How does the Elixir community feel about the ethical complexities of AI, OSS, and learning/teaching?

No, since they have also paid plans. I don’t believe a BigTech would respect OSS and don’t use any information from it in their paid plans.

Companies tries to force LLM to do everything like it’s a real “AI” which only ends up with frustration as the user expectations are not met. The best “AI” tools are the old tools based on the solid algorithms with only changed input method (chatbot).

I believe that if people would really focus on serious research then LLM may be excellent for learning new things and improving skills. However we don’t live in a world full of altruists that spends billions just to help other people.

So far the companies are trying to fire developers that “demand too much” while still expecting thousands if not dozens of thousands for their hardware (not for cloud/server purposes of course). It’s like expecting from gamers to use netbooks and using an LLM to “replace demanding gamers”.

We have lots of models in lots of variations. Let’s even assume that the original author have paid for using OSS projects, but what with others? Do you really think that you have full copyright for what LLM produces? Most people are not reading the terms of use. Those who read them most likely don’t have time to follow it’s every update.

If you don’t have a law team it’s terribly hard to tell what are your rights as everything may be changed. Do we even have a law regulations for that? Depending on law system we may not even know if the chatbot creators have a responsibility to inform us about a problem. If we are safe can we say same about ≈200 other countries? Do all other countries respect our copyrights?

It’s one huge and frustrating mess that is basically out of control. What stopped corporations from moving factories to China? Nothing? Then what’s stopping them from moving “AI teams” if our law becomes “too restrictive”? As always the biggest evil in a new tool is their owner/user.

I have some intellectual property experience. I’ve never been persuaded by this point. Legally I don’t think it’s infringement. Morally, I don’t see how it’s against the spirit of the MIT or most OSS licenses.

I’m finding the opposite: I’m ultra-senior but new to customer-facing Phoenix and completely new to LiveView.

It’s been a fantastic learning experience. I think this might depend on one’s learning style. Maybe also how one came to be a programmer. I’m self-taught beginning when I was 13 or so. EDIT: And then 4-year college Computer Science and a bunch of grad school.

I don’t think there is a single restrictive license that could be applied to any other and even if … are you licensing your code with it?

Do you know original authors or original license? Assuming it’s MIT it’s like I would assume the world exist to fit my own needs and everyone is supposed to help me achieve what I want. How does it sound that way?

Look that it’s not “AI”, but LLM. Why does it matter? It’s a Large Language Model - would it really surprise you if 2 data from sources that have conflicting licenses are used to generate a single solution? You don’t even have to worry about your own license if the code standalone is completely illegal.

For sure you don’t “trust” the “AI” source as same as with license, right? If you did then you’ve learned a lot of fake information. How can we even trust a tool that cannot correctly do simplest tasks without hallucinations?

I really can’t understand how people could blindly assume the best case all the time and ignore possible consequences. Imagine the case that a person that is known from TV as a thief says: take this money and trust me they are from a legal source. Could you be so sure that you are safe in the court? Or is it maybe kind of “low risk, no problem” idea? Now let’s play with words and say that originally the source was legal until the money were stollen … I don’t see even a single line of defence here …

I’m worried that this hype becomes too dangerous and therefore could lead to unpredictable things. In worst case when people would stop questioning things then we could end up with any progress. If that would happen globally then we would have even a technological regression in the long term.

In my experience, the tool can correctly do the simplest tasks.

How can we trust co-workers to do the same?

You may want to have a look at some of these threads too:

• How useful do you think AI tools are/will be? (Poll)
• What are your feelings on AI in general?
• In an age of AI, what can programming languages do to survive/compete?

You’ve sort of answered this in your question (bit in bold). Most OSS contributions were made before AI existed, and if we look at what’s going on in the wider space we see there are numerous lawsuits against AI companies for essentially stealing other people’s work (lots of threads relating to this on Devtalk). It’s a bit different with OSS since most OS licences pretty much allow anyone to do whatever they want - tho it could be argued that the intent was significantly different back then - and that those licenses were more to offer ‘free to use’ type rights while offering guarantees in case the maintainer/s died or lost interest or changed their minds.

