They're not saying it's urgent; they want to bring urgency - convince people to move faster.
It seemed appropriate to ask AI about the meaning of the phrase:
The phrase "bring urgency to something" means to inject a sense of importance and immediate attention to a specific issue, project, or situation. The aim is to motivate people to prioritize the task at hand and to act more quickly than they might otherwise.
I can’t go into specifics for obvious reasons, but I can tell you they’re a smart, capable crew, with lots of resources, and time is on their side. If I had to place bets on which lab will invent agi, they’d be near the top, precisely because they’re exploring ideas that others don’t seem to be focusing on. It’s the early stages, so it’ll take a few years to get to results. But they’re laser-focused.
There are a bunch of things I thought were cool. John has a worklog going back four years or so, with entries practically every day. He’s nothing if not thorough.
Everyone tends to focus on John, but the other members are just as professional and dedicated. They’re a delight to work with. Also the investors are worth mentioning — Nat in particular has built a super cluster of H100s for startups to use, and has done a bunch of other work to support AI.
If I had to place bets on which lab will invent agi
I'd bet against "invent AGI" being a coherent or agreed upon thing, and I'd also bet against the "wake-up" scenario of a system saying "I'm AGI"
Which lab invented deep learning? Certainly U Toronto had the breakthrough that made a 30+ year old field exciting, but there was a lot of work in Montreal, NYU, Google, DeepMind, OpenAI, and many other places
Hell,LLMs can and do output strings of "I'm AGI", so in some sense we're long past that. The problem is as humans we've never defined what it means for a system to be AGI.
My back of a napkin definition is that AGI, given the same information and tools as an IQ 100 human, is able to do all of the same tasks that you could ask a human to do, at the same quality level, as well as recall the experience some time later, like a human would likely be able to do.
The particular issue here is that is subjective as any other subjective definition. 100 IQ is subjective and a moving target. 100 IQ doesn't even mean you can survive and navigate in the real world. Even worse is all humans cannot do all tasks other humans can do and we'd consider them intelligent.
Furthermore this "must behave like a human" blindness represents both a danger, and a way to miss intelligence capabilities leaving them unused/under capitalized. Humans behave like humans because we have human bodies with particular sets of strengths and weaknesses. Other 'intelligent' actors with different embodiment will necessarily behave differently.
Couple that with humans have become far more intelligent and stronger by using tools, that the ability to 'universally' interface with external tooling, at least in my subjective opinion is what will be the defining factor of higher intelligences.
Didn't Schmidhuber's lab predate all of those in their inventions except ReLU (which I've heard is now out of favor as compute outpaces memory bandwidth and other activation functions have better properties).
Sounds like a -terrible- job for LLMs, because this is all about attention to detail. Order of operations and specific constructs of how floating point work in the codes in question are usually critical.
A human has to have the knowledge of what the code is trying to do and what the requisites are for accuracy and numerical stability. There's no substitute for that. Having a translation aid doesn't help at all unless it's perfect: it's more work to verify the output from a flawed tool than to do it right in this case.
> Why not do the same thing credit cards or access smartcards do?
I think you would be surprised how bad the security on these systems is.
The credit card security relies mainly on the ability of the bank to rollback in case of "a shit happened" and in the payment terminal itself.
Probably not something you want to see to protect against identity thief nation wide. And you also can not trust individuals smartphone to do the right thing.
I think this is not true in most of the cases. The (security) technology behind the debit/credit cards using the SmartCard chip (IC) is pretty ubiquitous. It is the same as the security technology guarding the SIM cards in your phone and even your eSIM. Basically the protocols and the interface specifications are the same. In the end, they are just smart cards. Imagine this technology not being strong enough, because I remember the days when the security of the pre-paid public phone cards was quite rabish and any kid with some skills and knowledge could forge a card with unlimited credit.
It very happens that the father of the smart card technology to be a french guy [1] and the current biggest provider of this technology is the french aero-space/defense/security company Thales Group[2] followed by another frech company called IDEMIA.
There is a very nice biography of the technology [3].
> The credit card security relies mainly on the ability of the bank to rollback in case of "a shit happened" and in the payment terminal itself.
That's certainly not right, the electronic chip in a payment card relies on cryptography, and if you used than together with a PIN the bank has a strong argument to not rollback anything. If you're using a debit card like most of Europe, you're going to have a hard time convincing them.
thanks! I am still curious about the sensor inputs part.
I am trying to replace the 2.4ghz controller on my electric skateboard to make 0 to 5kmh and braking more pleasant and maybe use gyroscopes to do away with the controller altogether. What would tokens be in that case? Do you create a CAN style representation and feed that to the llm? What kind of throughput do you foresee being possible on which hardware?
In this case, I would ask the LLM to suggest an algorithm to minimize acceleration, jerk and snap based on the expected sensor input data, and then just implement that. Probably in memory on whatever runs the board.
Straightforward control problem of bringing the board from 5-0kmh smoothly?
Basic control theory will work better than AI here. The mathematical models used in control theory have been used in computing since at least the 50's (Kalman Filters). I suspect you won't have issues with computational power.
Figuring out exactly which model to use and how, may take some work. Also understanding control theory will allow you to do things like traction control, etc.
So how should a company that makes tech that people want while being as environmentally friendly as possible present itself? What other company is more environmentally conservative while making the tech that people want (for example Xiaomi, Huawei, etc) than Apple?
"Don't make it" is not a good answer. People buy it, so someone will make it.
And I'm happy that Apple educates their customers about the environmental impact.
"People buy it, so someone will make it" doesn't apply when they're at the size of Apple. They absolutely make things that no one else could make if they didn't.
