blackle

I have a flipper zero which I use monthly to provision keyfobs for new members of the hackerspace I run. Great little device! This new one doesn't have an nfc/rfid reader/writer so the use-case is a bit befuddling. I'd love to hear how people would use this and how it might beat out, say, a used thinkpad.

There's lots of cool details like this which you share in this thread but aren't present in the writeup, which is a bit of a shame! I'd love to hear more about how you arrived at this sequence of bytes. Was it the same kind of process as Memories or did you have a different approach this time? How do you mean by "unfolded what's left?" Were you generating these experiments manually or automatically?

The same author wrote a long writeup a number of years back on their other demo "Memories." It is a great read! http://www.sizecoding.org/wiki/Memories

this is gorgeous! really illustrates the fundemental rules in a very aesthetically pleasing way

For AMD you can use the Radeon GPU Analyzer: https://gpuopen.com/rga/

Any percentage of people being de-personed is bad. If the state is permitted to withhold travel documents of people indefinitely (and the supporting documents they sent in to get their passport renewed[1]), do you really live in a free state?

Also, and I know people knee-jerk at the comparison, but historically speaking Jews comprised less than 1% of the population of Weimar Germany.[2] The smallness of the percentage shouldn't be cause to dismiss the harm of their discrimination as "no big deal." It's been shown where that leads.

[1] https://www.usatoday.com/story/travel/news/2025/01/28/state-... [2] https://germanhistorydocs.ghi-dc.org/pdf/eng/JEW_RELIGIONZUG...

Sure: https://github.com/blackle/GnuCashAutoInvoice/tree/main

It's kinda a mess tbh, and it actually also requires some non-exposed symbols to work properly, so it needs access to the GnuCash source code. I wouldn't recommend doing this unless you're ok with maintaining your own unsupported GnuCash feature.

I use GnuCash for the accounting of my hackerspace. It was either this or a site called "wave" which the treasurer of a nearby makerspace recommended. After signing up for wave and playing around, I still wasn't sure. A few weeks later I decided I would use wave, but then I found they had locked my account for no reason. GnuCash it is!

It's good software! I eventually wrote code that dynamically links with the libgnucash library so I can auto-generate monthly invoices for the member's dues.

I fully agree, however if you're the kind of person who orders groceries, then what ends up happening is that you start collecting reusable bags that you have no way of reusing. I hope that someone starts some kind of "reusable bag recycling" system so I can give the scores of bags I have to a good home

I've noticed that adding a colour can make it go from having a solution to not having a solution. Maybe it should try to come up with a solution using fewer colours in that case, since it's not obvious that manually removing a colour will lead to a solution.

Relevant xkcd: https://xkcd.com/2595/

Increasing temperature just makes more uncommon tokens more likely to appear. This can include both uncommon ideas and uncommon spelling and grammatical mistakes. It also won't make ideas that the LLM isn't capable of thinking appear, unless you crank the temperature way up and get lucky (like monkeys on typewriters lucky)

Godspeed, Charles Entertainment Cheese. We will meet again in Valhalla.

My understanding is that minimizing perplexity (what LLMs are generally optimized for) is equivalent to finding a good probably distribution over the next token.

There is a way to do this same compression by utilizing the raw probability distributions that the LLM produces as output. Fabrice Bellard has some experiments with doing this with transformers: https://bellard.org/nncp/

The idea is that if you can produce an accurate probably distribution over the next bit/byte/token, then you can compress things with an entropy compressor (huffman encoding, range encoding, asymmetric numeral systems, etc). This comment is too small of a space to explain fully how they work, but it suffices to say that pretty much every good compression algorithm models probability distributions in some way.

I love grunions. Such funny little creatures. I'd recommend finding a video of them in action where they bury themselves in the sand, only to wiggle around and "pop" out of the sand again.

I assume you'd go to the library and find photo catalogues, but this is inconvenient and before my time. Nobody is suggesting everyone use this blocklist, but it's here for the hobbyists who want to use it.

I see the utility of this as removing AI image results from google images. Many artists need to search for real life photo reference in order to improve their drawing or painting skill. If you've never heard "photo reference" before, you may assume this is like tracing, but that is not the case. What you do is "study" a reference then paint on your canvas separately. This is to improve your ability to render particular subjects and materials.

Although AI generated art is fairly adept at rendering materials at the small scale, they often have issues of large scale consistency (e.g. clothing wrinkles turning into seams in an unrealistic way) or being inaccurate (e.g. a certain type of wild cat having ear tufts when it shouldn't). These are real examples that I have seen.

Needing to sift through AI art in order to find true photographs for reference is frustrating. A blocklist helps with that.

Consider a situation where the human operators only get the problems the LLMs can't solve. If LLMs are good but still limited, then the leftover cases will be those that are unusual, difficult, or downright impossible. Think of things like complicated fraud resolution, or the customer needing to change a piece of data on their profile that the engineers never considered could be changed.

If the switch to LLMs is largely a cost-cutting measure for organizations, I could see that the human operators—though downsized—would continue to receive the same compensation as before. In short, they will be paid the same to do more and harder work. If their performance metrics are based on how quickly they can close a case, these cases will never receive the amount of effort they need to get properly resolved. That is bad for the customer, who can't get a strange but pressing problem solved, and it is bad for the employee, who has to work harder at the same rate as before. The only person who comes out ahead is the capital owners.

I've sat with help-line operators for a medium-sized consumer tech company. It seems like 80% of their time is spent troubleshooting very niche issues, with the simple ones sprinkled in for levity. People need wins in order to feel good about their jobs. If it's all difficult problems—at bad pay, then that's just torture.

A lot of (low level) code is written with the assumption that the clock rate is fixed. Sometimes bus protocols like i2c are implemented with "bitbanging" where the code "waits" for the right amount of time by doing a bunch of nops in order, assuming that each nop takes the same amount of time.

I've used blender VSE for all my video editing since 2020 and it's so awesome to see it improve. I already considered it to be the best video editor on linux due to its stability and featureset (you get to use all of blender's existing animation tools!)

This is what I do with Firefox developer edition. It doesn't have that problem where it refuses to work if it's been updated in the background, it just adds a little green dot to the hamburger menu that indicates that it wants to restart at your leisure.

It is a little more sophisticated. They say they use an exploit that was found where a URL with five characters with a dash will get autocompleted by YouTube (I wonder why that is.) That improves sampling by 32,000 times apparently

The sound effect of a Disco Elysium skill chimed in my head when I read this

I hope you're not talking about beanfield. if they go it would be very very sad

generates oxygen by splitting water vapor found in the body

This will also generate hydrogen. Is there any mention in the paper about what happens to the waste hydrogen? I can't get past the paywall and I'm curious

Thanks! A thread was just easier for me to reason about. If we wanted to do this all in one thread, we could have the main loop use a system where it keeps track of how often it needs to poll the input and how often it needs to generate mouse events, and waits for the soonest deadline for each.

Those who tried but ran out of disk space

author here: they are randomly generated packets, extended using the packet loss concealment feature. you can fill a packet with random data and it will still decompress into a sensible sound, and that's where all the sounds in the drum machine come from

author here: this is correct. the opus protocol is so efficient, you can fill a packet with random data and it will still decompress into a sensible sound. this is where all the sounds come from