roflmaostc

I'm not sure what just happened but this webpage literally damaged my monitor.

There is now some flickering over the whole screen (it's not my mind, I filmed and checked my mind with another screen)

Not at all. tinygrad is a deep learning framework.

If you check the dependencies of TinyGPT-V you'll see that it does not depend on tinygrad but rather torch... https://github.com/DLYuanGod/TinyGPT-V/blob/main/environment...

Since it's based on 3D Gaussians in space, is there a way to obtain sharp images? Inherently, Gaussian functions extent infinitely, so images always look blurry. Don't they? Of course, \sigma can be optimized to be small, but then it converges to some point representation, doesn't it?

Maybe some CV/ML people can help me understanding.

HTML, the Programming language, right? https://news.ycombinator.com/item?id=38519719

From my point, there is still too much change around automatic differentiation libraries.

There is Zygote.jl, which is used in Flux.jl. However, it's more in maintenance mode. At some point, Diffractor.jl was hyped but it didn't take off yet. And then there is Enzyme.jl which people hype now.

But for me as a user, it's not clear what I should really do to make my code well differentiable for those libraries.

If you stick with torch, jax or tensorflow, everything seems to work better regarding AD.

[0]: https://github.com/FluxML/Flux.jl [1]: https://github.com/FluxML/Zygote.jl [2]: https://github.com/JuliaDiff/Diffractor.jl [3]: https://github.com/EnzymeAD/Enzyme.jl

Some of the response in the Julia community: https://discourse.julialang.org/t/discussion-on-why-i-no-lon...

>Sadly Julia is pain too if you don't do REPL/notebook (which you shouldn't)

Can you be more precise on this? In my case, I do a lot work around visual outputs (such as images). So notebook based development (such as Pluto.jl) feels perfectly fine.

once observed similar behavior with my bluetooth device years ago.

I find it kind of absurd that people claim that clock alignment has negative impact on their health. Studies show that but:

You do it only twice per year. But how often do you stay late up because of an event or whatever. And how often do you travel to another time zone, sometimes even >6 hours.

I think the latter effects outweigh the health impact of clock alignment significantly.

Yes but it seems it is at least an order of magnitude (or more) to slow/small to compensate CO2 emission of humanity.

And a lot of plants don't bind CO2 for a long time. Even trees have to be replaced by ~100 years and then if burnt or rotted they release CO2 again.

If you wait very long and you apply the right physical conditions, they might turn into oil again which was a good long term storage.

Let's say a tree takes 25m^2 in space. The average tree can bind roughly 50kg of CO2 per year. If a human emits ~10.000kg of CO2 per year, you need to maintain 200 trees for each human. Let's take Germany as reference. You need to plant 80 million (people) times 200 trees. In total this is an area of 4*10^11 m^2=400.000km^2

Funnily, this is roughly the size of Germany! So that Germany is long term CO2 neutral, you need to plant (and keep healthy) a forest as big as Germany.

No scale this up for the world, impossible.

That's the problem.

> You can radiate heat from the surface of the earth into space via the sky, but you don't lose much energy that way.

That's not quite true. Black body radiation is the only reason why earth does not heat up much more. It's the only factor with a negative radiative forcing.

Solar radiation is the only main incoming source of energy, and black body radiation the only outcome with almost the same magnitude (which is huuuuge btw)

ok to correct myself. Due to lack of much air, heat would only dissipate by Black body radiation. One would need to calculate if that could cool down the chips efficiently.

Space is around -270°C cold. The heat would dissipate so quickly, much better than even liquid nitrogen cooling.

I agree on the content but disagree with the title.

If you write tons of doctested documentation, that's a good thing?

Imho, there is never enough documentation. Even if it's outdated, it's still better than reading directly the source code and guessing everything from scratch.

Of course that is true, if you wait not significantly long enough.

Imagine the situation that you flip the coin (starting at H) and you grab it in air immediately. Of course, you will get H as result.

But let's say, the time to stop the coin can be a relatively long time T. Then, I think the probability is some kind of sum. Let's choose \Delta T= 10ms as time discretization:

P(H) = 1 / T * (10ms-0ms) + (30ms-20ms) + (50ms-40ms) + ... = 1/T \sum_{i=0}{floor(T / (2 * \Delta T))} \Delta T

P(T) = 1 / T ((20ms-10ms) + (40ms-30ms) + (60ms-50ms) + ... = 1/T \sum_{i=0}{floor(T / (2 * \Delta T)) - 1} \Delta T

For T -> \infty P(H) and P(T) getting more similar.

But, in practice you wouldn't wait equally distributed in time but more like a Gaussian distributed time period. Hence, each term of the sum would get weighted differently. And the variance and the offset of the Gaussian distribution can shift the probability in favor of H or T. It's really dependent of the concrete parameters. If you grab always after 35ms, then you'll always get T for example.

Are you aware the the light reflected from moon to earth compare to direct sunlight is probably more than 4 or 5 order of magnitudes smaller?

This excludes any possibility for generating reasonable amounts of energy from moon light. There is a reason why it is much colder during the night.

Author here. We thought it's funny and hacked that together. We would be happy to include more ambiguous papers!

Lol, I just tried again. And it indeed it gave the correct description

He's Professor at university. Nowadays, you have to go for the more famous ones

It's funny:

I asked ChatGPT-3.5 for my supervisor. Half a year ago, it could describe him, say his field of subject, his institution, ...

It replied today: "I don't have any specific information about [...] in my training data up to September 2021."

You simply should use a vessel which gets narrower to the top. If you put the crown, the water level rises. The smaller the cross section of the vessel the more the height has to increase to amount the volume.

The trolley problem is idealistic and in reality there won't be almost never a either ... or.

In practice, these days most likely a neural net decides what is the best maneuver to avoid serious harm to the people in its car. As alternative the net could try to enforce rules that "global harm" should be minimized.

And then both options as in the trolley problem are most likely not equal, in terms of probability.

It doesn't matter which functional language you use, any will do its educational purpose.

But as for other languages, there is pros and cons for each of them. I never tried F#, only Haskell and OCaml.

I wish more programmers would take an intense course on functional programming.

I learnt so much in my OCaml classes and really enjoyed programming with it.

Fun game, just out of curiosity:

Never received any training, I mostly use LaTeX so I'm a little aware of spacing and style.

Had 82/100 as average. What's your score?

Puh, I'm (EU based) so happy that I use pay mail services such as Posteo, Mailbox or Protonmail.

You essentially pay 1€ per month, I get full privacy and I can adjust the level of security, ...

I don't have a recovery option since I know that I don't loose my password and no one else would guess it.

Solution might be ShiftedArrays.jl which allows to do a circshift with a view. However, performance is slightly worse than an allocated fftshift.

We're currently working on wrappers doing that