https://kotlinlang.org logo
Join Slack
Powered by
*`IS KOTLIN'S GOLDEN AGE OVER?`* It’s a bold ques...
# announcements
m
IS KOTLIN'S GOLDEN AGE OVER?
 It’s a bold question to ask @jw, someone who was there when Kotlin became an official Android language. His answer is not quite what you might expect. His takes on AI, Swift vs. Kotlin, Flutter, Gradle, open source, and the Kotlin feature that annoys him might also surprise you. Watch Jake Wharton in conversation with @Vitaly Bragilevsky👇

youtube.com/watch?v=9WjPdgOK-68▾

homer disappear 2
kodee loving 29
jake 35
s
Jake is an absolute legend! 😎
true 19
a
Based
➕ 4
r
When it comes to AI, isn't the point of any tool to enable us to do something we wouldn't be able to do without it?
s
AI isn't just a tool anymore if it can replace me. No linter or IDE ever did that. And because of hallucinations, it's also a tool that's inherently difficult to control and contain. I've used it a lot lately, but I'm still not fully convinced. I feel like I can't trust it, at least not in the way I can trust the tools I've used before.
And what a tool it is... Imagine if your IDE's refactoring features messed up every fifth renaming slightly, for no apparent reason and with no reproducible pattern.
👍 4
I find it really valuable for code reviews, but actually having it generate code still feels difficult.
r
All tools in the history of civilization replaced some people (who did the work earlier) with other people (who control the tool). AI doesn't seem to be any different, at least yet 😉
👍 3
j
Calling it just a tool is such a cop out on the ethics behind it. Slavery is also a tool that lets you do things you otherwise couldn't without it. You know what's better than paying someone to do something? Not paying them!
💯 12
☝️ 2
☝🏾 1
➖ 2
➕ 1
r
I am only addressing the claim that using the tool is bad because it allows you to do more. I simply find that unconvincing. I am not speaking about the ethical aspects, as I lack the knowledge to either agree or disagree with them.
j
Nuclear bombs allow you do a lot more
But who can say whether or not they're ethical, right?
Even if you aren't capable to weigh one ethical aspect of these tool the great thing is that they have so many fundamental and systemic ethical problems that there's still ones you definitely can.
r
You're comparing AI to slavery and nuclear weapons. Sorry, but aren't you exaggerating a little bit?
👍 1
j
Not really. Climate problems will kill far more than nukes ever did. And guess what devours disproportionate non-renewable energy and dumps heat into water systems to sustain itself?
👍 3
r
Meat production still destroys the climate far more rapidly than AI ...
j
Cool. Eat less meat, too! Whataboutism is such a lazy argument. You're ignoring what we're talking about! Except food actually sustains living humans and we still can't feed everyone. So instead of watching a spinning butthole logo make a generative image of child porn we could stop world hunger AND use less meat. Great argument there.
👏🏿 1
👏 3
😆 1
You'll forgive me for no longer engaging in your bad-faith arguments.
s
I agree that, aside from the lost jobs, the other big issue is its energy consumption. It’s also driving up the prices of gaming hardware. I’m very sceptical about AI too, so I watched the talk with a lot of interest.
r
I didn't plan to play devil's advocate. I am merely arguing that AI is not something inherently bad. But ok, end of discussion.
j
Sure, but you can't separate the two sides. No one claims the output of LLMs is entirely bad. It's also unethical to enslave a billion monkeys sitting at keyboards at the other end of these chat interfaces, but they produce nonsense output (and, statistically, at some point, the full text of Moby Dick). The only reason we have to talk about the ethics is because the output is good. In the slave example, someone who's not a slave receives great value! Their output is great! These systems are designed so you don't see or feel the impact directly, and the people who build them lie and use their power to commit crimes behind the scenes. They make the tools available because they only care about control of the market, not some altruistic sense. If you actively choose to ignore the harm you are free to do so. Lots of privileged people ignored the harm of slavery as well! If the comparison is uncomfortable: good!
a
Putting AI at the top of the list is reasonable, but many of the other items resonate with me as well. I left the corporate software industry at about the same time as you. If and when my bootstrapped startup doesn't become viable, I'll make use of a similar list to help filter out the slop.
y
Looking at what happened to two math researchers, Tristan and Levent. They used OpenAI's Codex tool for their work. OpenAI took their research, and then used a marketing campaign to pretend OpenAI's AI solved the math problem by itself. When they got caught stealing the work, OpenAI quickly tweeted "congrats" to the researchers just to make the drama go away If they did that with math research, imagine what they can do with business production code !
