Mastodawn

RE: https://mathstodon.xyz/@dougmerritt/116349163827428010

Dissent.

The author's thesis makes a fundamental assumption: Code is now cheap.

This assumption is false.

When machine code was supplanted by assembly code, code did not suddenly become cheap. When assembly code gave way to compiled languages (COBOL, FORTRAN, Algol, Pascal, C, etc.), code again did not become cheap.

If we assume LLMs are the next step in programming (and I don't), then we've simply moved another layer out -- another layer of abstraction. But nothing will get cheaper.

Show thread

ewhac 5d ago

Ever watch a newcomer to programming write code? Too often, they're programming by auto-correct. They type a plausible-sounding function name, look at the fuzzy-find hovertext that pops up, pick something plausible, and see if it works. Rinse, repeat. A valid approach for a beginner, but let us not pretend that, at least at this stage, the coder actually *understands* what they're doing.

They're lab rats, poking the button until treats come out.

Show thread

ewhac 5d ago

Thanks to the Anthropic code disclosure, we now know LLMs do exactly the same thing, only much faster.

This is not productivity. This is not engineering. This is, to coin a phrase, "Guessing at scale."

Show thread

Talin 5d ago

@ewhac Code is not an asset; code is a liability. Being able to spew out code faster means that you gain liabilities at an increased rate.

I recently heard someone refer to the process of writing code as "enbugging".

Show thread

Paul_IPv6

@talin @ewhac

i look at code as much like prescription drugs:

- the less you take, the better
- the more you take, the more likely there will be bad interactions
- the newer it is, the less tested and opearationally/medically understood it is
- taking it just because someone else does isn't healthy