3 ms·

"Would you on principle release unmaintainable code? If you do so because you're rushed" The term for this is "technical debt", you deliver when it works but h

by xlayn 10y ago

"Would you on principle release unmaintainable code? If you do so because you're rushed" The term for this is "technical debt", you deliver when it works but have a debt for fix until it's good enough. "would you want the same situation to be applied to say, genetic engineering?" World works on this premise, every drug for every ailment has a secondary effect.

bordercases 10y ago
But these are inefficiencies that you don't explicitly want; instead you manage them. In some systems these inefficiencies cost lives and instead of managing them you decide to prove them out of existence, or avoid ruinous failure altogether. Truth be told now that you bring in the drug metaphor I'm torn, because I've been recently convinced that there are opportunity costs and institutional inefficiencies that come from being too risk averse on the drug market. So I would believe that tolerating some error, i.e. some casualty or someone's death, could on average be OK if more people were saved. This is a part of what we mean when we say we're OK with secondary effects in drugs. Usually the prescription ensures that the worst won't happen. There might be differences to tease out between this example and genetic engineering, however. If a drug fails, you kill off a couple of people and make recalls. But these deaths happen more-or-less independently. The same goes with some software systems that tolerate error: so what if your shitty exercise app goes dark? Or, so what if a single consumer doesn't get served? You lose some money but life goes on. No one dies or no one gets cancer. The reason why dealing with genetics is more scary, and this applies to manufacturing retroviruses, is that the error can spread or be transferred more effectively than in the case of drugs, whose sale is tightly controlled and whose information doesn't propagate. A virus spreads or a seed scatters or a person breeds and you end up exploring new unexpected territory. This isn't bad per se. But it's harder to control. If I were to deal with that kind of thing I would attempt to see what was invariant with scale and see if I could test on that case, so that if a spread did occur, the "patterning of the network" would be sound (like designing a component of a fractal).

by xlayn 10y ago

bordercases 10y ago

But these are inefficiencies that you don't explicitly want; instead you manage them. In some systems these inefficiencies cost lives and instead of managing them you decide to prove them out of existence, or avoid ruinous failure altogether. Truth be told now that you bring in the drug metaphor I'm torn, because I've been recently convinced that there are opportunity costs and institutional inefficiencies that come from being too risk averse on the drug market. So I would believe that tolerating some error, i.e. some casualty or someone's death, could on average be OK if more people were saved. This is a part of what we mean when we say we're OK with secondary effects in drugs. Usually the prescription ensures that the worst won't happen. There might be differences to tease out between this example and genetic engineering, however. If a drug fails, you kill off a couple of people and make recalls. But these deaths happen more-or-less independently. The same goes with some software systems that tolerate error: so what if your shitty exercise app goes dark? Or, so what if a single consumer doesn't get served? You lose some money but life goes on. No one dies or no one gets cancer. The reason why dealing with genetics is more scary, and this applies to manufacturing retroviruses, is that the error can spread or be transferred more effectively than in the case of drugs, whose sale is tightly controlled and whose information doesn't propagate. A virus spreads or a seed scatters or a person breeds and you end up exploring new unexpected territory. This isn't bad per se. But it's harder to control. If I were to deal with that kind of thing I would attempt to see what was invariant with scale and see if I could test on that case, so that if a spread did occur, the "patterning of the network" would be sound (like designing a component of a fractal).