petra

LoL.maybe the question is not phrased in the best way, English as a second language and laziness, but it's a pretty decent question.

Given LLM's is being smarter still a worthwhile to optimize with supplements?

I wonder what perf. this cpu can get in bit/ternary inference, and whether the 512AVX units(or whats it's called on AMD), get throttled.

You're right. I've used reader mode(the paper in the url bar in Firefox) to solve this.

That's the current Netanyahu and right wing extremists government. Nothing good can be expected of them.

Elections are going to be held soon. Hopefully this government will be replaced, and better things start happening.

There's also digital probabilistic computing, with a processor working on AWS, and for the right workload ,claimed to be offering a very significant speed boost:

https://www.eetimes.com/probabilistic-computing-is-already-h...

Towards the end of the article.

We already drown in useful new things and especially knowledge. There's a limit to how fast we can adopt them.

Llm's do have discipline and persistence, though.

The ability to correct the plan/prompt.

True, that's probably the biggest issue.

But the only solution here is very expensive marketing, so...

I agree. The market doesn't work well around this.

You can ask chatgpt.not a great way.

And when you ask it you the secure answers cost 3x more. And than require an installar. And some(Google) require a monthly subscription.

And even about the good systems, the chat recommends, since there's no mathematical guarantee for security, that you "Switch them off or physically cover the lenses while you are home.".

Consumers just don't care about security. It is what it is.

No pricing, "call for demo". Not sure this is for personal use.

Or maybe just a way to use chatgpt/Claude with files , and get accurate references to the page/paragraph in the source?

I think the next popular web framework would be something that would be optimized against llm code generation. So that the end result would be secure. correct. scalable(and scaling should be done by an llm).

That's an interesting question.

What happens if we take the most abstract libraries in any given field - and:

1. Bound to the llm to only use those as building blocks. Does it affect his reasoning ? Will it think more abstractly ?

2. Train the llm on those, so maybe it will get a feel for abstraction ?

We know there's a big problem with addictive ux in general.

And of course we know that small changes increase engagement/addictiveness.

Html interfaces are easily configurable, technically.

The fact companies at least don't offer easy ways to configure websites to be less addicting, and some even block those, does tell us something.

Because using the other methods of learning, possible coupled with having a search environment inside the chat for the future, or creating memorization material for the stuff i specifically want to memorize - seems more efficient and effective.

For me, as an avid reader of non-fiction books, for learning, i'm starting to question the value of reading them, compared to a good in-depth discussion with an LLM about a subject, together with reading academic papers and long articles/blog posts.

There are plenty of businesses that could be built under current regulations.

It might be a bit Harder, but business is difficult.

I wonder if they can at least create a small scale NBR factory.

In cases of emergency, like COVID, is butadiene supply limited or for the right price the US, and in general, the world, could get enough butadiene ?

I like the effort.

It needs some improvement.

Elon musk is among the top of the list. He is also the founder of companies that created and advanced a lot of technological wealth in the world. A huge contribution.

But it's far from certain that the recent SpaceX stock will create a lot of wealth for retail owners. Maybe even the opposite.

- the extent to which solutions could be implemented as text: not sure about that. AlphaFold is basically a mechanical/geometrical/Chemical problem. There are other scientific transformer based models.

- the extent which solutions exist online - if you have a strong verification tool, you can generate examples, you can generate feedback, i think you could start with small/smaller prior art

- the extent which solutions could be specified and checked - if you have a lot of priort art, maybe llm's can find the good "patterns" and compare against them, and at least get close to a good results - but you'd still need human verification.

I use Facebook mostly for groups.

Facebook groups have destroyed online forums and with that killed long discussions on the internet.

But sure,somehow Facebook doesanage to suck you in and wate time on bullshit. For that it's awesome.

The story about the contribution of Bell Labs' patents to the world was exciting. And the benefit was certainly much greater than the extra amounts people paid to bell labs.

In 20 years, thinking about llm's contribution to new technologies, to improved accessibility of valuable knowledge, to solving problems.

How would you tell that story?

IDK. i played with Fable. It's coding abilities are really impressive.

This actually feels like something that could be a popular app on a baby monitor.

A few reasons: -2.5 years is a pretty short time for a new tech development, even if it fails eventually - usually when a new tech is introduced, the biggest gains happen when the environment is changed to fit it. That takes time: libraries, api's, verification tooling, rl environments, skilling users, etc. - possibility of orders of magnitude hardware cost reduction - Optical. Analog. Rram. Many others. Something will work. And internal improvements in the model architecture. And there's scaling in reasoning time.

No idea. I'm not from the US.