remote/open-slopware
1
0
Fork 0
mirror of https://codeberg.org/small-hack/open-slopware.git synced 2026-07-13 23:14:32 +02:00
Code Issues Projects Releases Packages Wiki Activity
open-slopware/why_not_llms.md

19 KiB
Raw Permalink Blame History

Why not LLMs?

Legal Ramifications

LLMs are often trained on, and thus prone to regurgitate (either completely, or in-part), chunks of code that are licensed under terms which have specific legal requirements that a contributor using LLMs may not understand or even be aware of when making a contribution. Regardless of this ignorance, it falls to the repo's owner to comply with the terms of any and all licensed code integrated into their project.

Legal Cases And Law Problems

License Problems

Legal, copyright, and ethics problems arise, especially with copyleft licenses such as the GNU (A/L)GPL. With the "help" of AI the copyleft code may be "license-washed" very easily.

There are ongoing problems with AI "license-washing" in the FOSS world:

Stolen Training Data

AI companies use data from across the web to train their models, most often without the website owners' and users' consent. Big tech companies like Google and Meta are scraping data from the users of major FOSS projects, such as Mastodon, WordPress, and other ActivityPub-powered and self-hosted software.

  • In 2023, the Washington Post published a list of sources in Google's C4 data set. A multitude of fediverse instances and personal sites were included. The fediverse is known for its userbase being major proponents of privacy and opt-in consent, making this especially jarring for those who have chosen to use decentralized social media for control over their data.
  • In 2025, a similar leak of Meta's sources was published. Meta's list demonstrates how their integration of ActivityPub into their Threads software has enhanced their ability to scrape content without authorization. Threads is widely blocked in some parts of the fediverse, but their scraping of server CDNs has allowed them to get around that. Notably, both the CDN domains of the managed hosting services masto.host and fedi.monster are included in the list; large servers like mastodon.art, which is hosted by the former and has many artists who've left sites like DeviantArt and others due to their AI scraping of user content, had media unknowingly scraped.
  • In March 2026, a research paper showed that simply fine-tuning LLMs resulted in outputs containing up to 90% of entire (copyrighted) books, contradicting LLM companies' previous statements in court that their models do not store copies of training data. After fine-tuning exclusively on a single author, the researchers were able to cause the LLM to output works from over 30 completely unrelated authors across different genres. None of the models were explicitly trained on these books by the researchers, which indicates that LLMs always carry with them a considerable amount of copyrighted materials from training.

FOSS projects listed in this repo are using tooling that blatantly disregard licensing and violate of Codes of Conduct, making said tools antithetical to FOSS' purpose.

Environmental Impact

To start learning a bit more, you can checkout the wikipedia page on Environmental impact of artificial intelligence. We're very open to people contributing other explanations, links, and resources to learn more about this. Here's what we've gathered so far:

Labor

AI usage and normalization contributes to labor violations in many ways that are obvious and some you may not be aware of.

On one hand, many things that you think are "AI" are actually humans in another country pretending to be an AI chatbot for you for either extremely low wages or in some cases, no wages e.g. prison labor. This is particularly common for "friend"/"sex" bots, but it is also extremely common in the image/video identification. You can find a bit more info at the following links:

Poor Code Quality

Vibe coding / agentic workflows result in poorer code quality, and relaxed oversight practices. These effects may be compounded by the common practice of using additional LLM-based tooling to provide code-reviews.

Deskilling

There is increasing evidence to show that LLMs negatively impact developers' coding abilities:

  • Brains show less activity when completing tasks with LLMs compared to completing tasks with search or completing tasks without digital help.
  • Developers who use early-2025 LLMs reported higher subjective performance, but were measured to have lower objective performance. This gap between subjective and objective performance was considered notable.
  • In an Anthropic study, learners using LLMs demonstrated lower learning rates on average compared to learners not using LLMs.
  • A recent study uses the term "cognitive surrender" to describe the way humans tend to offload key critical thinking skills onto LLMs, even when the output is wrong.
  • A paper entitled "AI Assistance Reduces Persistence and Hurts Independent Performance" from April 2026 by academics from MIT, Oxford, UCLA, and Carnegie Mellon showed alarming evidence that performing a variety of tasks with the help of AI for only 10 minutes causes "inpaired unassisted performance and reduced persistence". The researchers noted that "although AI assistance improves performance in the short-term, people perform significantly worse without AI and are more likely to give up"; they also pointed out that "these findings are particularly concerning because persistence is foundational to skill acquisition and is one of the strongest predictors of long-term learning".

Infosec risks

LLM usage results in massive security holes.

Health and Safety

There's been a number of high profile incidents that have resulted in endangerment or death. Here's some examples:

LLM use has also been linked to new-onset psychosis.

Ties to the War Industrial Complex

A lot of AI companies also work directly with nation states for use in their Departments of War (sometimes called Defense) which in turn leads to further AI usage during war and invasions. This is coupled with NYT: Palantir, Anthropic and small start-ups are generating rewards from their investments in defense tech.

As another example NPR: OpenAI announced Pentagon deal after Trump banned Anthropic which was due to the USA Department of War launching an AI acceleration strategy.

Due to the nature of LLMs being only kind of as good as the data they are trained on, this can lead to additional civilian deaths and housing/infrastructure damage either intentionally or not. Examples:

Content Warning: War details, death

All of this to remind you that if you use AI, you're helping to support these companies and the additional activities they participate in, outside of generative code or images.

Effects on Policing

Police have quickly embraced AI, which has already directly led to people being jailed for things they've never done. As examples:

This is, in part, due to companies such as Amazon Aggressively pushing police to use AI which they do through both facial recognition and offering compute for predictive policing. With regards to facial recognition, here's an example of how it too can lead to false arrests: Face Recognition on Flawed Data.

There have been warnings about AI in policing, particularly around racial bias, such as:

Maintainer Fatigue

Having to deal with the onslaught of many LLM written pull requests and issues, causes real maintainer burnout that stagnates projects as maintainers become overwhelmed with half baked, poorly written, insecure code. Here's some examples:

Effect on Hardware Prices

The demand for construction and outfitting of new data-centers to host AI/LLM compute capacity has overwhelmed global supply and production of multiple hardware components. Memory, Storage, and GPUs have seen massive increases in price for both consumer and enterprise models upward of 400% in some cases.

The lack of supply has led large system-builders to purchase production capacity from OEMs well in-advance of delivery leading some manufacturers to end consumer-oriented product lines in favor of enterprise capacity.

The down-stream effects for consumers is that near all electronic devices which contain memory and storage will see their prices rise and availability decline. Those who already own existing electronics and computer hardware components may also find themselves unable to have their devices repaired or replaced under warranty.

This all results in shrinking the pool of people who have access to building home computers for any purpose, from gaming to coding to home labs, which in turn makes the tech industry less diverse due to people who have been historically marginalized having less financial resources to learn the skills at home. When this is factored in with the price of college being unaffordable in many places, we will see a sharper decline in disabled people, people of color, women, and the queer community entering the tech industry.