snek_case

Yes, SSRIs trigger change in the brain. That's why the effect is not instant when you start taking the pill, the change needs some time to start to happen. In particular, it's believed that depressed people have lower levels of neuroplasticity. Their brains are less able to adapt to their current circumstances and they're stuck in negative thought patterns. SSRIs help increase the brain's ability to grow new connections and help the brain adapt, which helps you get out of negative thought patterns.

If you're interested enough to watch a whole talk about this (you can jump to around 26 minutes): https://www.youtube.com/watch?v=k2AdebIbx5Y

Actually, I don't think he's being pedantic at all. LSD binds to a wide array of proteins and receptors in the brain (including dopamine receptors) and the binding profile of this substance is most likely very different from that of LSD. Likely to be much more specific to 5HT2a, closer to psilocybin than LSD. IMO they just said "LSD-like" because it would make a nice headline.

See this diagram on wikipedia for LSD binding affinities (lower means more affinity): https://en.wikipedia.org/wiki/Lysergic_acid_diethylamide#/me...

They could run multiple games per GPU depending on the resolution and how demanding the games are, but yes, it's definitely more expensive than just serving a stream.

It seems likely here that since revenue numbers keep going up, they don't really realize that there is a problem. Why rock the boat and change anything with the culture when you can just keep collecting your bonuses?

With deep learning you'll eventually be able to just have a very realistic avatar that looks and sounds like a real person. Might eventually be hard to tell if people are using something like this or not.

The problem is that Google has a conflict of interest as both a browser developer and a platform for people to sell you ads. People selling ads to Google must be putting pressure on them to do this... Or maybe they have internal pressure to maximize revenues.

At the end of the day, yeah, it seems like it might cause people to switch. I'm also honestly wondering, how long until a real competitor to YouTube surfaces? I realize that this is hard to build (though we do have multiple streaming platforms in existence), but Google has been playing this slowly boiling frog experiment with ads on YouTube. Now you have multiple ads per video, coupled with ads baked within the videos themselves, and sponsored content. If you don't pay for YT premium, it's kind of unusable.

The worse the product gets, the more of an opening is created for the competition. Seems like a matter of when rather than a matter of if? What if Amazon or Apple created a YT alternative that was ad-free, maybe with some paid content to sponsor creators (pay $2 a month to get bonus content from this channel). They could afford to sink money into it and it could be very disruptive, but it's like they don't dare. Maybe they're afraid to associate their name with a product that could fail, kind of like Google Plus.

It's not just a matter of different modalities. It's still a matter of sophistication.

The end game is endless generative music or video streaming customized to your preferences. Being able to describe a story, or having the AI model take a guess at what you mind find interesting/entertaining and generating a whole TV show or movie for you to watch. Or generating background music while you work and automatically adjusting to your tastes as well as adjusting if you're finding it hard to concentrate or you need to take a call.

> It is hard to scam someone with an in person exchange of cash for goods.

You would think so, but some asshole tried to do this to me. We had agreed on a price ahead of time, and when he showed up, he tried to low ball me with a 50% lower offer. I refused to sell the item to him, and then he became aggressive and I had to call the cops to get him to leave.

I won't get into the details, but based on what I saw, I think this guy runs kind of a side-business acting as a middle man reselling tech gear. I think he does this routinely, probably low balls people and then tries to re-sell the same stuff on marketplace afterwards. I think tech gear attracts assholes/scammers more than anything else because it has a potentially high resale value.

So, just be careful out there. Never share your phone number with people from marketplace, and don't reveal your home address if at all possible. Try to have a friend over if you have people coming for transactions or do it in a public place.

Asking the right questions...

What bugs me immensely about this whole situation is that we can't really discuss it in public. I feel like if I dare express disagreement with current policies and predominant views, I could be fired. I could be publicly attacked and shamed, even for what seems to me like rational arguments along the lines of "let's take a step back and look at this problem from multiple angles".

Being attacked and labeled as a racist because you dare question these kinds of policies essentially amounts to gaslighting, and the idea that we're not allowed to express disagreement against these kinds of policies is well, censorious and oppressive.

I took a machine learning class in ~2006 and granted the prof wasn't specializing in studying NNs, but he seemed to have little awareness of the impact of dataset size, convnets were never mentioned, the suggested approach was two-layer MLPs and sigmoid activation functions. The course was based on the book "Artificial Intelligence: A Modern Approach" 2nd edition, which barely just mentions neural networks in passing.

