| ▲ | Fluorescence 6 hours ago | |
It doesn't even need "elimination". There will be negative biasing from their user base, the profitable use cases and the company incentives. Those designing for human comprehension, simplicity and maintainability are less likely to be the ones dumping their code-base into a model. When actively hostile to AI, they will be less likely to even open-source their projects. Maybe there is a future for restricted openness with circles of trusted human collaborators. However, the users with piles of faulty dog-shit they don't understand or care to even try, will be the whales. The perfect customers, utterly dependent and burning tokens every day for trivia. How long until the companies intentionally farm such dependency e.g. gently obfuscating code output? There is a dystopic future for regulated full lock-in where only binaries are sent to the client not code, model-hopping is blocked with legally mandated decompilation guardrails and secure-compute hardware that refuses to run open models. | ||
| ▲ | Applejinx 5 hours ago | parent [-] | |
From what I've seen being adjacent to some of these folks as they try to parse what Claude etc. say about the code being AIed, this is already happening, and I wouldn't call it gently. There's a whole AI 'get really arcane and highfalutin' language of talking about its own code and I've occasionally called it gaslighting. I got a bit of pushback on the grounds that there is nobody actually there to gaslight you, but all the same I don't think this is subtle. | ||