That reminds me of a joke.

A museum guide is talking to a group about the dinosaur fossils on exhibit.
“This one,” he says, “Is 6 million and 2 years old.”
“Wow,” says a patron, “How do you know the age so accurately?”
“Well,” says the guide, “It was 6 million years old when I started here 2 years ago.”

Apparently Inflection AI have bought 22,000 H100 GPUs. The H100 has approximately 4x the compute for transformers as the A100. GPT4 is rumored to be 10x larger than GPT3. GPT3 takes approximately 34 days to train on 1024 A100 GPUs.

So with 22,000*4/1024=85.9375x more compute, they could easily do 10x GPT4 size in 1-2 months. Getting to 100x the size would be feasible but likely they’re banking on the claimed speedup of 3x from FlashAttention-2, which would result in about 6 months of training.

It’s crazy that these scales and timelines seem plausible.

GPT-4 was fine-tuned on English and Chinese instruction examples only (source). There’s clearly some western bias in the historic events, but it would have been interesting to also discuss if there was a bias towards Chinese events as well. And, if so, what other languages or prompts may elicit that bias.
As an example, could you get the model to have an English bias with “I’m from America…” and a Chinese bias with “I’m from China…” even when using English?

Haha, thanks for the correction. If you have to use your degree in ethics, perhaps you could add your perspective to the thread?

How many people will read the title without the comments and leave with the wrong idea?

Not that I think you should take the post down, but the title is quite definitive, and confirms existing biases, so people are unlikely to research further.

If you can get past the weird framing device, the Plinkett reviews of the Star Wars prequels are an excellent deep dive into the issues with those films: https://www.youtube.com/watch?v=FxKtZmQgxrI&list=PL5919C8DE6F720A2D

Jenny Nicholson’s videos are great, but her documentary on “The Last Bronycon” is special, as the realization dawns on you while watching that she has more connection to Brony culture than you might have guessed: https://www.youtube.com/watch?v=4fVOF2PiHnc

According to consequentialism:

1. Imagining sexual fantasies in one’s own mind is fine.
2. Any action which affects no-one but the actor, such as manifesting those fantasies, is also fine.
3. Distributing non-consensual pornography publicly is not fine.
4. Distributing tools for the purpose of non-consensual pornography is a grey area (enables (2), which is permissible, and (3), which is not).

From this perspective, the only issue one could have with deep fakes is the distribution of pornography which should only be used privately. The author dismisses this take as “few people see his failure to close the tab as the main problem”. I guess I am one of the few.

Another perspective is to consider the pornography itself to be impermissible. Which, as the author notes, implies that (1) is also impermissible. Most would agree (1) is morally fine (some may consider it disgusting, but that doesn’t make it immoral).

In the author’s example of Ross teasing Rachel, the author concludes that the imagining is the moral quandry, as opposed to the teasing itself. Drinking water isn’t amoral. Sending a video of drinking water isn’t amoral. But sending that video to someone dying of thirst is.

The author’s conclusion is also odd:

Today, it is clear that deepfakes, unlike sexual fantasies, are part of a systemic technological degrading of women that is highly gendered (almost all pornographic deepfakes involve women) […] Fantasies, on the other hand, are not gendered […]

1. Could you not also equally claim that women are being worshipped instead of degraded? Only by knowing the mind of both the consumer and the model can you determine which is happening. And of course each could have different perspectives.
2. If there were equal amounts of deep fakes of men as women, the conclusion implies that deep fakes would be fine (as that is the only distinction drawn), which is probably not the author’s intention.
3. I take issue with the use of systemic. The purpose of deep fakes is for sexual gratification of the user, not degradation. Only if you consider being the object of focus for sexual gratification to be degradation could the claim that there is anything systemic. If it was about degradation, wouldn’t consumers be trying to notify targeted people of their deep fake videos and make them as public as possible?
4. Singling out “women” as a group is somewhat disingenuous. Women are over-represented in all pornography because the majority of consumers are men and the majority of men are only attracted to women. This is quite clear as ugly women aren’t likely to be targeted. It’s not about “being a woman”, it’s about “being attractive to pornography consumers”. I think to claim “degradation of women” with the caveat that “half of women won’t be affected, and also a bunch of attractive males will be” makes the claim vacuous.

For microcontrollers, quite often. Mainly because visibility is quite poor, you’re often trying to do stupid things, problems tend to be localized, and JTAG is easier than a firmware upload.

For other applications, rarely. Debuggers help when you don’t understand what’s going on at a micro level, which is more common with less experience or when the code is more complex due to other constraints.

Applications running in full fledged operating systems often have plenty of log output, and it’s trivial to add more, formatted as you need. You can view a broad slice of the application with printouts, and iteratively tune those prints to what you need, vs a debugger which is better suited for observing a small slice of the application.

Cool, you posted the original with the Tim Minchin callout.

The approach requires multiple base stations, each in the path of a ray which is detected at both the station and receiver, and the receiver’s position can only be known if there is communication with the stations.

I had to go find the thread being referenced, it’s here for anyone else who hasn’t seen it: https://kbin.social/m/lemmynsfw@lemmynsfw.com/t/42348

I took the time to post my own thoughts in the thread, which I’ve reproduced here:

It seems a lot of people have very strong negative reactions to anything which is too close to children in pornography. I don't think that's an especially bad thing, but in my mind the only reason to be against minors in pornography is to protect those same minors from harm.

As an analogy, I find scat porn disgusting. But that doesn't mean it should be banned, as long as all parties are able to consent and no harm is done.

When it comes to non-child-coded drawings (though with childish proportions), I don't see the harm or lack of consent. And if that is the case, it doesn't matter what my personal feelings are about the content, it shouldn't be banned.

So, thus far, the cost of ITER is less than the Manhattan project, but it has taken longer. The adage that it is easier to destroy than to create comes to mind.

It does seem like ITER could be more transparent, but the article is overly hyperbolic about one of the most important civil works going over time and budget.

America has spent 5x the ITER budget on Ukraine so far (and rightly so). I wish we lived in a world where that money could have supported research projects like this instead.

Yeah, on closer inspection it looks like kbin is still having federation issues

There’s already:

https://kbin.social/m/ai
https://kbin.social/m/ArtificialIntelligence
https://kbin.social/m/machinelearning

I don’t think the UI is doing the heavy lifting to make these links easy to use outside of kbin. To join from, for example, lemmy.world, I think you write: https://lemmy.world/c/ai@kbin.social

But unfortunately, federation is still a bit broken.