Hi all, wanted to share what we've been working on at Pyroscope.
This is a ChatGPT based Flamegraph explainer. Analyzing flamegraphs can be challenging and we often get questions from users about the best techniques for finding insights from flamegraphs.
We thought maybe we could teach an LLM do this task and turns out it ChatGPT does it pretty well.
You can check out a blog post [0] for a longer explanation for how it all works, or you can upload your own profiles and get insights quickly by going to flamegraph.com [1]
hehe, if anyone's interested, I built a similar machine around 2018-2020 [1], although I never automated the nail pick-and-place part of the process. I also used epoxy instead of plaster.
I would suspect you could make some reasonable side money with a large scale version of this, maybe with colored wire, especially with commercial art in mind.
I did consider it, and I think you're exactly right RE the direction I would go into (colored wire, larger pieces, etc). But then I had a bunch of other things happen in life and so I had to put it all on pause.
The world of art (and especially commercial / corporate art) is very foreign to me, but I bet exploring it would be a fun challenge, so I am planning to do that in a few years.
I work in automation. My livelihood is making things, unattended. I see my career.
Personally, I think artist should be compensated well. The whole "starving artist" thing isn't just a trope, and I don't think they should be punished like that, "in the name of art!". There are people in my family who were some of the most creative people I know, with art degrees, who have to spend time waiting tables, rather than on art, because because they have stomachs, and don't like the rain.
Money isn't bad, especially if it's from making beautiful things that others value.
My favorite little anecdote that I like to tell is that the first thing people often see when they add Pyroscope to their apps is that it takes way less CPU than other signals like tracing or logging. It's pretty common to see logging taking 5-10% of overall CPU utilization.
The other 90% is usually spent doing serialization / deserialization (half-joking).
I work on Pyroscope, which is a continuous profiling platform and so I see a lot of profiles from various organizations.
If you want to save the world some CPU cycles I would look into optimizing deserialization. And it’s not just JSON, binary formats like protobuf are not much better.
It comes down to the overhead associated with allocation and tracking (GC) of many many small objects which is unfortunately very common in modern systems.
Great article and all but there's no way I'm going to remember any of it. I think I'm going to stick to "click different blending modes until it sort of looks right".
Exactly, also I typically layer a ton of different colored gradients on top of each other while clicking through each blending mode for each layer.
There's no way I'm going to compute in my head "Well this gradient with three colors is going to look like this after I mix it on top of these 3 bottom gradients each with different colors, oh and an image underneath"
Clicking a button beats computing an image in my head in terms of mental energy cost.
Speaking of energy, a physical dial, or scroll control, would be nice to cycle through quickly.
Edit: I do really admire this author's technical deep dives though, very beautifully made
At least in some versions of Adobe software, on some platforms (was years since I used Adobe so I can't remember if it was on Windows or OS X), once you've selected one blending mode, the drop-down control is still in focus so you can use the scrollwheel on your mouse to cycle through them.
Mouse scrolling through the blending dropdown is still a thing in Photoshop CC. It sometimes catches me out when I want to mouse-zoom the window immediately after choosing a blending mode
This is a ChatGPT based Flamegraph explainer. Analyzing flamegraphs can be challenging and we often get questions from users about the best techniques for finding insights from flamegraphs.
We thought maybe we could teach an LLM do this task and turns out it ChatGPT does it pretty well.
You can check out a blog post [0] for a longer explanation for how it all works, or you can upload your own profiles and get insights quickly by going to flamegraph.com [1]
[0] https://pyroscope.io/blog/ai-powered-flamegraph-interpreter/
[1] flamegraph.com