HNHacker News
TopNewBestAskShowJobs

aey

340 karma · joined November 10, 2015

http://aeyakovenko.github.io
submissionscomments
aey··on Highlights from Facebook’s Libra Senate hearing
That’s a good question. My US centric view:

The open market faction in the US would love to see those trade barriers lowered. Money is expensive to move.

From an ofac perspective, an immutable ledger would allow investigators to trace the funds. Small time criminals use venmo already.

USD as a world currency. I really can’t imagine libra being anything else but an extension of USD.

aey··on Highlights from Facebook’s Libra Senate hearing
The main difference is that self generated private keys are the bank account.

Someone earning dollars in the US, can send libra to their folks in Ukraine, and they can spend them without conversion fees.

aey··on Studies: Premarital Sex, Cohabitation, and Divorce
Not at the cost of forcing people to stay with the wrong partner.

What’s the likelihood of a person who is unlikely to engage in premarital sex to stay with a bad partner? Seems like set has large overlap because religion would be a common factor for both.

aey··on Studies: Premarital Sex, Cohabitation, and Divorce
Lol,

reducing risk of divorce != increasing happiness

aey··on Cloud TPU Pods Break AI Training Records
According to the post, it’s only 2x cheaper. So for the same dollars spent, I would cut my time from 48 hours to 24.

But, TPUs are not standard, can’t be used for any other usecase. So those savings might not actually be realizable when everything is taken into account.

aey··on Cloud TPU Pods Break AI Training Records
My personal experience with crnns and lstms was that cutting training latency from 48 hours to 24 hours didn’t make a difference in our progress. Some architectures took 4x longer to run.

48 to 1 hour would have allowed us to run a few experiment per work day. That would be huge.

If you told me that the next generation of TPUs could run the current 48 hour GPU training set in 1 hour, that would be something.

But 2x faster won’t make a difference. Generally, startups for a new chip need to demonstrate 100x improvement over the general purpose approach to get funding. So this seems like another Google vanity project.

aey··on Cloud TPU Pods Break AI Training Records
What’s the perf per watt difference?

It’s hard to compare perf per dollar. Your electricity costs and GPU costs maybe vastly different from mine.

aey··on Cloud TPU Pods Break AI Training Records
Results are disappointing. 2x speed up over general purpose GPUs doesn’t justify a whole new hardware architecture.
aey··on France has approved a digital services tax despite threats of retaliation by US
No matter how the law is written, if the result is that the exceptions only apply to French companies and not to the US, it’s bullshit :).

They could pick arbitrary ways to bisect the set until they get the desired result.

aey··on How to write idempotent Bash scripts
If it’s IO, then the inputs and outputs can be tracked as files. If it’s not IO, and just a pure computation it can be reproduced on demand.
aey··on How to write idempotent Bash scripts
Why wouldn’t you use make? Side-effect tracking and cleanup is the job of a build tool.
aey··on So you think you know C? (2016)
> I feel like the questionnaire piss off people who really knows C.

I hated this test. I’ve spent 12 years working on C targeting various flavors of arm and x86.

Just because the behavior is undefined when compiled without warnings and run on a Soviet water integrator doesn’t mean the language is undefined for the 99.995% of the industry uses.

Behavior of c89 or later with -Wall -Werror on modern clang, gcc, icc, visual studio, is well understood on arm, x86, mips, risc, ppc, Cortex-m and just about every other hardware architecture.

But, C is a pia, and I’ve been using rust instead :)

aey··on Most Unit Testing Is Waste (2014) [pdf]
That’s funny. I only write tests to prove that the program does what it’s meant to do and not to find bugs.

The goal is to prevent a future programmer, including myself, from breaking any of the declared properties.

aey··on Trade War Has Damaged U.S. Chip Industry
Yes, the concern is that in a direct democracy demagogues will be elected far more often and with far more extreme views.