I think there just weren't enough people looking at neural networks back then. At the time, they were still considered a niche machine learning technique, just one tool in the toolbox, and not necessarily the best one.

Also, it's one thing to say that the best of the best had access to computation, but if your university has just one supercomputer, which is shared among everyone, and you have a limited number of hours during which you can use it, that's not increasing your chances at running very thorough experiments. If anything, it's not that we didn't have the compute to do deep learning in the 90s and early 2000s, it's that more access to compute makes a broader range of experimentation a lot more accessible. That makes it a lot easier to find the right training recipe more quickly.

Back in 2009 the enthusiasm around neural nets was starting to grow, as there were some early successes with deep neural nets, but I remember some people were still trying to argue that support vector machines were just as effective for image classification. It seemed laughable to me at the time, but I think that everyone who wasn't a connectionist was starting to feel threatened and wanted to justify their chosen research specialty, which they would keep doing until they couldn't anymore.

> Even that was not the decisive problem - given enough compute multilayer NNs could be trained just fine.

As someone who has worked in deep learning, this isn't true. Figuring out the right recipe is extremely important. The regime in which you can effectively train a neural network with good performance is just very small. If you structure things just slightly wrong, training just won't converge.

IMO we totally could have trained something like a smaller ImageNet with a decently large dataset on a supercomputer in the late 90s. We just didn't know how. In the late 90s and early 2000s, NN researchers didn't know about (or didn't appreciate the importance of) batchnorm, or ReLUs, or which hyper-parameters to use. Convnets had been invented but hadn't yet been popularized.

In the late 90s, you might have tried training an 8-layer deep MLP on a dataset of 50K images on your Cray T3E supercomputer. But you would most likely have failed, because you weren't using convnets, your sigmoid activation function lead to a vanishing gradient, and your hyper-parameters were off. You might let this run overnight, but 12 hours later, your loss barely went down, and you were out of compute budget on a supercomputer you were sharing with other researchers. Given the right code, the right recipe, you could have made it work on that 90s supercomputer and with the compute budget you had, but the right recipe just wasn't known at the time.

We needed time to figure out and to publish about the really basic ingredients needed to train bigger neural networks successfully. Having access to more compute can help figure out what the right recipe is through trial and error, but ReLUs are a super basic innovation. They are a very simple function, and essentially a theoretical innovation. Someone needed to sit down and figure out that vanishing gradients were actually a problem we needed to think about.

I see a lot of cynicism around renewable energy and electric cars and I frankly don't get it. The typical cynical approach is to point out one potential problem, and dismiss the entire solution as if it was worthless, which is obviously false.

As others have pointed out, there are now HVDC lines that can go over thousands of kilometers (transcontinental scale!) with reasonable losses. There's no reason to believe that we've hit the peak of HVDC technology either. The need for this technology to exist will drive further innovation.

Even if the OP's estimate is off. Even if we "only" get to 90% renewables by 2045, that's still pretty damn great compared to what we have now, isn't it? Is your cynical opinion that we should just keep going with the status quo because we can't get 100% ?

Even if we need 15% more wind and 15% more storage to than the OP's estimate reach 99%, that's not impossible either. It will create a lot of jobs. Why not just roll up our sleeves and do the best that we can? Thankfully, the world is not only made up of cynics, and a lot of people already are.

We aren't done innovating on the renewable front either. There's LOTS of research into cheaper, safer, more scaleable grid-scale batteries. The current state of the art is cobalt-free and nickel-free lithium iron phosphate, but people sodium-ion batteries look like they're going to be viable soon.

I also feel pretty confident that we'll eventually manage relatively efficient ways to store energy over the long term. If we could turn electricity into methane or ethanol with even 25% efficiency, that would allow us to store energy from the summer to use in the winter, or to power transatlantic airplanes with renewables. That seems pretty exciting to me.

> please, change this metric to transistor per square millimeter. Thanks!

I think there's problems with that metric too. Not all transistors are created equal. Depending on the switching speed that you want, how much leakage current you can tolerate, etc., I bet you can vastly change how many transistors you can fit in a given area.

From Google: "a beginner cyclist may average around 75–100 watts in a 1-hour workout. A fit participant will average more than 100 watts, and pro cyclists can reach 400 watts per hour."

So 1KWh would be the output of a pro cyclist going hard for 2.5 hours, or a decently fit person biking for maybe 8 hours.

Also curious to hear more about this...

Wait, floating-point cache? Does it cache FP operation results? Where can I read more about this?

