ahoho

Kissinger picked targets in Cambodia. He gave the instruction to bomb “anything that flies, anything that moves” [1] [2]

[1] https://gsp.yale.edu/sites/default/files/walrus_cambodiabomb...

[2] https://nsarchive2.gwu.edu/NSAEBB/NSAEBB123/3%20%20Kissinger...

This was exactly my undergrad research topic! My knowledge is a bit out of date (perhaps this is in TFA), but the running theory is that there are two number systems. The first operates noiselessly and immediately for numbers under 4. The second works in the exact way you are describing, where the magnitude of errors are linear in the size of the set (the “analog magnitude” system). As I recall, there is substantial evidence supporting the idea of two separate systems rather than one.

I don’t really follow your point about integers. We know the brain does object detection at a low level, so it makes sense that the number systems will operate over countable elements.

sentence-transformers is excellent. It will be slow on CPU but, because it’s on top of pytorch, does have apple silicon/GPU support.

sentence-transformers has reasonably good tooling around this

This is spot on. I have two close relatives in their early 60s who lost their jobs during the pandemic. One has diabetes, and searched for months before finding a tenuous low-level position for the insurance

Right, and single thread performance won’t matter as much if it becomes easier to implement multithreading. This hurts legacy code, but I imagine it would be worth it in the long run.

I absolutely see the use case, and I believe AR is already being tested in some heavy industry sectors (eg oilfield services). That said, it’s not really in Apple’s wheelhouse as a primarily consumer tech company.

Not directly relevant, but I’m realizing the relative ubiquity of Apple Stores are a competitive advantage. I assume you can test an Oculus headset at the occasional Best Buy, but I suspect the experience isn’t as polished

Do you know if there's there a good reason to favor 3.11 over 3.10 for this use case?

I think that if OP finds The Matrix too adult, Ex Machina is even less appropriate.

Her is a good recommendation but I struggle to imagine a 12 year old identifying well with alienated-middle-aged-man-malaise ;) (although I loved Lost in Translation at 14 or so)

Arrival is a great recommendation. I had some quibbles with its science, but it’s good Sci Fi. Then he can introduce him to Ted Chiang’s other writing

In its defense, I saw The Matrix at 12 or 13 and adored it because I felt like it was something more serious—-that I could appreciate it was a sign I was maturing, and I was grateful for that

Of course all kids are different. I always loved the Back to The Future movies from a young age. Moon is also harder fare while being less violent than The Matrix.

You might also enjoy watching the modern Doctor Who together? Or Star Trek?

As for more books, maybe introduce him to Hitchhikers Guide?

“It was a chicken-and-egg question that has been the subject of debate for a century,” says Karl Deisseroth, a neuroscientist at Stanford University in California

Huh, I'd have said this idea was already well-established. I'm curious how this finding is different than the existing evidence that bodily sensations can influence emotions. I mean, I can't be the only one who feels anxious when I've had too many cups of coffee or ate something funny

It seems odd to pick on one of the most universally reviled Beatles' songs to make your point. Three of the band's own members despised it. I don't even think the dozen fans of Maxwell's Silver Hammer would argue it's high art

It is pretty remarkable that I, someone born decades after those songs were released, can immediately bring each one to mind as I read that list

I've had the same thought, and I'd love to see an analysis of spreadsheet software adoption and (e.g.) financial sector employment to situate LLMs in their proper historical context. A brief glance at FRED data doesn't seem to point to any big changes in the job growth rate in that sector, but there's obviously too much exogeneity there to really say.

As an LLM researcher, I think a large hurdle is the set of last-mile problems like those that have plagued self-driving cars. Yes, stakes are comparatively much lower when automating a job posting, but correctness matters.

Inference costs are non-trivial, and I wouldn’t be surprised if the cost of running ChatGPT (given the 3M/day figure) has surpassed that of training it. Without optimizations, training only uses ~3 times the memory as inference, so exponential parameter/cost scaling still affects both.

There’s ongoing research to reduce the computational costs of inference, but to my knowledge they only offer linear improvements (although I wouldn’t bet against more substantial reductions in the near future, particularly as these techniques are compounded).

These are not universals. I’ve seen the no-shampoo meme repeatedly in online comments. As someone with long, fine hair, I’ve made two attempts to go without, waiting almost a month for my scalp to “reset”. It always looked extremely greasy (like Gary Oldman’s character in Slow Horses).

I have had luck opting for gentler shampoos and not using them every wash, but I wouldn’t speak for anyone else. There is large variance in peoples’ bodies.

It’s about disease prevention (mad cow, avian flus, etc)

Eh, not so long ago you used to be able to get a decent bacon egg and cheese from a bodega for $3

Depending on your department and funding period, this may not be a like-for-like comparison. The 34k is an academic-year minimum; the CS stipend at Berkeley is $50k over the full year [1] (of course, to be fair, it is Berkeley). That said, I do think other UCs pay considerably less and do not guarantee summer funding

[1] https://jeffhuang.com/computer-science-open-data/#verified-c...