I expect the pushback from OSS will come a little later than say with artists, musicians and other industries which are being impacted more severely right now - because the impact in the dev industry may be felt a little later, through job losses (tho it’s already started) or AI companies essentially killing off languages once they’ve used them to get where/what they want.

Many would argue we should exercise caution - history tells us that big tech/capitalism on that level simply cannot be trusted:

There will probably be pros and cos, though some early indications are not good:

Does ChatGPT harm critical thinking abilities? A new study from researchers at MIT’s Media Lab has returned some concerning results.

The study divided 54 subjects—18 to 39 year-olds from the Boston area—into three groups, and asked them to write several SAT essays using OpenAI’s ChatGPT, Google’s search engine, and nothing at all, respectively. Researchers used an EEG to record the writers’ brain activity across 32 regions, and found that of the three groups, ChatGPT users had the lowest brain engagement and “consistently underperformed at neural, linguistic, and behavioral levels.” Over the course of several months, ChatGPT users got lazier with each subsequent essay, often resorting to copy-and-paste by the end of the study.

It may be too early to tell - one thing may well be a given though, we probably don’t want to walk into the age of AI blindly or without caution…

Thank you all for the thoughtful responses. I’m actually very encouraged to see such a diversity of viewpoints right off the bat. Even if there are no clear answers (which is not a comment on the quality of discussion here!) knowing that these perspectives are strongly represented in this community creates a stronger draw for someone conflicted like myself to engage with it.

It may be too early to tell - one thing may well be a given though, we probably don’t want to walk into the age of AI blindly or without caution…

Well said. Both enthusiasm and skepticism are essential, for a community which hopes to evolve in the right directions. I wonder if the same can be said of an individual.

I find myself in agreement with much of what you’ve said @Eiji, and I unfortunately am finding it increasingly difficult to reconcile these beliefs with the daily life, influences, and pressures of a working software dev who’s main priority in life (at present) is to provide for their family. I’m finding myself gradually start to ask questions, “do my own personal principles really matter, if something is already here, it exists, and perception of my success in my job seems increasingly predicated on embracing that thing?” I don’t think my ideal world would have AI in it, but I have to accept that other’s have different experiences, feel differently, and none of us live in our own personal ideal worlds. At the same time, principles matter, ideals matter, I have immense respect for that. We should all do what we can to influence the world around us for the better.

@dogweather Thanks for weighing in, I’m glad you’re finding valuable experiences in working with these tools. It’s fascinating how disparate reports seem to be with this ~ learning style does seem highly relevant. If you did feel like elaborating on that with any further thoughts on your personal experiences, I’m all ears. What aspects of a learning style, or one’s approach, might make for better (or not) growth while using LLMs?

@AstonJ Appreciate the broad perspective and summary of the state of play. Thanks for compiling those links, will definitely be reading in depth. Very interesting study as well, I hadn’t come across anything that went as far as EEG analysis. There’s also the recent METR study showing programming tasks taking longer with AI assistance. Again, so incredibly perplexing how these will likely go against the reported experiences of many around us (numerous colleagues on my part, not to mention so. much. discourse online). I am getting the sense that LLM-assisted/driven software development may be a different beast entirely than generative-AI tech applied in other areas, for better or worse – not to say that they don’t share many of the same pitfalls and moral quandaries. What disturbs me most is the relative decrease in discourse I’m sensing across the board about topics like

LLMs behind these tools likely trained on data taken from OSS authors without their permission , and without attribution ~ do the benefits of these tools for OSS outweigh this?

It feels like maybe we just got collectively tired of talking about this, but the more powerful these models get and the more enthusiastic their proponents become (nothing against them) the more important it seems to me to keep these ethical complications in mind, rather than trying to ignore them or put them behind us. I’m feeling more and more compelled (sort of against my will) to engage with these tools than ever, and I’d rather things be complicated if that’s where we’re headed.

I’m thinking about my answer for you.

But a quick question: you haven’t had good learning experiences working with AI?

