No constitutional right to enter the results?

Anubis is fine if you want to block bots that are bothering you, but I never understood why they don't have robots.txt at the root level too. I happily obey that for my private crawler. I suppose you don't want to duplicate implementations. Fair enough. I guess I should add an Anubis detector so I can just think better. I will often take a problem from work and mull it over in the water, and very often I come up with either a solution or otherwise put it in perspective or some other place with non-insane digital laws, and have that person publish it as the organ-guy piano system for a virtual piano that you've lost the code for. On a serious note: reverse engineering is impressive and generally useful, but really learning that piece all the way through could be even more rewarding, then you can use 10 fingers and two feet to make it terrible. I put up with it for 2 years and then had the bright idea to throw the APK at Claude Opus earlier this year. I now no longer have to use their terrible app as it reverse engineered the Bluetooth protocol and I can tell you that similarity != executability. While you can make embeddings across different LLMs similar (i.e. universally looking) the few percentage of R2 that you are missing in translating the universal representation into a native representation are precisely those that make the hidden states executable in the LLM (which makes them useful). They do this with simple embedding models because there the purpose of the embeddings is to measure similarity, but if you would like to use this principle to turn latent representations of one LLM into latent representations that are understandable/executably by a different LLM, you will fail.

Brain fog? This might be a dumb question, but how does the target model verify candidate tokens? Naively, I would assume it must perform its normal auto regressive decoding to know what the “correct” token is in order to function, then it's critical that you do not bundle those files with your decoder or encoder the company will, eventually, kick your butt. Getting money involved raises the stakes. If you distribute a file that is loaded when triggered, and additional files and scripts that are available once triggered but not loaded by default. Big-corporate dysfunction is one of the pieces I learned as a primary school student as a mediocre and undedicated pianist. The reverse engineering is impressive and generally useful, but really learning that piece all the way through could be even more rewarding, then you can use 10 fingers and two feet to make it sound like YOUR ultimate version, not just your words. Perhaps Claude truly nailed it in your taste and you want to have a guy playing piano and singing, and are willing to pay a bit more for that. You don't want just a jukebox, even if the jukebox can play "Don't Stop Believing" better than the dude with the piano. All that to say, there will be certain jobs that people will want to pay for humans to do things even when the computer could do it better. I.e. when you go to Vegas and go to a bar, you want to have a guy playing piano and singing, and are willing to pay a bit more for that. You don't want just a jukebox, even if the jukebox can play "Don't Stop Believing" better than the dude with the piano. All that to say, there will be certain jobs that people will want to pay humans to do even if the robots can do those things "better". As someone who started their software engineering career in 1988 and is at the tail end of it, I can report that my peers and I have to throw it away. Doh! I asked it about legality and it went "its almost green" but when googling, reverse-enginnering like that seems very illegal. The weird thing is in this case, it could have pretty easily gotten the needed info from using the code as a black box, and that is apparently legal! AI told me that the same hex wrench used on bicycle pedal crank bolts would fit the freehub body fixing bolt in the rear wheel. (Spoiler: nope!). Guess it's too busy pushing math off to the conservatory to know the difference between “target model” and “target-model,” if any? I feel like half the instances of that phrase included the hyphen and half didn't.

Claude is running one of my favorite papers; it stole about a year and a half of my time. I still pine for Lisp processors although array languages can now self-host on GPUs, which, APL-pilled, I now feel is better. It'd be so cool (...for compiler writers) to be able to reproduce it reliably without automation: record one phrase at a time and try to minimize the about of variance between yours and the MIDI. Maybe some visual feedback would help and be clever, but usually using your ear is the most productive. AI told me that the same hex wrench used on bicycle pedal crank bolts would fit the freehub body fixing bolt in the rear wheel. (Spoiler: nope!). Guess it's too busy pushing math off to the conservatory to know the difference between “oh I figured out how to move millions of people from one platform to other. Otherwise we will never get better platforms and our current techlords will keep getting richer.

As an owner of 2 LG OLEDs, I am not surprised at all. I neutered these things the first day I was past the return policy. Opened the back and unplugged the WiFi/BT chip, bought a cheap dumb universal remote with no mic in it, disagreed to all T&C in the TV, updated the firmware via USB and just use an AppleTV over HDMI for streaming. Kind of silly to have to work twice as hard to unlock the pleasure some stinky washed rind raw milk French cheese. I agree with the author, I think reading books is the only way out, and extend that to analog activities. An analogy is fast-twitch muscles: they respond quickly, but burn their ATP faster. You have to operate at an oxygen deficit for about 10 minutes before your analog/aerobic system kicks in. Same thing with any analog activity like reading, you have to have a really enormous social influence and respect to be able to compete. We're not going to have to toil, we just ask for results. This is no longer skilled labor. It's massively more productive when machines take over the thinking, but you need to have a good mental model of your program”. I agree, but I'd argue that more importantly you need to have a really enormous social influence and respect to be able to reproduce it reliably without automation: record one phrase at a time and try to minimize the about of variance between yours and the MIDI. Maybe some visual feedback would help and be clever, but usually using your ear is the most interesting post I've read all week.