frankchn

There are reasons why drug approval can be slow and conservative: https://en.wikipedia.org/wiki/Thalidomide_scandal

To add on, most people have never experienced what “full braking” is in any car since it is so uncommon to even use half of the ultimate braking power a car can deliver in a non-track day or non-emergency situation. Even a modern Toyota Camry can sustain on average >1 g of deceleration force from 70 mph to 0.

Thus, people think they have applied full brakes but in reality they are only 60-70% of the way there.

Car dealers now prefer financing (if you have good credit). They can make more money that way since the banks will give them a kickback for originating the loan.

The Apple M-series CPUs with unified RAM is interesting in this regard. You can get an 16-inch MBP with an M2 Max 96GB of RAM for $4300 today, and I expect the M2 Ultra go to 192GB.

The drop is in the aftermarket immediately after the 8-K release, so I think we can attribute most of it to the market's gut reaction to this piece of news and not the broader SVB impact (which would have occurred during market hours).

Agree that the market might change its mind quickly, especially if estimated recoveries are projected to be high or the depositors completely made whole.

If the deposits are going on the parent bank's balance sheet, it would be subject to the same controls.

Treasury bonds are securities as well.

Venture-backed startups are weird in terms of cashflow compared to traditional businesses.

They appear out of thin air with no history and millions of dollars. They then constantly burn money in the tens to hundreds of thousands of dollars a month until they disappear or get injected with millions more dollars. Often, traditional banks don't know how to deal with these entities and it sets off tons of alarms in their fraud and risk management departments.

SVB and related departments know the business and financial models of startups a lot better and are more willing to help.

A high-end chip usable in the Mac Pro will draw more power than the M1 Ultra at 60W. I don’t think they are gonna fit a chip like that into any wearable unless the device comes with a backpack with the processing elements (and battery) in it.

That chart doesn't include dividends right? I think it will look slightly better if you include dividend reinvestment.

Simple, you just allocate a bigger contiguous chunk of memory and copy the entire vector over when the current chunk maxes out.

My impression (I might well be wrong about this) is that the energy from that forms part of the cosmic microwave background radiation.

The initial version of MapReduce was conceptualized and implemented by two people (Jeff Dean and Sanjay Ghemawat). Those were the only names that appeared on the paper: https://static.googleusercontent.com/media/research.google.c...

You can get paid to work on FOSS though. Google, Intel, etc… employees contributes to lots of FOSS as part of their jobs.

64 cores and close to 1 TB of RAM will probably consume not more than 2.5 kW or so based on available TDP/power consumption figures.

In 24 hours it consumes about the same amount of energy (2.5kW * 24h = 60 kWh) as charging a Model 3 LR's battery from 0 to 75%. So definitely more than heating up two dozen gallons of water even to boiling temperature (8 kWh), but still within the realm of household consumption with EVs.

The main node has 64 physical cores and 1TB of RAM, and the 24 storage nodes have 8 cores and 16 GB of RAM (192 cores and 384GB of RAM total).

So we are using 256 cores and 1.5 TB of RAM, so probably about 7.5 kW or so. Over 5 months, the compute would have used 27 MWh, so about the energy content of ~750 gallons of gas.

The storage will obviously use more energy (I don't know if they are using spinning disks or SSDs), but I think ~50-100 MWh for the entire computation is a reasonable estimate.

Transfer rates for Stanford has been extremely low historically: 42 admits out of 1,959 applicants in 2016 (https://admission.stanford.edu/apply/selection/profile16.htm...). I believe Caltech is even lower (single digits) as they only matriculate 250 freshmen every year. I certainly won’t count on getting admitted to either of those colleges as a transfer student even if you get straight As.

Just wanted to pop in to say congrats on making GA! Really happy to see the program develop from the early days :)

It is usually not quite that drastic. More often it is like GAM offer X, F offers 1.2X.

Then there isn’t a problem — at least for FAANG. People are willing to jump through more hoops for more rewards in all aspects of life.

If a less prestigious company with worse compensation tries the same thing, then they will quickly discover they can’t hire anyone. It will be self-correcting.

Yeah, a valid philosophy is that a company should have to spend at least as many man-hours interviewing me as I spend interviewing.

I don't think the math part of the Gaokao tests memorization.

Looking at a sample paper (https://medium.com/@yujia_jo/2016-jiangsu-gaokao-national-hi...), it seems like the examination really tests for the ability to do math and problem solve under fairly heavy time pressure (150 minutes total for all questions if you are in the science stream).

Speaking from personal experience, it does help in at least two ways:

(1) Being surrounded by other hard-working students pushes you to do better and exposes you to more advanced classes and research early on (a majority of my friends in CS started taking graduate courses by their sophomore year and did undergraduate research at least for a summer in one of research labs on campus).

(2) Recruiting/Ability to get interviews. It isn't a problem in getting interviews for software engineering internships/full-time positions if you have MIT, Stanford or Berkeley on your resume.

Does it use Display Stream Compression or something? Raw 5120x1440 x 240 fps x 8 bits x 3 colors = 39.55 Gbits/sec. DP1.4 has a maximum data rate of 25.92 Gbits/sec. DP2.0 on the other hand does 77.37 Gbits/sec.

That's because all of Google's main consumer properties didn't get branding changes and are still all "Google Maps" or "Google Photos", etc...

WhatsApp and Instagram are much stronger brands, and are not already branded "Facebook WhatsApp" or "WhatsApp by Facebook".

The maxed out configuration officially supports 3 ProDisplay XDRs (6K), one 4K screen via HDMI and the built-in screen.

Yeah EUV uses mirrors, which Zeiss also supplies: https://www.zeiss.com/semiconductor-manufacturing-technology....

There is an introductory PDF presentation from Zeiss about their product for EUVs: https://www.euvlitho.com/2019/P24.pdf

The DOT and the NHTSA considers that value to be $6.3 million per "statistical life" in 2009 dollars [1]

[1]: https://www.nhtsa.gov/staticfiles/administration/pdf/Value_o...

Yeah, the most advanced systems can switch between 4WD and RWD while the drivetrain is under considerable stress too. For instance, the F90 M5 can engage the front wheels on command while in the middle of a power slide and let the front wheels pull the car out of the slide.

S3 became strongly consistent only recently (https://aws.amazon.com/blogs/aws/amazon-s3-update-strong-rea...) while I think GCS and Azure Blob Storage has strong read-after-write consistency for a while now.

In any case, Cloud Spanner provides strong consistency in multi-region deployments.