Thanks @dogweather, very interested to hear more. Sorry about the delay, that’s a great question. As you’ve probably gathered, I’ve so far fallen mostly on the skeptical/reluctant side of these issues, but I do work for a company that has recently developed plenty of AI enthusiasm and multiple AI products (bolted on, not the core offering). So full disclosure, I have only a surface-level familiarity with these tools and integrating them into my work and learning processes. I understand there are many “prompting skill issues” and “using the wrong model” issues when it comes to AI; I don’t assume I can use it well.

My main context here comes from observations of co-workers, working in two modes – we can call them “supercharged autocomplete” and more recently “hands-off agentic coding”

“Supercharged autocomplete” (generally via Copilot, Cursor)

• What concerns me here is an observed decrease in methodical thinking and working, possibly due to increased distraction/noise while coding.
• While collaborating with colleagues operating in this mode (remotely, watching their screenshare) I observe more “throwing spaghetti at the wall” rather than building understanding down to the details at the syntactic level.
• In my own learning process over the years, I’ve gradually (too slowly) developed an intuition that working methodically is essential, to continuously “lock in” progress on some problem and avoid “circling back on your own path.” I feel my colleagues have had more trouble getting lost while trying to figure out tricky bugs and problems lately.

“Hands-off agentic coding” (Cursor, Claude code)

• The tradeoff here seems obvious ~ you’re no longer working directly with the code. My peers have told me this lets you operate on a higher level, you still need to understand the code, you still apply rigorous code review. Regardless of how effective this may be at scale (of weeks to months to years working with a codebase) I’ll take their word for it, but the bigger concern I have is that the process of code review has always seemed more like a “hole poking” rather than “deep understanding” exercise to me. Speculatively, I don’t believe I can learn and grow in my craft through code review alone. This recent mastodon thread that’s been circulating has some interesting parallels, providing a “red/blue” perspective on LLMs and the code review process.
• My experience has been that I learn best when tinkering and experimenting, building up a system from the smallest of parts and interacting directly with it through as-fast-as-possible feedback loops along the way. Give me a REPL on an airplane and I’m happy as a clam trying to figure something out with limited resources; give me a set of docs to read through and I’m miserable (though good docs are essential ~ when paired with hands-on activity). Similar to writing out something by hand solidifying it in your memory, working directly with code in live operation seems to help me internalize it and remember how it all works naturally. These low-level experiences feel important.
• On a tangential note, in the web development space I’m not at all happy about the movement towards the “path of least resistance” for building UIs with LLMs ~ they seem best at working in a Tailwind-centric style which tends to produce an HTML “soup” very difficult to inspect and make sense of any structure through inspection from the browser, due to lack of any meaningful terms associated with DOM elements. Part of the original promise and gift of web technologies was that you could look under the hood and figure out how everything worked ~ this was a boon when my initial interest was sparking and I was starting to learn. This aspect of the web was eroding prior to LLMs, but I think they’ll act as a serious accelerant; I’ve already started to see arguments for coding styles, practices, and technologies that cater to LLM involvement as a priority over human learning and understanding.

As someone who’s written million lines of OSS code I have some perspective here. not all OSS is the same. My projects are made available under the MIT license. I have given my permission to use it for whatever purpose you or anyone else likes. Intentionally.

But as for the ethics here, the things that were wrong and illegal are still wrong and illegal and we should pursue companies to that end. I’m more than happy to see investigations and lawsuits for using copyrighted material, and I hope that any found infringements are punished substantially. The individuals, and the company.

For this one, honestly it’s just hard to say and I haven’t seen convincing evidence one way or the other on how it will impact things. There is compelling evidence in both directions that actually using an LLM to learn will or won’t help etc. But “LLMs existing while you learn” is not something I’ve seen measured. It could perhaps have a negative effect via encouraging bad habits in beginners, but also beginners have always had bad habits and reality is what trains it out of them. (i.e you’ll get fired or reprimanded for bad work or slop, and thus have to invest time learning things). Very up in the air for me though.