It used to be possible like showHN posts go on to become smashing success. But Dropbox like posts seem like an impossibility now.
I've been having serious mental crisis from this realization.
Why does AI make you feel this way? It feels the opposite to me — like meals that formerly required a master chef to make, but soon anyone can make for themselves at an acceptable level of quality with meal prep kits
Not sure if that's changed but ever since seeing that line, I've been put off of that world. I still occasionally click on HN links advertising new methods which can be run on "consumer" cards and every time it's just a 3090/4090...
I don't have that much money.
The p40 was on the inference side of the tree and you can get one on eBay for $200 or less.
But when I got into software in the 90s, it was about $4000 for just a run of the mill desktop PC, which is about $8000 in today's money. And it didn't even have 3d acceleration.
In the grand scheme of things, a 3090/4090 is not expensive.
You don’t need thousands of dollars to start cooking.
I have friends in the who cook on this stuff at home all the time. We in the west have elevated our kitchens into something luxurious. Nothing wrong with that, but it’s overkill for many people.
Yes writing a todo list app is now faster with copilot and stuff but thats now what i meant.
Products that we build today represent many millions of decisions and trade-offs.
If you believe something should exist, please build it and don't worry about what anyone else is doing or saying.
I was working on an application in the early 2000's and learned that Gigantic Inc. launched something to solve the same problem. I immediately bowed out, and they immediately let that first-launched beta languish for more than 10 years with less staff than we had.
In the intervening years, another startup built a similar competitor and sold for hundreds of millions of dollars.
As another example, I eventually worked at Gigantic Inc, and the org that I was part of had been failing to deliver a useful product to the public for about 7 years, and continued to fail for many more. This was an organization with hundreds of people and many hundreds of millions of dollars of budget, and they were being absolutely clobbered by a combination of their own unbelievable incompetence and the brutality of the market around them.
One of the biggest lessons I've learned in my career, is to never assume that just because Big Tech has some budget, that they also have attention or competence.
They do? How so? I can only think of a small subset of applications that require large amounts of compute.
Try to satisfy yourself, and maybe that will lead you to a commercial kitchen with a Michelin star. I know nothing else will.
But there's a thriving community on HuggingFace and Reddit showing what you can do with the lo-fi versions. In particular, the evolution of lower bit inference (and I believe training as well even) has reduced memory requirements and because of that hardware requirements considerably. There is a lot you can do with your own local gen AI model running on your personal machine.
Now it feels like I have these incredible capabilities to apply LLMs in novel ways. I've had a personal project (an expense tracking app) where now I can see a path to easily do things like scanning receipts accurately, automatically categorizing CSV imports, etc. They were always within the realm of possibility - but would have taken so much more time to build as a one man shop. And we are just at the tip of the iceberg right now.