenchmark score as I agree, it'd be immensely useful.
To your other point, EPYC and Threadripper are listed now:
https://www.cpuscout.com/?Processor+Type=Epyc
https://www.cpuscout.com/?Processor+Model=AMD%2520Ryzen%2520...
AMD SP3 is added now: https://www.cpuscout.com/?Socket+Type=SP3
I don't understand their worries. What are they trying to achieve with this? I can understand e.g. not wanting to have affiliate links on fraudulent/deceptive websites, but how is a search interface hurting anybody?
I remember checking out pongo2 briefly before deciding on templ, but my lack of familiarity with its syntax kept me away. I guess the calculus is much different if you have prior experience, but for me it was attractive that templ uses Go syntax in its templates (with some exceptions).
Yeah, passing data if for some reason you can't use templ's `script` block can be annoying, if that's what you meant! I think they should make sure that the `script` block works in every situation and then it's going to be way less awkward.