Forcing the sale of TikTok _is_ reciprocity. Facebook, Google, etc. are already banned in China. This is the CCP's policies backfiring.
312 karma · joined December 9, 2010
Instead, they just redid the analysis based on a figure in the paper and found that the old model with slightly different parameters gave a better fit to the data. This is a valuable contribution, but a bit over-stated by the paper title, and the confrontational, "gotcha" tone of the paper is unwarranted.
A better framing would have been something like "Chinchilla Scaling: Reanalyzed".
The Khmer Rouge murdered 25% of the country! 2 million people!! https://sfi.usc.edu/collections/cambodian-genocide
The Viet Cong were also horrible (excepting, of course, their intervention in Cambodia against the Khmer Rouge!) and devastated Vietnam, and the CPV rules it with an iron fist to this day.
Communism really is and was extremely, horribly bad.
In South Korea, the seemingly insane plan of the US (install a pro-Western dictator as a bulwark against Communism, and eventually transition to a liberal democracy) eventually worked. The Korean War pre-dates Kissinger's rise to power, but his overall principle of oppose-Communism-no-matter-what seems kind of reasonable, even in hindsight.
Anti-communist dictatorships and militias were often terrible, too, but it is really hard to be worse than Communists.
Grant agencies cannot always get it right. The more funding they have, the more they can give to researchers that they aren't 100% sure about.
The AV did not handle it well -- it should have stopped immediately, rather than pull over.
But it's really important not to lose sight of the fact that this situation would have been completely avoided if both cars had been AVs.
They should have made it extremely clear in their original manuscript their CT results are against a non-pretrained version that was not always trained to convergence, and so the CT results presented are a just a lower bound on performance.
They got confirmation from a Google engineer on only one block, which they trained for 1 million steps. For the other blocks, they didn't train anywhere close to that many steps, and their tensorboards don't show convergence on total cost.
ML is difficult! It's not enough to confirm that your implementation and installation are correct - you also need to train to convergence. You are free to complain that this is onerous or expensive. It's a common problem in academia vs industry in other areas of ML -- ML takes a lot of compute, and academic budgets are extremely tight. That said, CT is still orders of magnitude smaller / cheaper than most modern ML models (not surprising! CT is pretty old at this point by ML standards (3 years is a long time in this field)).
Kahng's FAQ admits that pre-training led to a better final proxy cost in the Nature paper, and that it significantly reduced compute time, and that this is the method that the Nature authors recommend and used in their Table 1 evals.
This whole situation is such a shame. Prof Kahng wrote a really great article on CT a few years back, and it gave me hope that the whole field was going to embrace ML and really take chip design to the next level: https://www.nature.com/articles/d41586-021-01515-9
Most of the field in fact has, but Kahng himself seems to be backsliding here.
Unfortunately, I suspect that Kahng's been listening to some pretty toxic people (the same day his paper was posted to arxiv, it appeared in an SB author's court filing). It seems like his group ran into some difficulties jumping into machine learning, and rather than take that as an opportunity to grow and learn, they took that as an opportunity to attack.
While I'm glad that others in the field have built on the work successfully (see Appendix of https://www.annagoldie.com/home/statement ), it would be really tragic for Kahng himself to get left behind here, and so so unnecessary!
What is even the point of benchmarking against a learning based method of you don't let it finish learning? Quality of output matters way more than compute cost.
As it turns out, they do! This method has been used in TPU for multiple generations, including the most recent.
The concerns presented in the Register are way overblown.
Sometimes in a bankruptcy, bondholders take a "haircut" and agree to get less money back because the company simply can't pay them what they're owed.
If your business is operating mainly on loans instead of revenue, it's because the lenders believe that you will provide value to someone in the future, and that you'll be able to capture some of that value and give it to the lenders. The lenders can be wrong, but it's still about giving people things they want.
There are some bad businesses, for sure, which are about tricking people into giving you money without giving them something in return. And a lot of business activity is about jostling for a good negotiation position (e.g. consolidation), so that you can capture more of the value you provide. But because people control their spending, you ultimately have to make something people want (or make them want what you have) if you want to make money.
Overall, you almost make it sound like the global trade of calories/textiles/etc is a bad thing, but it is an extremely good thing! It is one of the best things!!!
Shockley may have been a talented scientist, but he did not invent the transistor, and tried to steal (not share in, but steal) the credit from the people who did.
Newton was a jerk, but he invented modern physics. Shockley was also a jerk, but that doesn't mean he invented the transistor.
Some games today do this, but it's rare.
Game systems being treated as art (as opposed to the story/visuals/sound) is rarer still.
Incidentally, the reason he chose to present simple formulae in the post was that he wanted to show elegant simplicity, not dazzle with mathematical wizardry.
Regardless, you don't need to dunk on someone who is already dunking on himself, hard, harder than most of us are even capable of, just because in your opinion he isn't doing it in quite the right way.
If you believe that telomere shortening is the issue, we should be investigating therapies to encourage telomere lengthening.
We have access to powerful technologies (e.g. CRISPR), and we can develop technologies that are more powerful still.
And, in any case, if the price of longevity is slower social change, I'll take it!
Perhaps a world in which a 5000-year lifespan is common will be a world in which people are more socially courageous, more daring, as they know they'll have plenty of time to recover from a mistake.
We are spending a ton of money on healthcare (analogous to the trains), and almost nothing on the root cause of most of these health problems.
The reason the fable is effective is that it frames aging as an adversary, which makes it much easier to understand the necessity of spending resources to defeat it.
Bostrom would say that your concern that aging and cancer are intertwined is akin to a scientist seeing that the scale is impenetrable to all known materials, and then giving up.
Somehow, whales live hundreds of years, and trees live thousands. What's different about their biology that accounts for this? Why aren't their bodies wracked with cancer?