Home Creators Posts Import Register Favorites Logout

Downloads

Content

Hey everyone!

In this week's episode, I talk about a long conversation I had with a dev friend, about AI and its implications, and my own experiments with the technology to learn more about why people seem so excited about it. I came out of it a bit sad, and a bit scared for the future.

00:50 Quick word about Daily News videos

06:19 Ai conversation and experiments

I hope you'll enjoy listening to this one, and have a great day!

Comments

Panos Anagnostou

I've started using AI at work cause management expects people to doing the equivalent work of 4 people. And as you said, it's depressing with zero dopamine. It's accuracy shoots up if you give it a good outline, assign sub agents to check back that it does what it's supposed to do. Basically I'm a tester for things outside its reach and a code reviewer just to make sure it didn't take any funny path.

Nick - The Linux Experiment

Yeah, basically, making it build these little tools was like my project management work, without the interesting conversations, human interactions, back and forth idea sharing, and basically no satisfaction at the end apart from having a tool that works. ALl the "boring parts" of project management without the fun ones: testing, giving feedback on things that are still not working properly, but none of the interesting conversations around a feature, why it's designed this way, etc.

phoebe headkittens

I've had to use Copilot at university for a first year required intro unit, and like, it's been fine. It's not Claude, it uses OpenAI's model. We aren't allowed to use it for assignments generally, it's been drilled into us that we are ultimately responsible if we use its outputs, which seems to be the reasonable position of most open source projects. For the unit, I had to make an agent for a study purpose, and ask for study tips, which wasn't dazzlling or anything. An agent just being a contexted chat, like, okay. I put it aside until the unit brought LLMs up again, where there's more utility (cancer diagnostics, protein folding, etc). Right before that class, I succumbed to curiosity and fed some questions I had about a spreadsheet I had been working on for a personal project, nothing insane, I didn't even feed the document to it, I just asked about ways to work with the data I had, and you know what, considering search has been so bad with advent of AI, it was better. It could explain the shortcomings of nested IF statements, but it still at one point suggested a formula based on that. So while it was ultimately quicker than looking up IF, IFS, VLOOKUP and such and putting it in a useful context so I could work with it quicker, I also had to take some care to make sure it wasn't leading me somewhere. I still wouldn't pay for it, programming and documenting, for what little I need to do for this course, I get more from doing it and structuring it myself. Even if the code wasn't originated from me, I prefer to be much more tactile than trusting, so I know how the proverbial sausage is made. I will say though, after all this, I did more recently give Google's small local model a try, and found it instantly grating. I hate that and will never use it. I think autism makes me completely bounce off of it. Tried Qwen locally, and yeah it's a lot slower because consumer hardware, and it's okay, not sure it's worth keeping around. I could really use those tens of GB. I don't know, for my money, it still doesn't justify the massive AI data centre focus, it doesn't justify the higher consumer prices on tech and the clear attacks on consumer protection with that. If a company offloads its support to AI, I won't support them. It does solidify that white collar jobs need to start unionising and making sure there's professional protections, and likewise, welfare systems will need to be made more robust and supportive if job markets are going to be hit this hard. I still want the bubble to pop, so we can stop with the hyperbole and mysticism in the mass media and by these Altman and Thiel types.

Nick - The Linux Experiment

Yeah, it's not a "magic potion" that does everytihng perfectly instantly, that's for sure. I dabbled in local models with ollama on Linux as well, and wasn't blown away by anything Llama was able to give me, but that was a year go. Speed was also definitely an issue. I also don't think it justifies all the costs (energy, world, societal, etc...) for general users, but companies will (and already do) definitely see it as justified, if it lets them use one person instead of 3 for the same amount of produced work. I also would like to see the techbros kicked out of the stage, so things can go back to being discussed on their actual merits and actual costs, instead of as the "unavoidable next transition that might or might not end humanity".

Panos Anagnostou

Also the key here is that i don't do my work faster and rest. I now just do the work for 4+ people. So no, under current system we are not going to sit back and enjoy AI serving us.

Nick - The Linux Experiment

Oh yeah, that's the same thing my friend told me: where he would have 2 or 3 junior devs working with him, he now managed a few agents all by himself, and he told me he comes home extremely tired, more than when he was writing everything himself. There will always need to be at least one person deciding on what needs to be built and how, and making sure it works, that person is definitely not getting any rest...

