bstockton

My experience, over 10 years building models with libraries using CUDA under the hood, this problem has nearly gone away in the past few years. Setting up CUDA on new machines and even getting multi GPU/nodes configuration working with NCCL and pytorch DDP, for example, is pretty slick. Have you experienced this recently?

I was little confused after reading the article exactly how this would work. I checked out the actual regulations and found this formula and explanation:

> (Intermediary global revenue) × (Canadian share of global GDP [≈2%]) × (Contribution rate [4%])

Proxy for “Canadian revenue” Intermediary global revenue This figure refers to the annual global revenue of a digital news intermediary. It excludes other unrelated revenues from the company operating the intermediary.

Doesn't seem like a very fair or accurate way to implement it regardless of whether this is a good idea or not.

Didn't really seem that vituperative to me. Seemed like he's really frustrated with a bunch of events tangentially related to Jill, gets emotional and could have been more respectful. I think he probably owes her an offline apology but I don't think this a reason to discount this man entirely.

>4) The basic theory that got us to the current AI crop was defined decades ago and no new workable theories have been put forth that will move us closer to an AGI.

I guess it really depends on what you mean by "basic theory" but my view is that the framework that got us to our current crop of models (vision now too, not just LLMs) is much more recent, namely transformers circa 2017. If you're talking about artificial neural networks, in general, maybe. ANNs are really just another framework for a probabilistic model that is numerically optimized (albeit inspired by biological processes) so I don't know where to draw the line for what defines the basic theory...I hope you don't mean backprop either as the chain rule is pretty old too.

I'm confused, what is the proposed framework supposed to fix, or how is it better? Is the goal really to reduce achievement gaps by limiting the advancement of top students? Surely, that can't be the goal...that's crazy. Furthermore, that has the possibility to exacerbate the problem by forcing advanced students to augment their math education in the private sector, something only available to wealthier families.

Also, the shifted emphasis on data science stuff is a joke. The very courses they're talking about minimizing are the building blocks of data science and there's no shortcut.

Well, one strategy would be to preemptively rebut common misrepresentations and meaningless critiques, I do agree scientists could benefit from this technique. And if it's just misinformation with no merit, use Hitchen's Razor.

However, established science has been wrong before about things there was a consensus on. We should investigate evidence that casts doubt on consensus if there is some merit, even if it is painstaking. It's one of the less sexy and tedious aspects of science, nevertheless important.

>Suppose [a person] had a basket full of apples and, being worried that some of the apples were rotten, wanted to take out the rotten ones to prevent the rot spreading. How would he proceed? Would he not begin by tipping the whole lot out of the basket? And would not the next step be to cast his eye over each apple in turn, and pick up and put back in the basket only those he saw to be sound, leaving the others? -Descartes

Science is done by clearly and logically addressing doubt. Sweeping doubt under the rug and showing prejudice in which evidence is presented is antithetical to the impetus of science (a disimpassioned search for unwavering truth). I'm surprised this is published in nature.

Edit: tried to format the quote, didn't work.

These have piqued a lot of people's interest around me. My friend recently quit a steady job to go work at Modal https://livemodal.com I have wondered about the necessity of a crane though, seems like a pretty big limiting factor.

What? Just saying a minor software bug (possibly Spotify not integrating correctly) is not the same as a transmission or brake failure. Also, in Tesla's case the minor bugs can be fixed relatively quickly and for free. I don't know what you're talking about being used to software bugs.

I think this is a bit misleading. The metric is "problems" per 100 cars. That could mean the car's transmission gave out or it could mean there is a software bug or paint issue that is causing the driver problems. In the case of Tesla it could be the case a lot of problems are due to software bugs that can be corrected over the air seamlessly for free. I think classifying each problem into categories with associated weights would help to make the metric more meaningful

Thanks a lot. The big impact vs specialized impact does frame things well. I almost see the PhD as a necessary condition for my career path; akin to an MD for a doctor. There are no laws governing the work like an MD. However, when I look at the people who are in the positions I want, less than 5% have masters, the rest are all PhDs. I'm in my late 20s by the way.