On the nature of consciousness
Hm... okay. Let's see. 500 words. Can't be that hard. I've done it before. I can do it again. Let's go.
AI! It's mighty important. totally underrated.
Monday morning... 8AM UK time is the foreverhaven deadline. I need to publish something. Crank something out. Produce, not consume. No AI slop!
Good that I set up that Bearblog. It's so neat. No one reads this anyway. Hm... I need to write something of timeless elegance. My opus magnum / magnum opus (either is akthually correct).
Maybe I'll turn on the coffee shop soundtrack on my cold turkey writer. Just to get some background sounds. It's pretty hard to be alone with your thoughts. What do I hear? Let me zero in mindfully. People talking, coins dropping, people walking. That's pretty much it. Oh just now somehow was chuckling. Is this even a real coffee shop? Probably. When I turn on the Coffee Shop soundtrack in my Cold Turkey Writer (you can pick between that and the Rain Soundtrack), it starts displaying 'Sounds by Coffitivity + Rainy Mood'. Coffitivity huh? It has its own logo and all. Was this a startup? Who came up with this? "You know how you can always really focus when you do work in a cafe? We should start a startup recording coffee shop sounds. Call it coffitivity'.
good this is inane, derivative drivel, Hauke. Why don't write something that really matters?
Let me hit that lost mary... maybe that'll get the old writing juices flowing. Hm... cherry ice is too sweet and so is triple mango. But it's what I have to work with.
Oh here we go, Stretchly ergonomics software is the only thing that can overwrite cold turkey superimposing itself onto my screen for my stretch break and breaking the monotony, telling me: 'Tilt your head to one side, then the other, then forward.' I do it dutifully, YOU SHOULD DO THAT TOO NOW; these stretch breaks are really good, they really break up my day nicely, reminding me to do some stretches, and be mindful and so on, I tried to add more prompts to it using claude code recently but somehow it seems to have overwritten those during the last update now the json seems to have reset, maybe I should write to the Ukrainian developer but he might have more pressing problems rn, maybe this app is on the way out with the new software that'll inevitably come out soon, like 'donethat' which is a new startup by this EA who I was once met, when he was running EA frankfurt and I flew me in for talk that I gave to the EA frankfurt group, but I screwed it up because I had to give it in german and my german was rusty and it was a stutterfest, how shameful, I shouldn't dwell so much on the past, this is getting more unhinged, you're not thomas mann or thomas bernard, not anymore at least, didn't they teach you how to sit down to write 5h in class essay in high-school, not sure if this is something I can ever reactivate.
Anyway, the thing about RSI is: coding superintelligence doesn't really help and it's all about algorithmic improvements on the transformer architecture and that's not helped by coding, but frontier math capabilities are probs a better proxy for that. but can you really improve based on the fundamental architecture based on first principles if we do not even have a theory of deep learning or do you have to brute force that with a lot of compute + genetic algorithms? Seems like it. I guess they're solving some erdos problems but then someone on x said they're not actually that hard. I wouldn't know. I guess it's anyone's guess really.
What if Dario is right?