I still don't see any data from school maximalists that schooling is anything more than signaling (beyond basic math and literacy), and that kids wouldn't be much happier if they could spend time on projects and pursuits they were passionate about instead. My son is 6 and in 6 months of a Minecraft obsession he's fluent in English (besides Swedish and Chinese), learning about all sorts of concepts like animal breeding, crop rotation and automating factories and farms and building all sorts of machinery. Meanwhile school is teaching one letter a week. I can guarantee that if you put kids like him together with a bunch of other kids with the same passion and let them go wild building stuff with some guidance from a technically competent adult, they would come out much more prepared for life at 18. Fine if your kid has no curiosity or internal drive to do anything, but then I question what they will get out of school anyway.
People take "Full" as meaning L5, but Tesla uses "full" as in ODD (Operational Design Domain). It can go anywhere, city streets, highways, parking lots, unmarked roads. In that sense it is indeed "full". This is clear when you look at the history of Tesla autonomy products, first there was Autopilot, which is only for highways. Then they release Full Self Driving Beta, which includes every type of driving.
Awesome! I've been wanting to make something like this that I can use with my son and nieces, but haven't had the time. This is exactly what I've been looking for. I expect we'll be spending all summer making little games together!
Or, it's the type of job where a degree doesn't guarantee that someone is a good coder, and that's easier to test with a whiteboard than other kinds of engineering for instance.
Yes, what I meant was deep learning is great at deriving those higher level abstractions from low level raw data. Words can be seen as something in between, bag of words can be fairly effective at simpler tasks, but LLMs embed words into higher and higher abstractions.
Deep learning really shines when the input is raw and at a very low abstraction level: pixels, byte pair encodings etc. Using deeply learning for classification on tabular data is just needless complexity, as the variables are often at a very high abstraction level already. Also with tabular data there are generally not many spatial or temporal relationship between the variables, which CNNs and transformers excel at.
Great thing to do when you're young, why not. I slept on a mattress in a closet for a year. Standards increase as you age, not just because you can but because you get increasingly tired. I held out as long as I could, sleeping in cheap hostels when traveling, but after one night of drunk people coming into the dorm causing days of fatigue I realized it's time to up the standards. Or just not travel.
My point stands: just because Tesla chooses to make FSD an L2 system at the moment doesn't mean that it's unable to act safely in the same scenarios as for instance Mercedes L3. So concluding that FSD is behind Mercedes is delusional. If you actually look at evidence available, (the thousands of videos available for FSD, and the ... tens available for other brands), and still claim FSD is behind, that is delusional.
I'm defending Tesla because these comments are just insanity, and it points to an extreme anti-Tesla or Elon bias, which is very common these days. If you can't impartially evaluate a piece of technology that is clearly more advanced in every possible way, you really need to reflect.
That would have an acceptable point to make if FSD was not capable of safely doing what all other ADAS systems can do, but it obviously can. It's just that Tesla has no interest in restricting its features and get L3 approval for such a small subset of driving.
If today you claim FSD V12 is "behind" the lane keeping assist of other manufacturers, that is just delusion, I don't know what to tell you.
As a customer I can't really say that I can tell any difference between all the streaming services, so presumably this type of optimization helps Netflix's bottom line somehow?
Then what are you claiming the difference is? Are music labels inherently less greedy? Is it harder to build music streaming vs video that leads to a monopoly?
Is it because labels have ownership in Spotify? I don't think the HBOs and Disneys will give up their streaming platform unless forced to do it.
I think TV and games are meaningfully different. Most people tend to watch way more number of shows in a year than buying/playing video games. Video games have more replayability, so it's something you need to "own" rather than stream. Therefore, subscriptions for games are not a big thing yet (I doubt it will ever be). As it stands now, if a game I want is not on Steam, that's OK, I can just install another game store. It's a bit inconvenient but not too bad. The problem with TV is cost. I'm OK switching between Netflix, Disney, HBO, Youtube, whatever, but I'm not OK with paying $70 a month,
> The services have taken the piss and now they'll get the repercussions of it.
Whose fault is it really that all content is not available on a single service? It's nobody's fault, it was inevitable given how things work.
