We'll be standing on the crusted desert of what was once farm land and people will still be saying, "Yeah but we don't actually know how hot it was in the Triassic."

It's astounding to me how many people think this is about "audio quality." It has nothing to do with audio quality. It's about user experience.

For all the UX gurus of the digital age, they certainly miss the reason why people prefer vinyl. No it isn't some hipster bullshit, it's just a better listening experience. The fact that YouTube, Spotify, etc. will just play forever is actually a negative. It causes the music to become background noise. Now maybe that's how you want to use your music (it is for many people), but for people that like to listen, having the music stop every once in a while is actually a positive experience.

When a record plays, it has a definite beginning and end. When the record needs to be turned, there is a break that you can allow to continue as long as you want. The user experience of a vinyl record is actually superior to Spotify BECAUSE OF all the little details, not in spite of.

What some people think of as friction, the enjoyers view as immersion.

Let me scan your eyes so I can protect you from the evil AI (which I am also creating).

that if the game was good, this wouldn't matter at all.

Seek help.

When you start to A/B test your product, configurations that make money, but worsen the service will emerge. Once a product succumbs to the pressure of shareholder demands for profit, these configurations are promoted as a means to an end. Products teams, analytics teams, and even engineering teams are designed to find as many growth-encouraging configs as possible. Their quest for the almighty dollar inevitably leaves the product a husk of it's former self. The service will never be the same.

As these "solutions" become more difficult to discover, and as growth slows, the quest for increasing revenue will alienate even the core users. As the audience declines, more product compromising solutions must be implemented until, finally, no blood will come from the stone. The product is sold and runs as a museum piece, an artifact of silicon valley and a marvel of the efficient outcomes of vampire capitalism, in the treasure hoard of a bigger, stronger, younger company.

Must be why all the clinical trials of the vaccine showed massive deaths in the vaccinated group compared to the unvaccinated group.

Oh wait, no, that didn't happen.

This comment is like the story that the father of statistics, RA Fisher, wouldn't say that smoking causes lung cancer.

Yes of course, smoking and lung cancer are correlated very strongly - but Fisher would never say that smoking caused lung cancer. He couldn't show it statistically, so he wouldn't say it.

Of course, at a certain point, it's up to the disbeliever to posit *some other reason* that lung cancer and smoking are strongly correlated if it's not causal. And the same goes for COVID deaths.

>cannot actually be used for much because of hallucinations

https://googlethatforyou.com?q...

Consider a model that is correct 90% of the time. Oh, but I guess we can't use it because of 10% 'hallucinations.'

The 'models don't work unless they are perfect' crowd needs to learn something about statistics.

>It's primary multiplication and addition

Oh my god, this is posted as a serious comment. As if there is just no benefit in using a fast linear algebra library instead of a slow one (or a fast architecture instead of a slow one).

Hey, technically I can solve these back-propagation equations in a notebook, why do I even need to write code at all?

I just did a quick google search out of curiosity, and NVDAs data center revenue is 10x AMDs.

Maybe that has something to do with it, and not your "a little bit worse products' claim.

Anyone that uses any generative AI tools on their local machine knows NVDA devices are much faster than AMD at a similar price point.

Certain models take a long time to train on CPUs. Nvidia has been working on this problem for years, and training those models using GPUs is much faster. If you're retraining models frequently, fitting models on CPU is simply not practical.

It doesn't provide a "better result." The model is the same. It just does it much, much faster.

As if you needed special congressional knowledge to buy nvidia stock 2 years ago.

I find this post interesting because of how little it understands about the current technology landscape.

> Fundamentals of NVIDIA business are gaming and CAD acceleration

It's like you've been under a rock for 10 years. You may not be aware of this, but every reasonably sized company employs data scientists and analysts to do research on the company data. They use immense servers, usually provisioned on AWS or the like, that are accelerated using GPUs. The use Nvidia GPUs almost exclusively. Why? Because the libraries Nvidia has developed to accelerate machine learning algorithms are the best and it's not close.

The demand for accelerated servers that these companies use to create value is increasing at a rate much faster than any bitcoin/gaming/CAD demand.

Nvidia is valuable because the market is pricing in the growth, demand, and requirements for more nvidia cards. But they aren't used for gaming anymore. They're used to fit random forest models, neural network models, and yes LLMs.

So to sum up - nvidia has the best hardware acceleration and it's not just for gaming and CAD. It's what every medium to large business is using to do analytics, and the demand is growing at an exponential rate.

>LLM AI is currently all about potential applications, with no actual profitable business existing today

This is just a completely ignorant statement.