> when querying for "barack obama" on a model trained on a dataset that has never seen Barack Obama the model somehow generates images of random black men wearing suits[1]. This is, in my non-expert opinion, a clear indication that the model's knowledge of the world is leaking through to the image generator.

That's super interesting. It's not just black men in suits either. It's older black men, with the American flag in the background who look like they might be speaking. Clearly the model has a pretty in-depth knowledge of the context surrounding Barack Obama.

I would say the image generation model is also doing a pretty great job at stitching those concepts together in a way that's coherent. It's not a random jumble. It's kind of what you would expect if you asked a human artist to draw a black American president.

The charging voltage has to mach your battery voltage, so twice the voltage in the charging cable means you can send more power for the same amount of amps.

Unpopular opinion: I know there would be other problems to solve if things were different, but I wish there was no such thing as private companies. As it is, private companies are an investment vehicle that regular people just don't have access to. It's a game of who knows who with very little regulatory oversight. Maybe early stage startups don't need a ton of regulatory oversight (they have other things to worry about), but maybe it also shouldn't be legal to have a private company with a market cap over a billion dollars.

Slightly worrying notion but, if Russia believes that global warming benefits them, they might keep on extracting and burning fossil fuels long after everyone else has transitioned to renewables.

It's kind of like copying and pasting code from StackOverflow.. Except the code hasn't had as many eyes looking at it as something that was pasted on StackOverflow, and it has a higher likelihood of seeming like a correct solution to your specific use case.

Honestly, it seems pretty dangerous. Seemingly correct code generated by a neural network that we think probably understands your code but may not. Easy to just autocomplete everything, accept the code as valid because it compiles, and discover problems at run-time.

I haven't worked with Unity, but I'm assuming they get a percentage of the money spent through microtransactions and monetized content? They're essentially a platform, kind of like the Apple app store, and taking a cut?

In that case, it's obvious why the Unity CEO would say this. He has a financial incentive to get game developers to try and milk gamers for profit (because Unity gets a cut).

I think that this could ultimately push gamers and developers to use other engines/platforms. If you try to milk gamers for profit, it will eventually lead to unplayable games where the only way to compete online is to buy a ton of virtual power-ups and accessories.

Someone needs to write a sci-fi short story where in the future, a google AI is trained to maximize human happiness, but its ability to predict human happiness relies on this dataset of mislabeled human emotions farmed out to underpaid Indian workers.

Oh. Well that puts a different spin on things then. Karpathy said he wanted to sharpen his technical skills. He might have left in part because he felt he was too far from the action, since he was at least 2, if not 3 levels removed from the actual research and implementation work.

Have the bot tell both users that it's randint(18,25)/f/cali, then connect both users together and get some popcorn ready.

Besides the feeling that you're doing something good for the environment, EVs do have some genuinely nice advantages, including much less maintenance, being easier to drive, having better acceleration and traction (you can drive different motors and wheels at different rates), and being able to charge at home.

Range can be an issue for some people when going on long trips (> 300 miles), but the technology will keep improving. I'm sure we'll see EVs with 10 to 15 minute charge times before the end of the decade. It would be a huge competitive advantage to have ultrafast charging, so it's an obvious thing to optimize for.

The prices will decrease eventually. The prices have come up because of high demand. Tesla is selling as many EVs as it can make right now, and with great margins. Those margins will ensure that production keeps expanding to meet demand.

It sucks that the prices are high because it keeps some people out for the moment, but production is expanding as fast as it can. Capitalism makes it so the people with the deepest pockets will get EVs first, but that's still many less gas-powered cars off the roads, and we're still going to transition to EVs.

The problem is that hydrogen makes no sense. Most of the hydrogen that we have is produced from fossil fuels, and we have no hydrogen distribution infrastructure.

You could argue that electric vehicles are the most conservative solution because we already have a national power grid, and there's a fairly clear path towards producing green electricity.

I think I'd pin the primary cause as the cost of living increasing rapidly and salaries not keeping up with inflation. Couple that with it becoming increasingly obvious that alcohol consumption has negative health effects and it's not surprising that people look for other forms of entertainment.

There's also just more to do in the modern world. There's things like raves as you pointed out, video games, the online world. There's also a million new hobbies which are now much more accessible thanks to the internet and online stores, and much more media content. Compare today's access to entertainment to the pre-television era...

And finally, to top it all off, we have technologies like instant messaging. Instant messaging makes it possible to group text your pals after work and decide to meet anywhere you like. Before that, coordinating with a large group of people was a lot more complex and took some planning ahead, so maybe it was just easier to default to always meeting to the pub after work.