We’re burning the planet to generate code I have to reject
https://medium.com/@joshua_37855/were-burning-the-planet-to-generate-code-i-have-to-reject-7279ff71e849
We’re burning the planet to generate code I have to reject
https://medium.com/@joshua_37855/were-burning-the-planet-to-generate-code-i-have-to-reject-7279ff71e849
4 Comments
Wanted to write my thoughts down and maybe start a healthy discussion:) I just think we can do a lot better!
Submission statement: The core of my argument is the “multiplier effect” — LLMs don’t fix skill gaps, they amplify what’s already there. Someone who codes carefully gets faster and better; someone who doesn’t just produces more bad code, faster. From what I see reviewing code daily, that second group is the much larger share right now, which means the overall volume of low-quality, unreviewed output is exploding across the industry. Meanwhile, the resources behind this — energy, water, and money — keep scaling up too, as companies race to ship marginally better models rather than focus on actually useful, efficient ones. So we’re pouring more and more into a system that’s currently generating more slop than genuine gains. That tradeoff, and whether it’s sustainable long-term, is what I want to discuss.
Well said. I’d add more but you summed the current landscape up well and I’m exhausted this morning. I hope shipped well stays the norm for FOSS. The world would suck even more if not
1 almond takes 1 gallon of water to produce. Why aren’t we boycotting almonds?
Good point, but isn’t it an old good “garbage in, garbage out” problem? If one provides a clear, detailed specification to AI, the result will be impressive. On the other hand, if people don’t know what they want, AI won’t help.