weebull

I had the same thing. I took the reader through the journey of "why" they should want to work in this way, but apparently that's confusing.

That actually the important part though. If people don't know why they are doing something, they don't do it.

I've found this article before and have to say, it put me off going any further when I got to...

> # tl;dr > > Here’s how getting code reviewed and submitted with Gerrit is different from doing the same with GitHub: > > - You need to add a commit-msg hook script when you clone a repo for the first time using a snippet you can find e.g. here (link)

Well, that was enough to make me think twice.

If it's anything like other Moore Threads products, this is an Imagination Technology GPU. Similar to Apple's GPUs (They forked 5 years ago).

Hardware is likely fine, but the drivers ... Well... I have no idea who's writing them.

> Something doesn’t quite add up for this dependence on CUDA. Sure, I’m installing some dependency hellscape Python research project from GitHub I reasonably expect idiosyncratic hardware dependencies, but surely the companies in this space are beyond that on the maturity timeline.

I think part of the problem is something you're touching on here. The ML software space is horrifically put together. It's a bunch of research projects cobbled together. Hence the dependency hell, and very few people who actually know how to port it. Cuda siits at the bottom of the stack with its design assumptions baked into everything above it (see PyTorch for a good example of that). Nobody wrote a hardware agnostic layer over the top, and now the thought of pulling out a Jenga block at the bottom scares everyone.

Cuda is an intentionally leaky abstraction layer over the hardware, and its worked.

> Thus my point is if AMD ported PyTorch to their hardware (and PT is not cuda-only) it would create a more level playing field on which they could compete

It is ported (I think by AMD). I use it daily. Biggest problem is that the people maintaining PyTorch aren't very interested as they don't use it themselves. Hence there are silly bugs in it. Users contribute fixes and they get closed down as "unsupported" or get plain misunderstood.

You don't mention whether PyTorch runs with ROCm though, and it does. RX 6xxx and 7xxx series GPUs do pretty well with it.

In an area so performance dominated people will spend the time porting to get the performance. I think this is especially true on the training side, because that's the area where companies have engineers waiting for hardware.

Unfortunately AMD only claim parity with Nvidia for training right now, and uplifts on inference. That's the wrong way round to get people to spend time porting as a priority.

It's more of a 3.5km trench of fire spewing lava, rather than what most would think of as a volcano. The article says it won't be as disruptive as other iclandic erruptions. That's only true if you don't live in Iceland.

A suspect there's also an aspect than longer journeys are safer to take into account. Merging onto a high speed road is a time where accidents are more likely, but you might only do it once between cities. Cruising at high speed on a good road0 is relatively safe.

Drive in a city and you may be merging 5-6 times as you move from road to road.

I'm just saying that assuming a linear relationship to distance is probably incorrect as well.

None of the things you've highlighted have ever, as a user, registered with me. What I want from a music app UI is a good search function, playlists, and then it's really down to content, content, content.

For an app like that I just need the UX to be "good enough".

The fedora proposal this is based on.

https://fedoraproject.org/wiki/Changes/fno-omit-frame-pointe...

That is an odd article you've linked to. There seems to be several factors in play.

1. People observe that there is a disparity in the number of pages on Wikipedia about men and women in science. 2. This leads to a number of people searching out people to write about that would redress the balance. 3. Unfortunately some are really only notable for their gender and/or ethnicity, and Wikipedia does not view that as a factor. 4. Hence, a lot of these pages get flagged for deletion.

The discrepancy is largely a reflection of our history. Until recently women and non-white ethnicities weren't in positions of scientific research. With fewer people in the field, there will be fewer notable people even if the chance of one person being notable is the same. It sounds like the "deletion of female scientists" is the system working correctly in response to people submitting articles for reasons other than refection the historical truth.

I'm sure there are cases of someone having their contribution ignored due to racism/sexism, but artifically inflating the achievements of people just to redress some perceived imbalance doesn't help. Now another person gets credit they don't deserve whilst the true talent gets buried deeper.

Here's an account from another WRNS P5.

https://bletchleypark.org.uk/wp-content/uploads/record_attac...

Sounds like they did the manual work of running the Bombe machines.

Sorry, but I think a lot of people see it differently. Whilst I think the most wholesome message would be one of "everyone of all types working together to advance humanity", a lot of people seem to see it as a scorecard. +1 on your side means a -1 in the "opposition's".