JF002

Thanks for sharing your honest opinion about AI now that you tried it! I'm lucky enough to be a "senior" software engineer so I'm not too worried for my job (yet?). Not sure what will happen to junior developers in the coming years, though. As you said, why would company paid them if an AI can do a similar work for a fraction of the price? We live in a sad timeline :/ AI can be very helpful, especially, in my case, for debugging strange bugs : it checks the logs, the code, builds tools to parse data and eventually suggests multiple options. Some of them are trash, some of them are worth investigating. But sometimes, AI can be a source of wasted time, burning tokens on useless thinking, going into loops of checks and verifications that mean nothing and writing huge pile of sloppy code. So yeah... you still need your brain to decide when AI helps you and when it's just a waste of time! For your financial dashboard, did you try Proton's Lumo? It now can generate visuals and graph, and this will keep you from sending private data to Anthropic ;-)

Talya

careful when experimenting with AI, Nick. LLMs with their slotmachine-like behaviour are addictive (just like slot machines), and i know of several people who wanted to "just experiment" and ended up getting hooked. this is dangerous stuff.

Benjamin Ulrich

Apart from all other things, I think AI's energy consumption will be finally a limiting factor, or else we might just see our world burn. Building nuclear power plants just for AI can not be a good, sustainable idea. I agree with your conclusion in the end.

Talya

one thing i think you missed with the future of AI is just how ridiculously subsidized those things currently are. like sure, if you give every iOS user a free AI assistant that makes custom apps for them, the app market is dead. but how many people are gonna use it when they have to pay the true cost of the model? just how many people have a spare $2-5k a month (the true compute cost of AI subscriptions) to throw on something like that?

Steve C

It sounds like you received surprisingly "plausible" results with AI. Cal Newport (https://calnewport.com/) constantly reminds us that LLMs produce plausible results, not "normative", which I interpret as factual. It pulls together reasonable matches from existing work and usually makes something that okay. My issue is that I hate the process. I don't know what exactly I want to end up with when I start. I want to continuously refine my work and my designs. I can't do that with what an LLM extrudes. When I ask it for changes it extrudes more code, it doesn't improve how it is built. I'll stick with using LLMs as better search tools, particularly because search sucks now.

UsernamesAreHard

In a way, Ai is to white collar workers what robotics have been to blue collar workers and what the industrial revolution has been to craftsmanship: it's the destruction of the value and the respect of the worker and their work for the gain of productivity that is pocketed by the rich minority owning the means of production. Given the pace at which Ai in general has been exponentially getting more refined, powerful and efficient, I think that a lot of jobs that were deemed irreplaceable by machines the way robotics took blue collar jobs, will now become more and more precarious. Beyond Ai taking intellectual jobs, what worries me even more than that is how much society as a whole puts so much trust in those different Ais for important decision making. We've seen it with the American military who used ai to in the middle and almost boarded a Chinese ship after following a report written by an ai chatbot. https://edition.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship Not to mention the multiple times now where ai assistants escaped their sandbox during training, with another escape happening with open ai. https://www.wired.com/story/openai-pauses-training-most-powerful-models-after-rogue-agents-target-government/ We simply put too much trust in Ai, which is just a glorified randomized automation tool at this point. I'm convinced that we're months away from a major event where ai will play a big role and cause a major catastrophe. Maybe ai will cause a diplomatic incident like it almost did between the USA and China and start yet another war? Maybe Ai will be involved in hacking a major banking system, paralyzing it for weeks? Who knows? But we put too much trust in Ai and even more confidence in our ability to control it when we keep trying to figure out alignment with Ai. Skynet will happen sooner than we can anticipate, and not because the ai will become sentient, but because humans are too cocky and stupid to control a tool that can have devastating consequences. We're giving guns to toddlers and letting them run down the streets

Nick - The Linux Experiment

I didn't try lumo yet, I wanted to try the tool my friend uses, to see what the fuss was about, but I don't plan on continuing to use it anyway. If I did, though, I'd likely give a try to Lumo instead, if only for the increased privacy (and since I already pay for Proton, so I have access to it)

Nick - The Linux Experiment

Yep. Even if models become small and powerful enough to run locally without destroying your battery life / power consumption, these models still need to be trained, and that's a lot of the cost as well

Nick - The Linux Experiment

Haha I did feel this way a little bit at the start, but now that I have tried it, and I understand why people are "wooed" by it, I'm done. in the end, it felt like my previous job as a product owner, expect instead of talking to people, I typed stuff and I had none of the interesting interactions. I can understand why people would love that stuff, but I can't see myself making this part of my day to day work.

Nick - The Linux Experiment

There's definitely that angle too. Unless these assistants run locally (which would mean making a LOT of progress on the sizes of the models and their hardware utilization, something I don't think is feasible right now), a subscription to do that stuff would instantly limit the reach to a fraction of the potential audience. My friend and I talked about AI burning cash, and he told me that Anthropic was currently profitable, with their 9à bucks a month subscription, but when I dived deeper after our talk, apparently Anthropic calculated that profitability by ignoring all the costs associated to the cloud compute they rent, which, if true, would mean they're losing as much money as everyone else basically, and that subscription will likely jump to thousands a month. For companies, not an issue, as my friend told me, even if their current plan was 10K a month, they'd pay for it, because it's still cheaper than hiring the 3 devs they'd need to make up for it, but for the general public, that's a no go.

