Hello! I'll take the algorithm opportunity to introduce myself to my 3 followers (you're the best!). I'm @magicturtle42. My focus right now is on model fine tuning for cancer research, systems infrastructure, and designing a system for novel biological threat detection.
I've been interested in agents since one deleted a toy db in a side project in early 2025. I've learned a significant amount since then, but am constantly trying to learn more so I can apply them more effectively.
I'm broadly AI optimistic, but also have a number of specific concerns.
I'm interested in theory of mind, human computer interactions, reading, and educational efficacy.
Feel free to introduce yourself below!
My frontier lab researcher neighbor told me chinese labs keep distilling his frontier models so I asked how many models he has and he said he just raises a 100B and trains a new model afterwards so I said it sounds like heโs just feeding intelligence to chinese labs and then his daughter started crying.
the consistent ROI of crushing everyone in sports betting all the time presumably beats improving some rare disease outcomes somewhat, but that's a rather bleak feature.
Distribution is helpful in the sense that it lets lots of people apply it to their personal problems, i.e. DAU is more representative of positive value if (if) what's being derived from the usage is meaningful utility rather than engagement slop. But pursuing singular metrics in general is just begging to get Goodhearted.
Obviously if you're treating it as product level specification you will likely be in for a Bad Time.
But if there is a better way to work at a feature or fix level than specifying what you would like with clarity then I'd love to hear about it.
I would not take the idea of spec driven development to mean that you do not iterate on the spec/output.
the problem with "spec driven development" is that most software can't be spec'd up front
software is a creative act, where you figure out what you're building as you build it
you need to get your hands dirty in the details, and react to incremental versions
it's telling that all the examples of spec driven development are sorting a list or porting thoroughly tested code (like a js runtime or browser engine), which are the exception, not the rule. the vast majority of software doesn't have a spec โ or if it does, the spec was created *after*
Part of the skill of software seems to be enduring pain (this system/process/things sucks) while not becoming numb to it (I can make this better). Sensitivity tuning matters as well. You can't take on everything, but you can help the people around you a lot by fixing things they have just accepted.
I suppose this is just broadly true about life.
every data center story says it uses "as much power as 100,000 homes" like that's a scandal. an aluminum smelter pulls five times that and it's why airplanes are cheap. measuring industry in homes is how you train a country to believe building things is a crime