Post B7VwaiJ5mCCaaQnnRQ by nicole4fox@datastream.cortexvoid.net
 (DIR) More posts by nicole4fox@datastream.cortexvoid.net
 (DIR) Post #B7VwaiJ5mCCaaQnnRQ by nicole4fox@datastream.cortexvoid.net
       0 likes, 0 repeats
       
       Tbh I dont give a fuck if AI was used during the development of software XYZ, I will still use / contribute to it.AI is just another tool in a long chain of tools, like compilers, autocomplete, frameworks, high-level programming languages, ... and the result matters more to me than the tools that were used in the process. Is the program secure, correct, usable, etc.? Good reason to use it. If it sucks by the common metrics, I dont care if it was fully written by hand or with AI it is still a pile of garbage. If two systems are functionally indistinguishable in output quality, then differences in how they were produced (including AI use) are not relevant to me. Also, by now so many tools in the chain now contain AI usage that it doesnt make sense anymore to draw an arbitrary line. Its almost impossible not to have AI somewhere in the chain.I admit that it can be morally questionable to use models such as GPT or Claude, but one can also use a locally trained model on the latest MacBook and people will still cry about it even though only local data has been used to train it. And most other arguments about why AI is bad / unethical can be countered by adhering to some basic guidelines about its usage.Reality is that if you currently work in anything IT related and dont use any LLMs at all (trying them, seeing what they are capable of, understanding them, seeing what they suck at and what they succeed at), you just handicap yourself for nothing.
       
 (DIR) Post #B7ctHYiSwDST8IvXKC by lunareclipse@snug.moe
       1 likes, 0 repeats
       
       @nicole4fox Saying it's just a tool kinda completely ignores the social context it exists in.Linters and LSPs never needed the sheer amount of exploitation to create that LLMs do.but one can also use a locally trained model on the latest MacBookAs far as I know, it's not possible to train a useful LLM model on a MacBook, within reasonable timeframes. You can use one that was trained with lots of nonconsensuslly scraped data like what Google did with Gemma, this doesn't solve most of the ethical issues.These large models tend to rely on exploiting people in impoverished countries for labelling and moderating content that then gets used for training, this is not possible to circumvent with the most common open-weight models.The problem available local models solve is ridiculous resource usage during inference, they don't solve any of the issues with training.AI is useful for some things, when you have so much data to sift through that it would be unfeasible to do manually it can be helpful, provided you can accept a level of failure in the task. That said domain specific ML models tend to be much better at that than LLMs specifically.But we've seen how people used Gemma to sift through code and find malware, or speed up reverse engineering, or empower disabled people to slap together makeshift accessibility solutions that noone else would make for them. There's good uses for these tools but the costs are also very real.The official goal for these AI companies is to make intelligence a subscription utility. That doesn't really sound like a fun world to live in for me. Nor does prompting a chatbot for programming as a job because at that point physical labor sounds more fun.Coding agents tend to remove your familiarity with the codebase, this is extremely bad for open source projects specifically from what I've seen because if you don't have people who know what is where and why it's there, you end up with shit like homebox where someone prompted their way to porting SQLite-specific database code to also run on PostgreSQL, except the end result is a nonsensical clusterfuck that only doesn't immediately break on SQLite because of its weird behaviors like silently setting a negative number to zero instead of throwing an error when doing pagination. The correct thing to do would have been to reject the whole idea of adding PostgreSQL support to that project, because it's at a scale that will never need anything beyond SQLite, and the preexisting code was database engine specific. Porting that kind of code is always hell and full of pitfalls.
       
 (DIR) Post #B7ctIPsvG4JY25E8Tw by luxliquida@critter.cafe
       1 likes, 0 repeats
       
       @lunareclipse Well said.@nicole4fox
       
 (DIR) Post #B7ctMFCGwBc62j6R2u by lumi@snug.moe
       0 likes, 0 repeats
       
       @lunareclipse @nicole4fox there is literally an all-you-can-eat buffet of reasons why genai is horrible, and local models partially solve a small subset of thosei'm not going to say they are as bad as the cloud ones, but they are still unacceptableanyways, good post, luna ​:neocat_thumbsup:​
       
 (DIR) Post #B7cwqE6FseUxBsHDIu by stilic@fedi.cuties.zip
       1 likes, 0 repeats
       
       @nicole4fox you're literally saying this on a GoToSocial instance, and GTS banned llms for good reasons mind you: https://codeberg.org/superseriousbusiness/gotosocial/src/branch/main/CODE_OF_CONDUCT.mdwhat a pathetic take
       
 (DIR) Post #B7cwrDTPhAGfXMa4My by lumi@snug.moe
       0 likes, 0 repeats
       
       @stilic @nicole4fox that's an amazing code of conduct. short and very to-the-point