nathanfig

Also liked Fast.ai, although they sometimes use Fast libraries instead of PyTorch, it does focus mainly on PyTorch concepts.

Relative term. In the world of LLMs, 7b is small.

Was just about to comment this. Wonder if anyone more mathematically inclined can see a relationship.

Deno Fresh uses Twind, which avoids the build step: https://twind.dev/

But IIRC it doesn't offer quite everything full Tailwind does. In general I was frustrated with Fresh for not being up-front about the fact that it is largely built on Preact and Twind, and you must buy into those libraries first.

OP is excited about the leap in capabilities, that is all it means. You can be curious about the inner workings of alien magic too.

"memorize" implies they can only recite things verbatim and that's ignoring the massive leap in being able to synthesize disjoint "memories" in new ways.

Whoa, was this for multiple parties or a month long stay or something?

Most households with 4+ people are going to have dozens of connected devices, including multiple doing HD to 4K entertainment streams, outgoing video feeds, and (at least on PS5) the occasional unexpected 80GB updates to a video game you were just about to play. This in addition to the apps/webapps that are often in the 10s of Mb of content served per click.

Ehh, I think platform has been a pretty squishy term for a long time already so it's not one I usually dispute, even if I'm inclined to agree that this project is more of a workflow at this point. ROS itself is not an Operating System as the name implies; middleware is probably more accurate.

That is cool, and especially useful since LLMs often get details like URLs to repositories wrong, so I could see how this system would increase the accuracy of the output.

Upvoting this because I love innovation in this space, and I use both ROS and LLMs daily. Kudos on the work here, I can see value for developers trying to break into ROS in particular.

That being said, and I might just need to try it out to understand, but it's not clear to my why this tool would be better than just talking to ChatGPT directly. Particularly GPT-4. There's the guided workflow aspect, but even that I could probably emulate by copying the description from this article and asking GPT-4 to do the same. It might just require more manual copying and pasting.

Not really. The ocean serves as a natural carbon sink to a degree- but this is acidifying the ocean. And some feedback loops accelerate rather than decelerate warming- such as ice loss decreasing albedo.

Life on Earth can adapt to a new normal, certainly, and on a geological timescale normal has shifted many times. Humans will adapt too, but will likely endure catastrophic losses on the way.

So with no power to predict the outcomes of my actions, or even predict how I will pick an action, I bear witness to myself acting anyway- and this we call free will?

What troubles me is that the more we take away any determining factors for free will the more it sounds like randomness.

Not sure this allows a meaningful definition of free will, since if nothing can predict me neither can I.

Can you explain how this creates free will?

Interesting point. Automated bans with no recourse will mostly deter legitimate businesses. And there is a finite supply of legitimate businesses, but infinite fake ones.

Often wonder how much life we've encountered without recognizing it.

Right, this just seems like adding an alternate definition of life without moving the morphology problem.

Survivorship bias, you can't count how many fossils you did not recognize.

2.5A 5V is not much power, and it would use considerably less when idling.

You're probably still using the default, there's a dropdown when you hover over the GPT-4 button at the top.

That's just the introductory rate limit. Be My Eyes has a closer relationship with OpenAI than most of us https://openai.com/customer-stories/be-my-eyes

"... lossless compression can be a scorecard of prediction which is the same as AGI."

Having not read the papers, this sentence strikes me as a bit of a leap. Maybe for very constrained definitions of AGI?

I think HN will eat this up because few of us enjoy being sold to but like most absolute statements it's probably only applicable in specific circumstances. Enterprises sell tons of developer tools and move millions because of demos done well. AWS and Azure are definitely getting on planes and doing demos.

I like having type hints, I really do- but I've definitely been on projects where TS began to feel a lot like a nerd snipe https://xkcd.com/356/

Thanks!

Had to scroll past a ton of doomsayers and jokes to find this obvious question. I'd like to understand the decision before speculating the meaning.

Spanish and French are both romance languages and will have massive token overlap. Not likely to be so lucky with whale songs.

If you can find a way to tokenize whalesong then you could probably build a next-token predictor but we're still missing the Rosetta stone necessary to help us map their language to ours in vector space.

I think about this a lot with regard to software accessibility. We easily fall into this trap of thinking that it's only needed for a small minority of people- but all of us will age and lose the dexterity and senses we rely on to use computers and phones, not to mention that capacity to learn quickly.

Not that a selfish motive should be required, but prioritizing accessibility features today is very likely to serve you tomorrow.