Yup, precisely. Turns out getting AI to be reliable at doing useful things is harder than we've all been led to believe by the dominant narratives.

https://www.normaltech.ai/p/new-paper-towards-a-science-of-a...

> Lets redefine work into discovering the world again. Let people do old handcraft jobs, let them do more sports, let them read more, let them write and make more. Let them enjoy nature.

Why leave something so important up to what AI does or doesn't do?

Because capitalism doesn't allow for that.

Only a fundamental change to our society will allow this for the masses when pressure to the rich skyerockets

> I certainly don’t think individual cells have any subjective experience or sense of agency.

There's definitely research and scholarship that would beg to disagree with you there. At least in terms of completely writing off the notion of "agency" when it comes to cells.

Dr. Michael Levin's lab is doing some pretty cool work. https://drmichaellevin.org/

> something like the universal approximation theory comes to mind, transformer architecture clearly has the shape of a universal algorithm approximator

That says nothing about emulating a human brain.

is human cognition not an algorithm? is it just woo? or vibes?

But what is the shape of the algorithm of the human brain? It has a complex physical structure. We know the folds on the surface are important, but why is that shape specifically important? The brain is made up of two hemispheres - why, what does that do? There are different "types" of brain inside the human skull. There are physical areas that perform specific tasks. There are different types of neurons. Then there chemicals that interact with the brain, changing how it function depending on things happening to the body. All that stuff and more is the "algorithm" of the human brain. It's not the same algorithm as an LLM.

I'm hearing different from PhDs. The bottleneck with much research isn't "trying out ideas" so much as it's all the bureaucratic minutiae, grants, mentoring PhD candidates, collaboration with other researchers, etc.

I've heard LLMs can be helpful in limited targeted ways. But not as some kind of "game changing" accelerant.

Understanding in what ways it can be useful and in what ways it can be counterproductive in long run requires a certain degree of experience itself.

I call it "I'm not like other girls" writing.

> 4.8 is also 2x more expensive for a "modest" performance bump. How refreshing.

Where are you seeing it's 2x more expensive? https://platform.claude.com/docs/en/about-claude/pricing

Don’t measure model cost by token price. Measure based on tokens used to achieve a task.

Others report in this thread that it’s about 2x more expensive due to outputs: https://news.ycombinator.com/item?id=48312774

I'd be kind of shocked if a model that came out six months ago is the same size and cost to run as one that just came out today.

Same size? Maybe by a bit. Cost? Absolutely. Newer flagship models are often slightly larger each generation, but not even 2x. But more efficient architectures are coming out all the time, and it'd be a waste to retrain an old model. So it washes out.

Well, it seems like collectively we are all struggling to perceive model progress, given that it seems like every reply to you is reporting different experiences with which of the models has subjectively performed best for them.

I have my original gmail address from 20 years ago still and even old youtube videos my friends and I uploaded from ~18 years ago.

The cringe is rough but at some point the cringe becomes so bad it loops back around to me just feeling nostalgic and grateful that there's proof I was able to do things, create, be silly, whatever without worrying about appearances so much.

Also, I figure if I ever become a megalomaniac then old youtube videos of my teenage self doing parkour should go pretty far in humbling me (although, honestly, I think 13 year old me was way cooler than I am now, so I guess it could backfire).