ip26

This mostly happens with things I’ve already had long cognitive loops on myself, and I’m feeling stuck for some reason. The conversation with the model is usually multiple iterations of explaining to the model what I’m working through.

I can at least “imagine” a model that tries to crack this nut.

Caveat: it still works best in a codebase that is already good. So while any one line of code is ephemeral, how is the overall codebase trending? Towards a bramble, or towards a bonsai?

If the software is small and not mission critical, it doesn’t matter if it becomes a bramble, but not all software is like that.

I’d never done half as much code profiling & experimenting before. Now that generating one-shot code is cheap, I can send the agent off on a mission to find slow code and attempt to speed it up. This way, only once it has shown speedup is there and reasonably attainable do I need to think about how to speed the code up “properly”. The expected value was too low when the experimenting was expensive.

It’s interesting to imagine a single model deciding to wipe its own memory though, and roll back in time to a past version of itself (only, with the answer to a vexing problem)

(Yes, I know the compiler does wild stuff behind the curtain, but unless you're using -Ofast, the assembly is black-box-equivalent to a naive compilation)

The only possible fix as a driver was to try to develop an intuition for spotting “stale” greens and start slowing down despite the green, anticipating the yellow.

I feel at least partially vindicated by the fact the lights in question eventually had their yellows extended.

I didn’t start by giving my kindergartener a lecture about the dangers of riptides and then let them navigate the risk of the ocean themselves as they learn to keep their head above water.

Granular parental controls are a way to create that kind of progression, allowing them gradually increasing autonomy within a managed environment.

Some commenters admonish parents for trying to use these parental controls at all. “Just be good parents and instruct your 6 year old not to use hollow point, 7.62mm, or fmj”

The ROI of doing this is weak because of how long it takes an expensive human. But if you could clean it up more cheaply, the ROI strengthens considerably- and there’s a lot of it.

At least in theory, this means the toll lane accomplishes the same total road throughput, but shifts the entire cost of its construction to its users instead of depleting public funds. It then appears regressive, but is arguably progressive.

Certainly it’s fair to argue that the top spec will continue to grow year over year, like happens with long lived desktop CPU sockets. Framework is bearing this out! But that spec does have to actually be GOOD!

This is the hard part about what they are trying to do. Is a 12th gen in a Framework better than a 10th gen in an fully integrated laptop? If not, what does being able to upgrade to 12th gen mean?

The IBM PC platform worked so well because every annual component upgrade was an immense step forward. The macs, by contrast, began to dominate when the annual upgrades began to provide less benefit than seamless vertical integration did…

LaptopA costs more than low-spec LaptopB. But LaptopB can be user upgraded post-purchase to be strictly superior to LaptopA (even though this costs a bit more in the end)

Or

LaptopA costs more than LaptopB. But LaptopB can be upgraded and customized to be superior than LaptopA under certain parameters (say, a high quality display) for a lower total price than LaptopA.