💯 6
s
Exactly. I just watched a YouTube video explaining the whole story. OpenAI trained the model on their prompts, and I guess everyone else’s as well, while claiming that they didn’t. That effectively makes another strong case for local LLMs. I assume the whole “we don’t train on your data” promise can’t really be trusted. But of course, local LLMs have the issue that they’re also trained on vast amounts of copyrighted material belonging to others. It's a huge mess. Just as Jake said in the interview.
y
Most developers can't afford the hardware needed to run these models locally unless their company provides it
a
@Yassine Abou even then that is hardly possible. Running something like GLM requires 500k in hardware and specialized power lines
s
@Arjan van Wieringen With Q8 quant a Mac Studio / Mac Pro with 512 GB Unified Memory will do. One per developer.
Self hosting bigger models is not impossible, just quite expensive. Companies could surely afford that.
r
In 10 years you will run this on your phone 😉
😄 1
👍 2
a
Still doesn't solve the training issue though. Training costs the majority of power and resources. I love how we went from calculating hashes for a fictional currency ledger to summing up neuron activation values. It is an improvement, I will say that.
s
Right. If Cloud AI goes away entirely, we’re stuck with the local LLM we have today.
Maybe selling local models would become a business, too, one day.
✅ 1
y
@Robert Jaros Give it 5 years, and some tech giant like Shopify is gonna tweet: "Yeah, we tried AI ... we're officially going back to doing it by hand. Here's why" And everyone is going to go crazy😂
😂 2
❤️ 1
a
There are already thousands of YT clips, linkedIn posts, and blog post describing exactly that. I think coding agents are just a tiny piece. It is way more worrisome that we have companies like OpenAI that are able to do massive cyber warfare attacks using thousands and thousands of agents and almost unlimited compute. It happened a few times when it was not their intention. Imagine when they or governments force it to BE their intention.
💣 1
➕ 1
d
I'm looking forward to the positive aspects that could result from continued advancement of AI. If AI can solve the problems with our nuclear fusion attempts, it could address pollution from all sources not just from data centres. It might also discover high temperature superconductors and other important advancements in material science. Perhaps cures to popular diseases such as cancer, diabetes, etc. High quality education for everyone, not just those that can afford it. Advanced robotics for dangerous tasks. Efficient food production to finally solve world hunger. etc. etc People that are starving of hunger aren't concerned that their food was created with an AI that was trained on copyrighted material. Rules and ethics are important but they aren't supposed to be absolute. You wouldn't hesitate to ignore a traffic light if you could save a child's life. Etc. Let's face it, all of us reading this are wealthier than 99% of the people on earth. Privilege affords more ethics than those struggling to find their next meal. Note that I'm not advocating for AI companies to break all laws with the end justifying the means. What I'm getting at is that all absolute hard stances on both sides of this are wrong. The right answer is somewhere in the middle.
➕ 6
s
That's only possible if we can drive hallucinations to zero and AI can genuinely be trusted. That's a big "if". I'm not even sure we can ever get there. It may be an inherent limitation of the technology rather than a problem we can eventually solve.
I recommend the channel FatherPhi on YouTube, which is mostly content showing off how unreliable AI is today. For example youtube.com/shorts/gPthZLTnzu8 and youtube.com/shorts/…. It's fun to watch, but also let us see serious issues in todays models.
Advanced robotics for dangerous tasks.
We have human controlled drones for that already.
I really don't want autonomous weapons or medical decisions made by ChatGPT.
d
Humans make mistakes all the time. That's why software is so buggy. Driving hallucinations to zero isn't actually a requirement (although it would be nice). It just needs to make less mistakes than an intelligent expert in that field for it to be a genuine improvement.
s
Humans make mistakes all the time.
A human can be held responsible; AI can't. A human can also build a track record of success. At some point, you can reasonably conclude that a doctor is trustworthy and knows what they're doing. AI is different. It can get the same job right four times in a row and then suddenly get it wrong on the fifth, for no apparent reason and without any reproducible pattern.
💯 3
I liked the days when computers didn't make mistakes on their own. Any error could ultimately be traced back to a mistake made by a human developer. With AI, we have a black box that can make mistakes of its own that nobody can fully explain or reproduce. That's scary.
Of course, I hope AI will get there and ultimately lead to a better future. But what has been sold to us so far is mostly utopia. What we've actually gotten so far is lost jobs, at least in my case, and a whole lot of slop.
r
A human can be held responsible; AI can't.
yet 😉
s
How could it ever?
The companies that make AI may be held responsible, sure (if laws change). But imagine you're running Qwen locally on your own hardware and it messes up. Good luck suing Alibaba...
Look at the video I shared above and imagine having to argue with ChatGPT instead of a human if you receive a certain medical treatment or not. Horror.
