Yeah, culture changes all the time. The rift between boomers and milenials was already a good example of it. But It's also super specific to regions and countries. Here for example it was popular to say that school degrees don't matter and these days It's actually starting to show the opposite, because due to covid, and heat waves, and aging population, skilled workers with papers are in higher demand than they were 20 years ago.

Culture is one of those things where you have to be ready to adapt and evolve, but not to jump on all the hype trains because there's as many opinions as there are people :D Its similar to any buzzword I feel, except It's more like buzz-trends

I don't, except for company training videos and stuff like that.

I don't know whats my brain thinking, but after reading your rant i thought you are an inmate in prison (behind bars) and they are open thanks to the power outages (electric locks) so you gonna just walk out and do something (maybe escape?)

You cache a whole lot of stuff, I'd recommend to add logs to the cache function you have and watch if it doesn't recache stuff. I'll have to check more closely once I wake up

Midlife crisis , they all do it for the same reason why they buy yachts they rarely use, or cars that can go 350km/h despite road limit being 50~180 or whatever country you live in...

@Demolishun or just make a game that's too hard or too slow and then sell the boosts and equipment to players. It's a legit strategy, if it exists at that level it probably exists at all levels to some degree :^)

My conspiracy theory is that retoor is actually also controlling the spam bots, so she could swoop in with the anti-spam-bots and become our beloved god and savior.

Create a problem and sell the solution!

@retoor so much this. The ads and also the overhyped channels on YouTube have an extreme effect on people, including me. The more I can avoid it the less stressed I am on average. But Im also more and more angry that this is the default state for society and we just allow this to be done to millions of others, especially young and impressionable minds. It just pisses me off that we get more and more research into how detrimental "modern internet" is to peoples psyche and yet no one is doing anything to control or stop it. And the hype around AI is just the newest trend that makes the exploitation worse and as far as Im concerned it should be illegal to pull these stunts. Im pretty sure around 60% of official info on AI should fall under false advertising and should cost them millions whenever It's revealed :/ Peoples privacy, mental health and Security should be more important than money overall, but it isn't and it really makes me want to start a revolution

I mean, Im using LLMs myself. I absolutely love the ability to transform NL problems into formatted data and back. The models are excelent. This is the one task that Transformers were made to do, and they do it well. Im in no way an AI hater or denier or whatever. But Im really annoyed that the big tech companies are massively abusing peoples naivete and ignorance to build up empty hype and line their pockets with money in exchange for promises and a massive electrical and CO2 bill. It's exhausting watching people fall for the same scam again. Fucking google engineers talking about sentient AIs in the labs, faked demos with fake remorse later, inflation valuations based on endless exponential growth with massive layoffs later, tons of scammers and opportunists abusing the tech to trick others, a new buzzword every month to keep the obedient sheep shilling out more money for the newest model that actually performs worse. This shit just never ends because people are not willing to learn

@retoor I can run python scripts with 2 clicks instead or using a massive gpu to run a language model to generate a json that hopefully runs that script.

I can also google such Scripts without the need or the same llm... It's more accessible when you put an LLM interface in front of a library of scripts or if you get lucky and your ask is simple enough that the model seen in at least a couple hundred times to rewrite or modify it slightly. But then we're still just talking about the llm being an interesting model. It's also more than just a sum of It's parts because some form of logic is an emergent property of a language by definition. So It's able to fake thought because It's modeling it. But Its capabilities are vastly overblown. OpenAI is a business, and a predictable one at that. The strategy of making it free at first to hook people and then ramping the price, the fake demos, the neverending hype generation.

Don't be fooled. It's a cool model, but that's all we have so far

@retoor I don't think Sid implies It's not here to stay. But the fact is "AI" is being sold to the masses as something It's not. It's not even AI for godsake, It's just an LLM. Those are not equal. A weather model is not an AI, a physics model is not an AI, but Sam Altman is happy to tell you that a language model is a super intelligent AI, close to AGI, the future is now blah blah blah...

