I also attempted to do real-time audio visualizations with LED strips. What was unsatisfying is that the net effect always seemed to be: the thing would light up with heavy beats and general volume. But otherwise the visual didn't FEEL like the music. This is the same issue I always had with the Winamp visualizations back in the day.
To solve this I tried pre-processing the audio, which only works with recordings obviously. I extract the beats and the chords (using Chordify). I made a basic animation and pulsed the lights to the beat, and mapped the chords to different color palettes.
Some friends and I rushed it to put it together as a Burning Man art project and it wasn't perfect, but by the time we launched it felt a lot closer to what I'd imagined. Here's a grainy video of it working at Burning Man: https://www.youtube.com/watch?v=sXVZhv_Xi0I
It works pretty well with most songs that you pick. Just saying there's another way to go somewhere between (1) fully reactive to live audio, and (2) hand designed animations.
I don't think there's an easy bridge to make it work with live audio though unfortunately.
Agreed. As engineers we build context every time we interact with the codebase. LLMs don't do that.
A good senior engineer has a ton in their head after 6+ months in a codebase. You can spend a lot of time trying to equip Claude Code with the equivalent in the form of CLAUDE.MD, references to docs, etc., but it's a lot of work, and it's not clear that the agents even use it well (yet).
Another factor with the vibes being off (at least in the US): mass outsourcing of jobs thanks to remote work. You used to have to be a multinational company with global entities and offices. Now you can be a 10-person startup with half your people outside the country.
When the world went remote many folks were happy with the better work-life balance. But it means that we compete in a ruthless global labor market.
That's why companies rejecting remote work is good for the American worker in some ways.
I believe an important reason for why there are no LLM breakthroughs is that humans make progress in their thinking through experimentation, i.e. collecting targeted data, which requires exerting agency on the real world. This isn't just observation, it's the creation of data not already in the training set.
Data labeling has been moving to onshore / higher paid work. There's still a lot offshore, but for LLMs in particular and various specialized models, there's a massive trend toward hiring highly educated, highly paid specialists in the US.
But as other commenters have warned: beware of labor laws, especially in CA/NY/MA.
I've had a front-row seat to this...our company hires + employs contract W2 and 1099 workers for the tech industry. Two years ago we started to get a ton of demand from data labeling companies and more recently foundation model cos who are doing DIY data labeling. Companies are converting 1099 workforces to W2 to avoid misclassification. Or they're trying to button up their use of 1099 to avoid being offside.
I do agree. But the thing I would add though is that at least when I was there (W2012 - we were in the same batch!) YC and PG in particular were vocal about principles they believed in, especially the idea that naive young founders get kicked around in Silicon Valley and ought to be treated fairly. PG talked about this constantly and of course it’s a big theme of his essays. I remember one particular instance where a potential investor was behaving questionably and PG offered to step in and talk to them if they crossed a certain line (which they did not in the end). He was clear about what he concerned acceptable and unacceptable and why. And we were nowhere near the cool end of the batch.
Yes YC acts in its interests but in my experience they live by good principles and that makes all the difference. The Chaos Monkeys anecdote is an example of that. So I don’t agree with the article’s framing that they throw their weight around to simply exercise whatever power they have over others for financial gain.
Maybe this is a dumb question but I thought valuations had fallen a bunch since mid 2022 and now lots of VC firms are struggling with companies (especially mid-to-late stage) who raised at much higher valuations than they can now get in the market. But this firm is saying that current valuations are too high to make new investments. Would it not be a good time to invest in later stage startups? Or is the issue that the forward growth potential of these companies is lower now for some reason.
I was part of the W12 batch of YC (which was a lot smaller, ~60 companies, back when Paul Graham was still leading YC).
YC funds a lot of companies and has always had super high variance in the companies it funds. Entrepreneurs are a wild bunch of people. There have always been companies where the founders turned out to be BS artists or sociopaths. Companies that folded immediately after the program started. Companies with messy cofounder breakups already brewing at the beginning of the batch. Companies that turned out to be slightly scammy. Some of the founders that were in those companies pivoted and became successful.
Picking on Pear AI (which I don't know anything about) as evidence of YC failing is silly. It's also a super early stage company and you really have no idea what they will do.
The test of YC to me is, can they keep attracting and picking some of the best founders (which you can't really tell for years). And providing the inspiring, warm, but pushy environment that best sets up founders for success, and in turn keeps them coming to YC. I'd apply to YC again in a heartbeat if I were ever starting another company.
It takes time to build a company to significant revenue. I'd be curious to rule that out as the primary explanation before reading too much else into this.
This is great. This is mentioned in the discussion topics section ("modeling randomness in gathering belongings"). But I wonder what the distribution of baggage-gathering time looks like? I'd guess pretty positively skewed, i.e. most people are pretty quick but then a few people take a long time or a really long time. The speed of the flush stage of a particular wave is capped at the slowest person in that wave.
I do think there will/should be a reckoning about the how training data is acquired and attributed. For example, LLMs could attempt to cite sources, or share ad revenue fractionally with all the sources of that inform the response they're presenting.
I think that as the magic wears off it's becoming clearer that LLMs are more like fancy search engine UIs than intelligent agents. They surface, remix, and mash up content that everyone else created, without the permission of the creators.
That doesn't mean there won't be economic fallout. Spotify may have figured out legal streaming - but the music industry is still much smaller than it was in the 90s
Second this. The 32x64 version is what I used to make this digital tide clock as a project with my nephew https://www.filepicker.io/api/file/UzFNawTTWq4h5h6YF9Im. It fetches tide data from NOAA to tell you next high and low tides.
We also made an arcade game with the LED screen.
Not the cheapest route but it’s relatively easy and fun to get it working
By the way, the best New York secret is "Fish Friday", when the Acme factory in Williamsburg sells its stuff directly to the public for pickup, including various experimental things they've come up with. It's easily the freshest, best-tasting smoked salmon I've ever had.
This guy has a great set of examples of rules in English that started as bastardizations, fabrications, arbitrary preferences of some particular writer, etc.: https://youtu.be/JTslqcXsFd4?t=444.
To solve this I tried pre-processing the audio, which only works with recordings obviously. I extract the beats and the chords (using Chordify). I made a basic animation and pulsed the lights to the beat, and mapped the chords to different color palettes.
Some friends and I rushed it to put it together as a Burning Man art project and it wasn't perfect, but by the time we launched it felt a lot closer to what I'd imagined. Here's a grainy video of it working at Burning Man: https://www.youtube.com/watch?v=sXVZhv_Xi0I
It works pretty well with most songs that you pick. Just saying there's another way to go somewhere between (1) fully reactive to live audio, and (2) hand designed animations.
I don't think there's an easy bridge to make it work with live audio though unfortunately.