chabons

I see them as fulfilling different needs. An e-bike lets you go further, faster, travel more safely on mixed surfaces, and carry more cargo. An e-scooter is smaller and more portable, making it easier to store in apartments, at work, or integrate into public transit.

If I remember correctly the fully connected layers after the attention block are [?, a*h] * [a*h, b*h] (for some scalars a,b and hidden size h), which means that transformers also scale with h^2. I don't know what fraction of the total FLOPs that section of the model takes for practical model sizes, but it would indicate that making the model narrower for the same number of params would reduce compute.

From someone with no knowledge of reverse sear: Why does that seem like a mistake? 130F is how you get medium rare with a sous-vide setup. If you're talking long and slow using an oven, it seems reasonable to me that you'd aim for the same internal temp as sous-vide.

You're correct, they are advocating that stroads be removed. The difference is that in your comparison to computers you assume that roads, like computers, have a quasi-linear trade-off between performance and power (ie: alpha x performance - beta x power = 0, obviously a simplification, but I hope it gets my point across) and thus a middle-ground between the extremes represents a good compromise for a variety of applications.

Strong Towns appears to be arguing that the trade-off between movement (roads) and value (streets) is highly non-linear, and that the middle-ground's combined performance is so much worse than either a street or road that it's not worth building. See the comment below about counter-tops vs. stoves for an example of an extreme binary case.

Reminds me of Felix the desktop cat from the Win 2000 era, which would walk around on your windows.

From the article: "[...] the BBC has not been able to confirm the claims, [...]" I don't see a problem with the word "alleges" in this context, regardless of the pedigree of the publication.

I see. My understanding was that part of the purpose of the deal to switch to Intel foundry was to set up for acquisition, so I think you're probably right there, but I'm not sure what the other strategic reasons were, if any. That move was 'fait accompli' by the time I joined. I've since moved on, but it's interesting to hear an outside perspective.

The delays surrounding Stratix 10 were indeed embarrassing, but I'm not sure how much of that was due to the switch to the Intel foundry vs architectural features new to S10.

I can't speak at all for sales/FAE support, but I guess that was a target ripe for "synergy", or w/e they call it.

> [...] spend years working on getting acquired instead of working on new chips (which happened to Altera) [...]

I'm interested in what you mean by this. I was working at Altera at the time the acquisition went through, on the software tools mind you, but I didn't notice a significant shift in strategy. Are you referring to before or after the acquisition was announced?

The better example of an abandoned route through the Cascades is the old railway route through Snoqualmie pass, which connects the metro Seattle area to the east side of the mountains. It's now a state park, and you can hike or bike from one side to the other.

As technology improved, a longer tunnel was built lower down the pass to decrease the elevation gain for trains, and the old route became obsolete. I suspect that this is not a unique story, and the question would be whether the old railway routes are still safely navigable.

The common term for this is Rockoon: https://en.wikipedia.org/wiki/Rockoon

I think he's referring to the Seattle area pro-transit volunteer organization "Seattle Subway" https://www.seattlesubway.org/

Anecdotally, it's 3/3 for me, one of which was Kids by MGMT. I'm seeing some fairly divergent results here, I wonder if it's highly sensitive to the particular user's voice.

There's a lot of this that I like, but wouldn't the auction just create a single market rate for H1-B employees that's independent of the market rate for the position they're filling? The clause where the employer has to pay the govt the remaining salary if the employee leaves early sounds like it's designed to encourage employers to pay more to discourage that, but I'm not sure what incentive other employers would have to pay more than the H1-B market rate to hire H1-B visa holders who are already here when they could pay the market rate to hire a new visa holder.

I had the chance to use one at a hackathon, and can confirm the device barely worked. It had huge difficulties differentiating between different input gestures, and the employee who demo'd it admitted as much.

I don't think the stock options would be public, since it's discretionary at most companies, and varies from person to person. Anecdotal as it is, from my experience at Microsoft the stock sign-on and annual bonus don't correlate with visa status.

Do you have a source for that "the pay is always lower" comment? As far as I know, visa status doesn't affect compensation for employees in the same full-time role at FAANG + Microsoft, though I only have direct experience with one of these companies.

Some people use it for everything, but I think it really shines in writing documents with heavy mathematical content due to the straightforward math syntax.

There was a paper posted here a couple months ago that claimed that even expert LaTeX users wrote documents slower than novice Word users, but the interesting caveat is that LaTeX users were far faster across the board when the text included a lot of math.

You probably only ever hear of the success stories. I can think of Terry Tao, John von Neumann, and Erik Demaine off the top of my head, not because they were prodigies, but because they have gained enough name recognition that I Google'd them and learned of their early education.

Even if the rest had very interesting research, you wouldn't hear about them outside of their chosen field unless they reached superstar status.

My high school education (Canada) worked like that. Calculus was taught in grade 12, so there wasn't any time to specialize more than that in the curriculum. My experience (granted, not in math, but in an abnormally math-intensive engineering program) was that colleges assumed you knew nothing more than basic Calculus, and built everything up from there. Even if you had gone to an advanced program (AP, TEALS, etc...) it really only bought you a semester head start at most.

The author also published a short summary of the findings in a link on his twitter as well, which includes a GitHub link to some of the code used for the analysis and figure generation (no data though). It's worth a read.

This, and the fact that they're a technical school by nature. Politics and business garners more news cycles than hard science, and hence skews peoples perception of the prestige of a university or success of their alumni. Look at the number of Nobel prizes per capita and Caltech blows Harvard out of the water. 17 Nobel prizes over 22k living alum for Caltech, versus 48 over 371k for Harvard. Even if you only look at financial success, CalTech still comes out ahead in both early and mid-career (https://www.payscale.com/college-salary-report/bachelors). They should try comparing to a state school with similar academic focus and ratings.

TL;DR: Small school focusing on low-visibility fields is a poor comparison.

The phenomenon you're describing gives rise to the last mile transport problem, which is notoriously hard to solve. You either want something which is compact enough to fit on existing transit systems or exists in a sufficient quantity at transit nodes to augment human mobility and smooth out the gradients between transit nodes. I'm hoping that as battery and motor technology advance we'll be able to miniaturize personal transit technology enough to make such devices ubiquitous.

Perhaps the raw number of jobs isn't the best denominator then, since many of the jobs in tech likely pay more to acknowledge this fact. A better measure would be something like the total of wages payed, or total of wages payed for the bottom 4 quintiles of workers by salary (cut out C-class executive outliers). Equity of course makes that harder, but most of the companies on this list are big enough that they're probably paying more in salary at this point anyways. I'd be interested to see where that places tech companies vs. more traditional businesses.