How could you "fix" it? You'd need legislation saying that all content must be made available for licensing at the same price for every service. But that's not enough, because why would HBO sell their content to Netflix and vice versa? No, you'd have to force them to make it available. But that's not enough either, as they could just set the price extremely high. And at this point what would be the point in having multiple streaming services at all? Perhaps we need legislation that enforces separation of content producers and distribution, like the Glass-Steagall Act but for TV, or Net Neutrality?
I've come to conclude a rule of thumb for (side) projects: if the project includes any form of web app, that part will take 90% of the development time and energy. Since I work in ML and not web stuff (which I don't really enjoy), this is not really how I'd like to spend my time.
Unfortunately, almost every idea I have, even if it deep down has to do with AI or ML, includes some form of web app, because you need some way to interact with the interesting thing. In the end, maybe that means just doing nothing, because with kids and a busy life I just don't have the energy to spend on npm dependency management, front-end frameworks and JS build systems.
That observation is completely pointless. Yes, even at 100% adoption it will take a long time to replace the whole fleet of cars, does that mean we shouldn't be doing it? Just the fact they bring this up shows they have an agenda.
Tesla is also recognizing this fact and it's one reason they're pushing for robotaxis, because robotaxis can increase utilization by 5x, the same as selling 5x as many cars.
Steam Deck OLED. Runs much better than my 3 year old Dell laptop that cost $2k, for $500. Steam for gaming is great value as I already have a library of games, and you can play pretty much the entire switch library as well. Simple enough for my 6 year old to use, flexible enough to install mods and whatever. Don't do much gaming anymore due to life and kids, but this device makes it much easier to get started when I do have time.
"A feud over the direction of OpenAI led Musk to step down from the board in 2018"
So, voluntary departure, but it's clear that Sam won some kind of battle over the future of OpenAI. Not sure what their differences were, but Musk has repeatedly commented how not open OpenAI has become.
People hate Musk because he says everything that comes to mind, but it's people like Altman that scare me, because he's actually good at this game and it's difficult to get a sense of his true motivations.
I'd be interested in the real reason Dario et al left to form Anthropic, it must clearly have been a misalignment with Altman's ambition. It seems likely that Altman is more of a Steve Jobs type, perhaps even better at politics, maneuvering and fund raising. He managed to get rid of Musk and turn a non-profit into a $90 billion company, very impressive indeed.
Easy to shit on Ilya right now, but based on the impression I get Sam Altman is a a hustler at heart, while Ilya seems like a thoughtful idealist, maybe in over his head when it comes to politics. Also feels like some internal developments or something must have pushed Ilya towards this, otherwise why now? Perhaps influenced by Hinton even.
I'm split at this point, either Ilya's actions will seem silly when there's no AGI in 10 years, or it will seem prescient and a last ditch effort...
Tesla is running these models on 4 year old hardware to control a car in real time (30 fps). You don't need a full 100B model to control a drone, and it doesn't have to be as good as a car to cause a lot of damage. Reportedly both Ukraine and Russia are putting together on the order of a thousand drones a day at this point, Tesla includes the compute to run this in every car they make already today. Hardware is also moving fast, how come people forget about Moore's law and software improvements? To me there's no question that this tech will be in tens of thousands of drones within a few years.
Open source models in general. Meta has for instance released DINO which is a self supervised transformer model. LLMs are also going multi modal (see LLaVA for instance). The name "LLM" has stuck but they should really be called Large Transformer Models. LeCun is working on self supervised visual world models (I-JEPA) which if successful and released could form the basis for killer drones. It's still a lot of engineering work to fine tune and put a model like this on embedded hardware on a drone, but at some point it might be easy enough for small groups of determined people to pull it off.
I'm not a doomer but I honestly don't understand this argument. If releasing model as open source helps researchers determine if it's safe, what about when it's not deemed safe? Then it's already out there, on the hard drives of half of 4chan. It's much easier and cheaper to fine-tune a model, distil and quantize it and put it on a killer drone, than it is to train it from scratch.
On the other hand I totally relate with the idea that it could be preferable that everyday has access to advanced AI and not just large companies and nation states.
The whole team that trained GPT1-3 and came up with RLHF left to found Anthropic, perhaps proving that key people are maybe not so key and that other factors are more important. This is a further blow though on top of the previous one.