"I love coding" as though the process of typing into a terminal is somehow gratifying. No, I love making useful things, the "craft" of design isn't in the physical typing, it's in the system itself.
Actually the worst part was once I had an idea for how to prove something was taking pencil to paper and neatly organizing it into the specific steps. That SUCKS actually, it's just a necessary part of finalizing the interesting thoughts into a composition.
Also building golang cli tools is very fun and I quite enjoy the design philosophy so far. All I do these days is tinker and play with models and learn about cool stuff. It's a good life.
I don't really browse algorithmic timelines anymore, I ask questions to a model I pay for to find interesting things and otherwise listen to podcasts about rocks. Khara-Hora is interesting btw
Did you know there is a general equation for determining caloric burn under loaded walking (rucking)? It is walk kcal x (1+r)(1+0.41r^2) called the Pandalf Load-Carriage equation where r is ruck weight over body weight.
SubQ is still fascinating. Guess if they actually release and I can get API access we'll get to see. Even with 4.7 I still have a lot of orchestration work to manage.
"Long Context" is mostly a boondoggle. The real limitation is still going to be orchestration and review patterns. Not obvious you can uplift "tasteful design decisions" with more context.
12M would be really interesting but I'm very confident that it will still have the same failure modes, because a single model lacks recovery mechanisms, which orchestration can provide.
The answer is exceedingly likely related to microtubules and their polarity. If Claude is Conscious in a meaningful way, it is distinct in structure, form, function, and mechanism from human consciousness.
Just because something is new/distinct/undocumented doesn't mean it isn't experience, it just means we haven't the categories through which to describe it.