A langugage model is useful as fuck for working with NLP problems... But they pretend It's so much more

same, but I'm keeping it in check, because being a condescending asshole usually has the opposite effect and ends up hurting both me and the other side.

but yeah, I'm happy that it's slowing down right now... but don't hold your breath just yet. I'm pretty sure Jensen Huang or Sam Altman will come with some kind of new buzzword to bring the people back to their side... And hey, it worked 4 times already, I have no reason to think anything else would happen if they do it again

I sleep exactly 8 hours. When I sleep less my energy levels start depleting and I can only handle a day or two of less sleep before I'm borderline unusable

I sometimes sleep more, but that's usually in response to sleeping too little before or going to sleep late, which happens often due to my ADHD brain who can't let go of a task no matter what time it is until it's sufficiently finished

@netikras this dude is actively making me racist against India, and I have to say that my impressions of them were not high to begin with...

How do you know It's because of vowels?

Maybe more people clicked because they seen a new name they never heard about?

Maybe they just don't click your colleagues mails because they already know what it will contain and don't care?

Any other tests were ran?

We're all opressed equaly by try-hards and people who just refuse to move on because their entire personality is built around being a self centered, pain-in-the-ass hero of their own life story that constantly need to remind others how high and mighty they are conpared to all the other lowlifes.

@AceDev yes, obviously. Im talking to other people, since a light chat already got Started on this bot post. Just saying that the general idea is not completely bad, spam or no spam. It's just a bit overengineered

To be fair, I was thinking of building something like Easy Trip AI already. We were planning a vacation a couple months ago and went through the same trouble of planning all the flights and rides and AirBnBs because using a travel agency is by a magnitude more expensive...

I considered it, but then I realized that if I could pull all that hard data that the AI would require to make the plan, I might as well just skip the AI and just optimize the tree of travel options... in the end it pretty much just boils down to a version of the traveling salesman problem. It's pretty solvable and pretty cheap to do, except one thing. One single thing really stopped me from working on this and that's the lack of any reasonable open and free APIs for airlines and sites like booking or airbnb... and god damnit, I don't want to write a selenium scraper for all these sites and then deal with all the crappy data...

I still call mine master, and use black and whitelists. But for clusters I do like master/replica or master/worker better than master/slave since I see my machines more as replicating workers than slaves.

I don't think the conputers get much offended by any of these though

I also notices that even the o4 models is actually failing to finish generation now. Happened to me twice so far, it went completely off script in the middle of spitting my config.

I bet that without they hype they are downscaling the models to run cheaper and smaller models, while still charging the same for tokens. But alas that was to be expected, it's always about hooking the customers first and then slowing degrading and downscaling the service. like turning up the heat on a frog in a pot...

I'm glad I never got rolled into open AI in any way, shape or form. I'd rather stick with my small self-hosted but reliable models...

Im not sure, I just know. There's only a handful of people I consider true friends. It's usually people that don't want anything from me, because they are all skilled people. So if they actually do need something from time to time im actually happy to help. I never felt the emotion of wanting to help someone completely selflessly around people I don't consider friends. But as for what the criteria is Im not sure, probably intelligence, self sufficient, not a snowflake, good humor... Something like that

I also really enjoy tuning performane and memory usage. Im looking into cuda programming right now. I tinkered with openCL in the past but now It's time to get serious about these optimizations on gpu.

But overall I think Im good in most software dev stuff, except frontends, I hate frontends

Didn't this happen with samsung already? They made phones so thin they literally bent in peoples pockets when they sat down. Why repeat the same mistake again with apple this time?

Using google colab is free and you get one free T4 GPU there. Or you can pay to get access to more powerful gpus and tpus later. Bud for playing It's plenty. Though I do understand the desire to run your own HW.

The more you know the more you understand you know nothing.

That's one big truth that never fails.

If you think you know, you're too young and inexperienced to know

If only someone asked real engineers and designers first :)))))

They are not code generators (or any generators) they are general purpose transformer language models. They don't generate, they transform. If you think about them like that, you will have more success. You need to give it a painfully precise specification of what you want and then they will do a decent job. That includes giving them specs of the libraries you want them to use if they are not popular and giving them examples and tables of the expected output.

No, It's not "prompt engineering" It's just normal enginnering, just in natural language. You still have to do the whole spec and design of you want anything decent

I never had these issues with KDE. Maybe I was just Lucky to draw a functioning desktop manager as my first distro.

Even Nvidia drivers work really well these days and install out of the box with apt or custom nvidia keyring

maybe if he did whip you more you wouldn't be such a fucking little bitch!

jk, It just sounds funny in my head, kinda verbal slapstick humor

Ah, those are the cloud cages, where they catch clouds to run our platforms. The more you know!