▲ 665 ▼ As Ubuntu embraces AI, Debian discusses banning all AI-generated code (www.xda-developers.com) submitted 2 months ago by sanitation@lemmy.today to c/technology@lemmy.world 148 comments fedilink hide all child comments
[–] db2@lemmy.world 28 points 2 months ago (2 children) permalink fedilink source parent hideshow 2 child comments replies: [–] mushroommunk@lemmy.today 33 points 2 months ago* (1 child) Except with stack overflow you can use your experience and pick the best code and onboard the comments discussing various aspects of the different answers. The LLMs don't do that. They can just spit out whatever the fuck they want and there's no guidance or votes or comments. Just an averaging of everything. permalink fedilink source parent hideshow 1 child comment replies: [–] radioactivefunguy@piefed.ca 2 points 2 months ago I agree, human vetting from experience is essential. But with agenetic, tool calling LLMs, you can (and should IMO) require multiple solutions be explored and presented with the tradeoffs for each option clearly outlined, and wait for you guidance to proceed with the advisee approach. LLM can definitely be used to churn out garbage code based on the averaging of everything, but can also be a super useful tool, if used effectively, to quickly see the most common ways a problem is solved, and then design and iterate a solution that works for your project. Similar to how I used to use SO, but with reasoned reviews of all the posted solutions and the ability to prototype what might work best for what I need. permalink fedilink source parent
[–] mushroommunk@lemmy.today 33 points 2 months ago* (1 child) Except with stack overflow you can use your experience and pick the best code and onboard the comments discussing various aspects of the different answers. The LLMs don't do that. They can just spit out whatever the fuck they want and there's no guidance or votes or comments. Just an averaging of everything. permalink fedilink source parent hideshow 1 child comment replies: [–] radioactivefunguy@piefed.ca 2 points 2 months ago I agree, human vetting from experience is essential. But with agenetic, tool calling LLMs, you can (and should IMO) require multiple solutions be explored and presented with the tradeoffs for each option clearly outlined, and wait for you guidance to proceed with the advisee approach. LLM can definitely be used to churn out garbage code based on the averaging of everything, but can also be a super useful tool, if used effectively, to quickly see the most common ways a problem is solved, and then design and iterate a solution that works for your project. Similar to how I used to use SO, but with reasoned reviews of all the posted solutions and the ability to prototype what might work best for what I need. permalink fedilink source parent
[–] radioactivefunguy@piefed.ca 2 points 2 months ago I agree, human vetting from experience is essential. But with agenetic, tool calling LLMs, you can (and should IMO) require multiple solutions be explored and presented with the tradeoffs for each option clearly outlined, and wait for you guidance to proceed with the advisee approach. LLM can definitely be used to churn out garbage code based on the averaging of everything, but can also be a super useful tool, if used effectively, to quickly see the most common ways a problem is solved, and then design and iterate a solution that works for your project. Similar to how I used to use SO, but with reasoned reviews of all the posted solutions and the ability to prototype what might work best for what I need. permalink fedilink source parent