Hence somebody who was one of one thousand doesn't get described "one of the army of people who helped defeat evil", but have their role inflated to an individual responsible for some key aspect. That way you don't have to acknowledge the contribution of others who, in truth, did the same as they did.

So why label Flowers as one of those that built their work on what the Polish did? Collosus was built (as you say) to solve Tunny.

So they've basically done it for you. Primary sensitive information is about is predisposition to hereditary disease. That's the same for you and your siblings.

Are you missing a "not"? Otherwise you're not making sense.

The flip side is that these people had a good job for a few years, and can hopefully springboard on. The alternative is that they weren't hired by Spotify, or potentially at all.

As long as they were hired in good faith, and not to just see which people worked out, I don't have a problem.

I never really accept bug resolutions like that. Oracle corrupted data because a network driver fed it bad information? Did this network driver manage to preserve the legality of the corrupted info? Unlikely, so why did oracle believe it? If the corrupted message was still legal, then it sounds like the protocol isn't robust.

I'm sure there's all kinds of nuance to your tale, but it still sounds like Oracle's bug to me. They just found what it was in your system that introduced the behaviour they hadn't anticipated, and then got you to remove it.

The danger isn't that we put some malevolentAI in charge. It's that stupid people think what we have is worth listening to.

For example: it's fairly easy to imagine somebody asking an AI for medical advice, and gives their child the wrong medicine, poisoning them.

Now imagine that person is a busy political figure with no time to do proper research on a topic.

Netflix invests in content creation, as do other streaming platforms. YouTube piggybacks it's business on amateurs, largely paying them with "exposure" and some fractions of a cent per view.

> GPT-4 can explain the concept when prompted and can evaluate logic problems better than most human beings can. I would say it has a deeper understanding of "true vs false" than most humans.

Sigh

GPT produces output which obeys the patterns it has been trained on for definitions of true and false. It does not understand anything. It is a token manipulation machine. It does it well enough that it convinces you, a walking ape, that it understands. It does not.

> I wonder why RISC-V wasn't designed from the ground up to use separate stacks for data and subroutines.

Nothing in the ISA precludes this. It's all about the software stack, and what the compilers do. All you need to do is assign another register the role of data stack pointer and define the ABI to use it.

Even the stack pointer is just a convention in RISC V

Uber is shit now. More expensive than a regular taxi and you're waiting 15 minutes to get a driver to accept you because your run isn't long enough for them to make money on.

I just use my local firm.

No driver wants to be stuck at a fuel station for 20 minutes.

Yes they will stop being fuel stations, but they're the last places I want to charge.

Stellar event produces a gravity wave and a light pulse coincidentally at the source. As they travel through space together the gravity wave is compressing and expanding spacetime in the vacinity of the light pulse.

I could very easily believe that interaction by itself would be enough to have measurable effects over interstellar distances.

> The transfer functions in your (rec.709) color space is non-linear indeed. However, the pixel values you store are in a linear relationship with each other. The difference between values 20 and 21 are the same as the difference between values 120 and 121, assuming an 8bit signal. I.e., the information is the same for all pixels.

Difference on what scale? ... because (hint hint) it's not number of photons that hit the sensor. Nor is it photons emitted from the display.

The truth is, it's a linear measure of the voltage which drives the electron beam of a CRT. Not a very useful measure anymore, but we've encoded this response curve into all of our images and, now, this is proving to be a mistake.

Working with images would be so much easier if we stored values that represent linear light (i.e. proportional to photons entering/leaving a device) with no device curves baked in. Log formats do this, but because the order of magnitude of light is more important than the absolute value, it takes the log of the value. It's a more efficient use of bits in the storage / transmission.

* wider f-stop * less chromatic aberration * less distortion generally * smaller circle of confusion

The chromatic aberration is an important but subtle effect. Remember that lenses are multiple pieces of glass, and every interface diffracts the wavelengths of light like a prism. One of the considerations in lens design is converging all those different wavelengths of light in the same place. Not just at one point, but at every point across the image plane.

Poor lenses might do the well in an area. Good lenses do it everywhere.

Disagree that the Rocm libraries are poor. Their integration with everything else is poor because everything else is so highly Nvidia centric, and AMD can't just write to the same API because it's copyright Nvidia (see Oracle's Java case).

The adoption of CUDA has been such a coop for Nvidia, it's going to take some time to dismantle it.

Blame Nvidia. They are the ones the got the industry hooked on a proprietry API.