My point of view might change if AI actually advances to the point you're hoping for, but that's far from guaranteed. I don't have high hopes.
d
Getting it right 4 times and then messing up on the 5th is a 20% error rate. There are statistical methods for reducing errors. Simplistically, if the model is known to be right 80% of the time then sample it a bunch of times instead of just rolling the dice once. An expert doctor with a proven track record may have had too little sleep the night before resulting in poor judgement under stress. A statistical model can maintain much higher consistency. A model would be validated extensively before being placed in production especially in medical settings.
s
We'll see.
r
Why do you want to argue with a human in such case? To make him pay you money? If so, we just need to wait till AI will have it's own money.
s
Yes, that was a healthcare example.
I don't get your point... AI's own money?
r
We do not know what the world will look like when AI becomes fully autonomous (and maybe even conscious). We don't know when it happens, but it's inevitable.
And yes, I’m an optimist - I rather imagine Lt. Cmdr. Data than Skynet 😉
s
I hope it's not inevitable. We should stay in control.
a
I dont agree that hallucinations need to be brought to zero. You should always have deterministic guard rails and validations in place. It is like expecting 100% guaranteed safety from an autopilot. Impossible. But 99% is extremely useful with safeguards.
s
AI is not there yet and it's questionable if it's possible - even with safeguards. I'm more on the pessimistic side than optimistic, maybe that's on me. Did you watch the two short YouTube videos I linked above? What do you say to that?
r
Of course it's not there yet. But it is rapidly heading in that direction.
s
And what makes you think that we won't hit a plateau?
I want a Star Trek future as much as you, but I keep my expectations low to prevent disappointment. 😅
a
The shorts show ways where AI fails, but they are also not at all related to the domain of coding at all. The field of LLM's at this scale is in its infancy. Nobody can make predictions of where it might go. We need to be careful though. On the aspect of responsibility. This is already the case with aircraft autopilots and full-self-driving. The exact same discussions....This isn't new territory.
a
I watched @jw’s whole conversation in that video, all answers are fully logical and based on ethical values. One very eye catching answer based on Aaron Swartz situation in contrast to what modern tech companies doing
👍 1
r
@jw Your comparison with slavery here isn't very illuminating. Slavery requires coercion and violence. LLMs in general, by contrast, are indeed tools in the normal sense of the word -- they do not inherently violate anyone's rights. You are comparing something that was parasitic by design in that its entire function was capturing existing values (slavery) with something that certainly creates new values (LLMs). A better argument would be if AI companies are violating the copyright of content-rights-holders (the example someone else gave of OpenAI ingesting the prompts of other researchers is a great one). If true, this is simply an IP compensation or contract violation problem rather than one of "slavery" and existing IP and contract laws should be enough to resolve it. This is not an argument against the use or consideration of LLMs as a tool more generally.
👍 1
j
Yeah... No...
But, I don't care about engaging here anymore. Sorry!
👏 1
Enjoy your blue pills
💊 1
s
@rocketraman There is no question that the copyright infringement is real. Facebook was already caught downloading massive numbers of books from The Pirate Bay to train its models. Almost all training data violates the owners’ licenses and fails to give proper credit.
r
@Stefan Oltmann Sure -- and I hope the copyright holders go after them for it in court and are duly reimbursed. It doesn't in any way show that LLMs as a tool inherently rely on that bad behavior. They don't.
s
I think I'm with Jake here and don't want to engage any further. Everything has been said, and we'll see how all of this develops.
👍 1
a
So you will stop all usage of AI now
s
That’s not what I meant. I’ll keep experimenting with it, but I’ll remain cautious and sceptical. I can see both its potential and its limitations. My experience using it as a coding agent has been largely disappointing, while I’ve found it much more useful for code reviews. I still don’t think these systems can be trusted completely, and I’m not convinced they’ll improve as much as some people expect. I’m still trying to figure out where to use them, where not to, and what guardrails are actually appropriate. I just hope we use AI responsibly and keep it away from critical tasks until it’s actually ready, assuming it ever gets there. Seeing things like FatherPhi’s videos on YouTube doesn’t exactly increase my confidence. Jake is 100% right about the ethical concerns. I’m somewhere in the middle: I can see where it genuinely helps, but I can’t exactly cheer for a technology that took my job away, even if I benefit from it in other ways. At this point, though, I don’t think I have much more to add. I’ve probably covered everything I have to say about it. If the technology improves, my mileage might change too. Disclaimer: I used AI to reword the gibberish I just wrote, so that’s probably still my best use case for it. 🤭
a
Sp you completely agree with Jake, but continue like the rest of us but from a higher horse. Got it 👍
👎 1
s
I'm not claiming moral superiority here. As I said before I agree with the ethical concerns, but I also see legitimate uses for the technology. I'm not judging any developer for using it, if it came across that way. But I am definitely angry at CEOs replacing developers with AI and killing jobs in the process. That's true. Ultimately, I don't have that much choice either, since companies increasingly expect you to be an "AI babysitter". I think that's a problem too. I'd rather use the technology on my own terms, where I actually see value in it.
43 Views
Open in Slack
PreviousNext
Powered by