Just had a discourse about AI at work... and I just have to vent about it.

LLMs are completely and utterly useless. I've personally never used any AI tool, and aren't going to. Ever.

My coworkers had argued "Why focus on writing an email when you can have an AI do it for you?". Well...

1. Your laziness needs a lot of computational power. AI is as much of a disaster to our environment as bitcoin mining was few years ago.
If we won't stop it, it will only get worse and worse. By choosing to use AI, you directly contribute to climate change.
Hope you enjoy being cooked alive, not to mention energy and water crisis... but you still should write that email yourself. That's what you're being paid for after all.

2. There's no privacy at all. You need to register with an account and everything you do in the prompt is being sold to advertisers.

3. You probably are leaking trade secrets without even realizing it.
Which is a big fucking deal, given you're operating systems directly responsible for safety of the European Union.
You've had to sign an NDA stamped with "NATO Secret", so you should be more aware of what you're saying and doing.

4. No, you don't need to use AI to encode/decode base64. You can install Linux completely for free and use base64 from coreutils completely for free, it's a lot faster and *gasp* it even works without an internet connection!

There are more reasons of course, many more. I can't be the only one looking at the whole AI bubble thinking "okay, OCR is legitimately cool, accelerating certain workloads with NPU as well, same for scientific research.

But all the crap bigtech is pushing onto us (consumers) is simply completely useless and pointless".

Oh and by the way - if you're one of those ""AI Artists"" being mad about real artists protecting their work from scumbags like you, you can sincerely go fuck yourself. You have no right to call yourself an artist just because you asked an advanced calculator to spit out a random image after being trained on billions of artworks without any consent from original creators. You would have to learn what "consent" means in the first place though.

It's clear that AI won't solve issues with climate change, poverty, finding jobs or any other issues bigtech promised it would. In fact, it actively makes everything worse - therefore I refuse to use it, and I will actively fight against it. For the future of our society and our planet.

@elly About (1), just recently have seen somewhere a post about "The sender uses LLM as 'take this short dry message and make it more formal', the receiver uses LLM as 'give me a summary of that ocean of words', so maybe we rather need to rethink communication standards than using computer resources?…"

@elly 4) can you even trust the base64 decode from an LLM, when it can demonstrated to fail at simple maths?

3) my company has some internal instance of chatgippity for people to use. I've not used it myself, but I've seen it severely wasted time of juniors that should've just asked a real person. Instead of an X-Y problem, there's an X-AI problem where you might end up debugging why the LLM provided solution didn't work!

@elly overall I agree with you but I should say that LLMs are absolutely amazing as tools to summarize docs and act as a natural language interface to them and the energy efficient local models do that very well, if it's a common doc or you use a RAG you'll spend very little time and power learning how certain things work (and double checking is easy) but I legit wouldn't know where to look for inspiration for many of the stardust APIs I want to implement. It's not revolutionary, it's just an evolution that's being abused.

For image generation, while most of it is AI slop there's some genuinely amazing use cases like krea.ai which lets you draw and it applies a filter meaning you can sculpt something 100% yourself and it makes it photoreal in realtime, so you adapt your drawing to get the output you want instead of generating from a stupid prompt. It's also very energy efficient, it's 1shot 1 iteration on not that powerful a card!

@technobaboo That's actually a very interesting point of view, thanks for chipping in!