I disagree completely. People wanted this for a long time, now it's available, that's it. People usually claim that Apple didn't invent anything new...
>So how should a company that makes tech that people want while being as environmentally friendly as possible present itself?
Not with self indulgent high budget folly like ‘Mother Nature’
I never said don’t make the objects I actually said I’m fine with how abhorrent a 3 plastic battery coffins that are AirPods are for how good the experience is.
Just don’t make marketing material like ‘Mother Earth’ when you’re shipping those products.
Anecdote: every single woman I talk to is afraid of AI and robots.
I don't see any data about how many AI startups are female-founded in the article. Perhaps just 2%?
And what's the correct ratio of investment into female-founded AI startups? 50%? What about all the other genders? According to Wikipedia, there are dozens of gender identities: https://en.wikipedia.org/wiki/List_of_gender_identities.
It's more like "we made a UI framework that speeds up development and runtime of complex apps considerably, though it also slows down initial load of web sites... but don't worry, we have a fix".
Development? Yes. Runtime? Highly debatable. At a bare minimum there are other frameworks with a significantly better runtime performance story than React.
Because they could have moved to a server based app not using RSC and seen similar gains, so the deception is they are trying to make it seem like RSC caused the gain (and riding on the back of a hyped new technology) whereas it's not really the case.
But RSC is how you do server based apps with React, and they’re clearly sticking with React.
Sure you can render to static markup without RSC but then you have to do a lot of work to mix server/interactive client components, RSC makes it a lot easier to achieve “server based apps”.
I think it’s less work to just use static content in places and avoid having to learn a whole new model of rendering, routing, data fetching and a lot of rules than it is to have a traditional ssr app that compiles a few static pieces of content.
RSC is how React figured out partial hydration. Before this, you would simply not move to a server-rendered app, as it was not worth it - the client bundle would stay the same and you're just increasing overhead.
Not true in the slightest. A server rendered app would deliver your html up front so your app is completely rendered immediately, this has been common for many years now and improves performance quite a bit. Couple with sending some static parts that avoid JS which is quite doable you have 90 some % of the benefit in a simpler model. That’s also why more direct a comparison of a SSR app to a RSC app is interesting.
Partial hydration is great but it’s a shame React is coupling it to a whole new routing model, component model, server rendering, and with such complex rules on use and inherent waterfalls as a side effect. There’s no reason to. You can get partial hydration any number of ways and avoid all the complexity.
> Couple with sending some static parts that avoid JS which is quite doable
If when you say "they could have moved to a server based app", you are implying that they could have done partial hydration easily some other way, that's a wild assumption. Until very recently partial hydration was not even on the table, as the community decided that downloading 10MB of JS for every website was fine, and not 'quite doable' for the majority of React apps; they are usually monolithic and built on top of frameworks that make it difficult. Not impossible (we were doing this back in 2016 with different tech), but very uncommon.
As I said above, this is exactly what RSC is solving for. It's not a new thing, but the first solution endorsed by React itself, and enabled them (and potentially all React apps) to gain SSR with partial hydration, reduce bundle sizes and actually improve load performance.
I'm not a big fan of any of this, just stating what it is. And I agree that it introduces monstrous complexity to an ecosystem that's already overflowing with it. Astro for example achieves this without introducing too many new concepts.
Totally agree that partial hydration helps more than full hydration, but note that if you compared SSR vs RSC the charts they posted wouldn't look so impressive.
Maybe I was too strong worded, but the point stands that without making it really clear they are "skipping a generation" basically, it makes their comparison look extremely favorable, whereas RSC actually has a big overhead: it serializes all props and the entire tree that's not hydrated into an arbitrary serialization format that JS then needs to parse and hydrate. They actually are far from as zero-cost as other partial hydration methods. And if you had compared SSR vs RSC you'd see that the numbers don't actually improve so much.
So is it like 10/10 deceptive? No, but it's also a meme thats going around and not the first article to try and gain attention by conveniently leaving out of the title and most of the article that 80% of the gain they got was just going from client-only to server-rendered, and not actually RSC. I absolutely stand by it being deceptive, if only on a more annoying than actually harmful level. The graphs aren't comparing SSR to RSC, they're comparing the (comparatively much, much worse) client-only to RSC.
> 80% of the gain they got was just going from client-only to server-rendered
That's what the article is about, isn't it? The gain comes from going from client-only to server-rendered. They did this using RSC. You could also do it using handlebars and jQuery if it was 2010 (with probably 100x better performance).
> if you had compared SSR vs RSC you'd see that the numbers don't actually improve so much
From the article: We saw a whopping 62% reduction in bundle size as well as as 63% improvement in Google's Speed Index. You would not get those gains by simply enabling SSR in an existing app, as the bundle size will remain the same. Unless.. you use Astro or RSC to enable SSR with partial hydration. Which is what they did.
No, speed index would go up massively. A huge amount of lighthouse improves with server rendered content.
Likewise RSC is not nearly as free as other partial hydration methods. You send a full serialized copy of the tree and props over that has to be manually de serialized. This nullifies much of the benefit. Further, you still send the large (even larger now) copy of React. This is exactly why the comparison would be interesting to see, and why I brought up my concerns - it seems people are pretty under informed about this.
I agree that's a bad choice, but it doesn't change the fact partial hydration was not available until very recently. Without Astro/RSC, you can either have SSG / SSR with large bundles, or an SPA with large bundles. These are the options.
It seemed appropriate to ask AI about the meaning of the phrase:
The phrase "bring urgency to something" means to inject a sense of importance and immediate attention to a specific issue, project, or situation. The aim is to motivate people to prioritize the task at hand and to act more quickly than they might otherwise.