I assume it’s multiple orders of magnitude more expensive than Google because of the use of GPUs (not to mention the ad revenue)

There is ample evidence that CA had no impact. They were charlatans trying to sell snake oil; IIRC the underlying algorithm was based on a simple factorization of a User x Likes matrix that did not have substantial predictive value.

There is plenty to lay at the feet of Meta, but I've always thought the CA scandal was an overreaction.

For ML, the bottleneck is often memory, which is the same between those cards.

I’m a PhD in one of the highest CoL cities in the US, so I wouldn’t say I’m speaking from a position of particular financial privilege

I appreciate the nudge to go for a walk/read—at home I can end up sitting on my couch mindlessly scrolling hackernews comments.

I had to work from home consistently during the pandemic and I despised it. Once a week or so is ok, all the time is hell—I can’t bounce ideas off my colleagues or have serendipitous conversations. In fact I’ve made some not-insubstantial personal sacrifices to avoid working from home. Everyone is different.

Commuting is not so bad when public transportation is good and/or you live close to the office.

I used to live a fifteen minute walk from my previous job. It was something I relished, and on days with poor weather, I just worked remotely. Now I take a subway and I can read for 20 minutes.

I wonder if employers might start setting up smaller, “hybrid-first” offices in less expensive, more residential areas (eg, in the outer boroughs in Manhattan). I enjoy working from an office, but if no one comes in, the appeal is diminished.

This is great for the descriptions of Shannon alone:

Claude also had a swing about 35 feet long attached to a huge tree, on a slope. We started the swing from uphill and the downhill end of the arc could be as much as 15 or 20 feet above the ground.

Claude’s neighbors on the Mystic lake were occasionally astounded to see a figure “walking on the water.” It was me using a pair of Claude’s huge styrofoam “shoes” designed just for this.

I’m not sure I agree. There are at least two independent websites designed to combat paid user reviews on Amazon alone.

Color me skeptical on BERTopic. Without human validation, I'm not convinced that it's an improvement over existing methods.

I'm an author on a recent paper about automated topic model evaluation [1], and we found that current metrics do not line up with human judgements as well as previously thought. To my knowledge, BERTopic has only been evaluated on these automated metrics.

For datasets of under a few hundred thousand documents, Mallet (LDA estimated with Gibbs sampling) can produce stable, high-quality outputs in minutes on a laptop [2]. Even larger datasets remain tractable, although depending on your use case you may be better off subsampling.

It's possible that I've missed something, but I'm not clear on what benefits BERTopic has that existing methods do not. I don't mean to be overly negative---it has a nice API and the approach seems reasonable---I'm just wondering what's really new here.

[1] https://proceedings.neurips.cc/paper/2021/hash/0f83556a305d7... [2]: https://mimno.github.io/Mallet/ [3]: https://maartengr.github.io/BERTopic/faq.html#why-are-the-re...

You’re parroting unfounded conservative talking points. We never defunded the police. Any cuts were minor and temporary

https://www.nbcnews.com/news/nbcblk/cities-vowed-2020-cut-po...

I should have been more clear---I just meant that his argument had a similar kind of structure. I don't think he's engaging in Western exceptionalism (nor do I think he's a closet racist).

To answer the question, my point is that there are countless examples of powerful, inventive, playful, varied works of modernist+contemporary architecture. I simply reject the premise. I brought up the Barbican because the core of his argument is "nobody likes these things because they're ugly," and I'm saying: I like them. They move me. Some of my most enduring early memories are of the Barcellona Pavillion and the Getty Center. And to read these architects, that was often the point. I guess I'm asking, why should I give a shit about this guy's taste? Don't yuck my yum.

He's just not putting forward a very strong case, nor invoking others who have studied these questions before him. I'm not claiming that there aren't any valid critiques, just that his isn't one of them. If he's only stating that middling traditionalist buildings look worse than middling modern ones: I buy that, and I think it's probably a question of economics, as he says. But to me, when recent architects set out to create the monolithic and meaningful, it's not clear they're less successful than their predecessors.

Regarding public tastes, a mea culpa: I had dismissed the survey too quickly, and on a deeper look it's reasonable (although it includes few instances of what I'd call great 20th+21st century examples). But, as the Civic Art Society says, it's hard to disentangle Americans' aesthetic preferences with their normative conception of what a federal building (or any important building) should look like. That will naturally tend toward traditional styles. I'm also not convinced that today's choice expression of populist preferences---the McMansion---is going to stand any test of time. Whereas I do think currently dispreferred styles will endure and come to signal beauty.

As for writing, I guess I found Le Corbusier's treatises convincing enough? But for defenses that post-date the modernist era, I'm not sure I have an answer for you. To me, the best defense is the work itself.