Bicameral parliamentary systems are even less prone to this since the prime minister is elected by the ruling coalition. But they have other problems.

aey··on Jitk: A Trustworthy In-Kernel Interpreter Infrastructure
I am quite happy with rust these days. But it still lacks great ‘no allocation’ libraries.
aey··on Jitk: A Trustworthy In-Kernel Interpreter Infrastructure
Bpf can handle 60 million packets per second, adding a user/kernel context switch will kill that by a factor of 1000x. So while your code may not care about it, there are definitely low latency applications where milliseconds equate to dollars.
aey··on Jitk: A Trustworthy In-Kernel Interpreter Infrastructure
Any sandboxing adds a performance overhead. There is no free lunch in Operating Systems.
aey··on SEC Charges Kik With Conducting $100M Unregistered ICO
Is writing software protected by freedom of speech?

Is encoding a set of account values into a genesis block posted on GitHub a security offering?

aey··on SEC Charges Kik With Conducting $100M Unregistered ICO
Except the reporting requirement are cost prohibitive for a company that is seeking seed funding to file an s-1 or a reg—a even.

So us, poor unsophisticated plebs, can’t invest in anything until post IPO. Which seems to be way after the majority of the gains have been captured.

Ironically people are trying to build fully automatic DAOs, such that the use of funds is fully transparent. Effectively baking in the reporting into the organization.

aey··on SEC Charges Kik With Conducting $100M Unregistered ICO
The shmublic investors don’t have access to seed/series-A deals.

I personally feel that rando cryptocurrency vs Uber post IPO has similar alpha, but obviously quite different beta.

aey··on How Qualcomm shook down the cell phone industry for almost 20 years
I just moved, and I have access to 1gbps fiber at my home for $50 a month vs a severely crippled 100mbps from comcast for $120 a month two blocks away. So land internet is not free, and 1gbps is rare.

Qcom charges Apple $14 bucks for the modem. That doesn't sound crippling or unfair or unreasonable. I think if consumers had the option to buy an ipod touch for 500, then add the modem like you do with a PC no one would be buying a fully integrated device from Apple for 1000.

I worked at Qualcomm, so take this with a grain of salt. But I find it laughable that Apple is complaining when they can sell the exact same thing at 2x the price by just adding a $14 dollar qualcomm modem.

aey··on How Qualcomm shook down the cell phone industry for almost 20 years
They haven’t lost a FRAND case yet.

In the ideal world the phone would be like a PC, and you could buy an iPod touch from apple and swap in the qualcomm 5g modem or the intel one.

Apple pays qcom $14 bucks for the modem. In the PC world I think qualcomm would be able to price theirs closer to 50 compared to intels.

aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Yep, worked at qcom. I think they should charge what the market could bear, like apple charges consumers :)
aey··on Field Notes: Highlights from Huawei
A large part of the dispute is due to China not enforcing and codifying its own laws. All US wants is for the laws to be on the books and fairly applied.
aey··on How Qualcomm shook down the cell phone industry for almost 20 years
If it cost apple less to build improvement into 5G that would have been licensed to everyone they would have done so.

Standards require all participants to license at a FRAND rate. If apple developed 5G IP they would have effectively received a rebate based on the relative impact of their contribution.

Qualcomm actually has great execution. You realize 5G is going to put a 1 gigabit network in the palm of your hand? I can’t even get that out of cable.

aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Qualcomm doesn’t own spectrum either. Apple should be investing in tech, not in cash. It’s crazy to think of tech companies sitting on that much cash. At this rate, apple will have 1T in the bank in 15 years.
aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Patents don’t last forever, and there are a dozen ways to solve each problem.

Apple is sitting on 200b in cash. What would be great for consumers if they spent some of their cash on 6g networks before they turn into a sovereign wealth fund.

aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Yea. For sure. Apple should be encouraged to spend their 200b in cash on an alternative instead of lawyers.
aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Yea. Apple can build its own. There is nothing wrong with them spending their 200b+ in cash on integrated modem tech.
aey··on How Qualcomm shook down the cell phone industry for almost 20 years
Yea, my point exactly. All the recent IP covers improvements since 2G. Apple can try to sell a 2G network connected iPhone based on expired IP without the Qualcomm tax.

Free market pricing is based on demand of the feature not the cost to make it. Qualcomm should be pricing the part at the point where the phone without the qualcomm part is equal value per dollar.

← PreviousPage 4 of 9Next →