HNHacker News
TopNewBestAskShowJobs

bnprks

230 karma · joined December 21, 2021

submissionscomments
bnprks··on [dead]
From the paper's abstract

"We found a very weak or no correlation between AlphaFold output metrics and change of protein stability or fluorescence. Our results imply that AlphaFold may not be immediately applied to other problems or applications in protein folding."

This seems like the paper is drawing the opposite conclusion of the current HN title ("AlphaFold predicts impact of single mutations on protein stability and function") perhaps a title edit is in order?

bnprks··on Framework announces AMD, new Intel gen, 16“ laptop and more
I'm also intrigued by the business opportunities of providing more ways for fans to spend money than just buying a new laptop every 4 years. If I can upgrade to a new mainboard and turn my old one into a functional desktop, I'm much more likely to add in a $500 mainboard upgrade at the 2 year mark.
bnprks··on Man beats machine at Go in human victory over AI
Yes, but it's more unusual to find an adversarial attack that can be described in human terms and successfully executed by a human against two different models
bnprks··on We're wasting money by only supporting gzip for raw DNA files
One useful perspective to consider is the costs of computation compared to the costs of the experiment.

Considering sequencing costs alone (ignoring costs of obtaining & processing a biological sample), Illumina's current cheapest sequencing kits - Novaseq S4 300 cycles - cost ~$15k for 3000 Gbases of data (whole-genome sequencing for about 24 people). As gziped fastqs, that data will occupy about 1.3TB of space.

Extremely high-volume purchasers may be able knock that sequencing price down some, but the storage challenge starts to seem less intimidating when you realize spending 5% of your sequencing budget on storage would give you a budget of $750/TB of data.

bnprks··on We're wasting money by only supporting gzip for raw DNA files
One recent method that does this is SPRING: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6662292/

Although I'm not familiar with all the details of the method, my understanding is that the reference used by compression can be extremely low quality for biological purposes while still providing good compression -- making it possible to make a quick-and-dirty reference with much less computational time.

bnprks··on Grad student unions strike deal with University of California
To add some relevant facts from my experience as a current Stanford PhD student: If I work over the summer my yearly take-home pay is about $50k pre-tax [1]. This is theoretically for 20 hours/week of research work "for my advisor", and the rest of my time is theoretically research work "for my degree completion". (In practice this is the same research)

The money flow shenanigans as best as I can tell are:

1. My advisor gets grant money from the government

2. Stanford takes about 40% as "overhead"

3. Stanford charges my advisor my tuition costs, paying for my "class" which is just doing research with my advisor. This is about $35k/year for years 1-4, dropping to $14k/year for years 5+ with a "TGR" registration [2]

4. I get $50k/year as the remaining money

Note that my advisor also has to spend some portion of his post-overhead grant money on paying Stanford for research space and paying for part of his own salary. Numbers are of course opaque to me, but it leaves me confused what "overhead" pays for if it's not fully covering building space and faculty salaries.

From my perspective, had I skipped grad school and gone straight to work at a big tech company, I would have 5 extra years in my career earning a take-home salary of ~$250k total comp. This makes the opportunity-cost of my 6 years of graduate school about $1.5 million compared to going straight to industry out of undergrad.

Of course, grad school offers lifestyle and work-focus flexibility that can be far greater than available in industry. I feel for myself the monetary tradeoff has been okay, but only because I had no debt from college and have the option to enter a highly-paid career track (software development) immediately after graduation. For students with debt coming in to grad school or less lucrative career prospects, even Stanford's $50k/year would be tough swallow in a high-cost area like Palo Alto.

[1] https://gfs.stanford.edu/salary/salary22/salary_tables.pdf

[2] https://studentservices.stanford.edu/my-finances/tuition-fee...

bnprks··on Acquisition of chess knowledge in AlphaZero
> Data, Materials, and Code Availability

> [...] However, sharing the AlphaZero algorithm code, network weights, or generated representation data would be technically infeasible at present.

Very interesting paper overall. However, the excuse that code sharing is "technically infeasible" is wearing thin nearly 5 years after the initial AlphaZero paper was released.

bnprks··on Zig 0.9.0
I'm pretty new to Zig, but after taking another look recently I was pleasantly surprised by the progress on SIMD operations via builtin Vector types [1],[2]. For an application of "I want to speed up my math-intensive code with SIMD while supporting x86-64, aarch64, and non-SIMD fallbacks" it might be suitable for you now, and Zig certainly exposes more readable SIMD syntax than with C intrinsics imho [3]. But if you have a particular CPU instruction in mind, you're still a little bit at the mercy of the compiler as operations on Vector types aren't explicitly tied to particular CPU instructions.

1: https://github.com/ziglang/zig/issues/903#issuecomment-45950...

2: https://ziglang.org/documentation/master/#Vectors

3: https://godbolt.org/z/P73MPncGP

← PreviousPage 2 of 2