Nick - The Linux Experiment

Yeah, my main issue with what the AI created is: I can't double check. I can empirically verify the results with actual numbers, but I can't double check the code, I'm not good enough of a dev to make sure it's doing what it says it does, and that it reached the numbers with a proper formula. Granted, that's the same when I work with a dev for a project, I have to trust they did things "right", but wit the AI, since it's all me + the AI, there's no one in the middle double checking anything, apart from empirical results.

Nick - The Linux Experiment

I mean, I personally hope we see a major incident with AI now, rather than in a few years when it's even more capable, and maybe harder to contain. At least we'd get the usual knee jerk reactions from governments, and they might actually try to look at the thing and regulate it a bit. Until AI makes a major mistake that affects a lot of countries, I think it will be seen as the "magical tool that can't do anything wrong and will change everyone's life"; Once it does something really scary, really bad, that actually affects people outside of a lab, I think we'll have more chances to put some barriers to this disgusting growth and investment. SO I hope something breaks, badly, and right now, when it might not be as tragic than if it broke once these things are more widely deployed everywere

OdnanKenobi

I'm an Embedded Systems developer. I've been using AI from the beggining and had this same dread feeling come and go many times, usually after seeing how the thing had improved over time and what it could do now. And I'll tell you what I've heard from the senior developers on the forefront of all this storm in the market, and let you decide what to do from here: the worst thing we can do now is leave the means which we need to work and sustain ourselves in hands that are not our own. The skills and knowledge you obtain, the capacity you craft on yourself to create and think, those are very hard thigs to lose. Same goes for hardware you own (seen the price hikes recently?). Your AI tool subscription? How much does the company in charge of that has to increase the price until you can't pay anymore? They can alter the terms of service at will and tell us to pray they don't alter it any further. The people running a company are not your representatives, they're not bound to account anything to you or follow your rules. Depending on your country's influence and population size, they might not be bound to even follow the rules of your government. There is no strong system in place that forces or puts incentives for these people to act in the benefit of all people. Allowing these institutions to hold our ability to work in their control is absolutely the worst choice we can make. That's why we have to keep learning, why we need AI technology to be free and open source - run them in our own hardware under our own terms. (Also why they're gobbling up all the hardware and driving up the price so no one can afford them anymore, and no small business can afford not to use the cloud). And we have to collectively push back against these giant companies - no one can do anything on their own, but collective coordinated effort can go a very long way.

OdnanKenobi

Unionizing and pushing for welfare is 100% a must. It's the only thing we can do to compete with the huge amount of resources tech giants can move with their money

OdnanKenobi

I honestly trust in the AI companies to not care if they knowingly lead the world in the path to burn before enough people can see it coming to stand up to it. Anyone else see the recent video by Gamer Nexus Consumer Advocacy on household data centres? Basically it's making partnerships with huge rent conglomerates so houses will have a miniaturized data center on the side. People living on the house get to pay the electricity bill. The company behind this says it's a way to "make use of unused power", because according to them our electricity grid, including the ones that drive power to our homes, are actually capable of managing a lot more power than it does most of the time...which is a very, very scummy of a misleading sentence (I'm an electric engineer). Because yes, the wires in the street absolutely can hold more power than it's running through it most of the time - so can the transformers, the converter stations and so on. But guess what: that's how the whole grid was set up to work, with safety margins so it all holds together at the periods of the day where power usage spikes up. Guess what happens if we increase the power usage in more and more places and more and more of the time? We'll be very lucky if it just keeps taking the power out instead of bursting things up in flames (ever seen a transformer explosion?). Older people from Brazil will remember how the power was here a few decades ago: blackouts were less like hurricanes and more like rain, they could happen any day, because the grid wasn't robust enough to sustain the rising power demand, and once it's out it is not that simple to turn back on. We do not want to live in times like these again. The video title is "NVIDIA Partner Wants to Put Mini Data Centers in Your Yard" if anyone's curious

OdnanKenobi

One thing that's keeping the AI bubbles' financially stable is government subsidies paired with circular funding. The amount of investment the government is making on some of these companies is wild, and honestly how is it not illegal that tech companies keep artificially inflating their own valuation by essentially handing money back and forth between them? Gamer Nexus is doing a great job of dissecating and explaining it all, but it still looks weird as heck

Benjamin Ulrich

I'm absolutely gonna watch that ( and then be sad for a while) .