deegles

not being snarky, but what is the point of using the model if you already know enough to correct it into giving the right answer?

an example that just occurred to me - if you asked it to generate an image of a mushroom that is safe to eat in your area, how would you tell it it was wrong? "oh, they never got back to me, I'll generate this image for others as well!"

I guess something like a kickstarter campaign would be needed to get together the millions of dollars needed per training run

Or Elizabeth Holmes....

I still miss Google Reader.

you could learn bookbinding and print out your annotated books. i feel like it would be a nice hobby.

My aunt is in her 70s and has 5 friends currently with cancer. My belief is that it's downstream effects from Covid wrecking people's immune systems and their ability to naturally fight cancer. Time will tell.

I live thousands of miles away and I can't help but wonder what natural disaster will hit me, eventually.

I still find it hard to believe that LLM methods will lead to "true" AI. No amount of processing power or data will be sufficient without something new.

LLMs as they currently exist will never yield a true, actually-sentient AI. maybe they will get better in some ways, but it's like asking if a bird will ever fly to the moon. Something else is needed.

I think people are laying the groundwork for "true" AI to be plugged in. even if they know it's not currently that effective.

The big question is if OpenAI will achieve "general" AI before their investors get fed up. I wonder if they used the success of ChatGPT to imply that they have a path to it. I don't see how else they achieved such a high valuation.

I would trust Microsoft with my voice more than some random startup.

I have tinnitus and can't hear much in a range around 4.5khz... I wonder if this would still work for me

My big question is what is being done about hallucination? Without a solution it's a giant footgun.

I think a few hundred purpose-built nuclear powered CO2 scrubbing plants could be an option. At 1,500 kWh/ton, a 1 GW nuclear plant could capture 5.8 million tons of CO₂ per year which is about 0.015% of global emissions. It would have to be coupled with emissions reductions and so on though.

Does it do diarization?

I think "LLMs hallucinate facts less often than people" is a strong stance to take at the moment.

I was amused to see this next to the thread about Ozempic on the front page

If they can solve the hallucination problem, sure. Otherwise we're going to have a lot of people who "know" things about the world that are simply made up.

We just have to look at the recent news about 23andMe planning to sell all of its user's data to whoever buys it. Sure, they can have a privacy policy and all that, but does it really matter when the company can be sold and they say "all previous agreements are null and void, your data is ours now"?

Just try a few models on your machine? It takes seconds plus however long it takes to download the model.

You're not just memorizing text though. Each piece of trivia is something that represents coherent parts of reality. Think of it as being highly compressed.

Nearly everyone's data has been leaked already. Any strong protections would only protect people who haven't been born yet imo.

I was thinking about what "actual" AI would be for me and it would be something that could answer questions like "tell me every time Nicolas Cage has blinked while on camera in one of his movies".

Sure, that is a contrived question, but I expect an "AI" to be capable pf obtaining every movie, watching them frame-by-frame, and getting an accurate count. All in a few seconds.

Current models (any LLM) cannot do that and I do not see a path for them to ever do that at a reasonable cost.

and they make us feel bad for using plastic straws...

This project had one goal: get a high visibility project out the door in order to secure a promotion.

Texas is the America of America.

I think it's fair to take a fee for payments through Apple's infrastructure... but 30% is egregious.

The issue is liability, just like self driving cars. Even if ML is more accurate, when it does get it wrong the buck has to stop somewhere. Those questions need to be answered first.

Well it's not too complex. Say $25k for the machine amortized over 5 years is about $400/mo. Let's say $500. Cost of goods would be around $7.50/lb for some nice unroasted beans. Let's say you sell for $10/lb. Fixed costs would be rent ($5k?), utilities (1k?) and whatever other costs to package etc. not even touching salaries. So at $2.50/lb revenue you'd have to process and sell 2,400 lbs per month just to break even on fixed costs. That's a lot...