HNHacker News
TopNewBestAskShowJobs

ZhuanXia

1,002 karma · joined October 19, 2018

submissionscomments
ZhuanXia··on Physician burnout widespread, especially among those midcareer, report says
The Doctor's guild does this to themselves by restricting supply to obtain their grotesque salaries. Zero sympathy.
ZhuanXia··on Anonymous shell companies buying American real estate
Land value tax.
ZhuanXia··on The Mechanical Muse
See Gwern’s automated poet: https://www.gwern.net/GPT-2
ZhuanXia··on [dead]
He did much better than most. This captioning task is extremely hard even for professional cartoonists. Few could have done better.

I admire Yang's intelligence, and think it is a shame that we do not filter our leaders more for cognitive ability, as Singapore does. This is almost enough to make me want to vote for the man.

However, his big policy item, BIG justified by Luddism, is confused economics. Comparative advantage basicly guarantees human employment until robotic labour is 100% as capable as human labour. True, after that point the value of labour declines below subsistence. But this implies human-level, or more likely, smarter than human AI. At that point, economic policy is pointless as all that matters for human welfare is the objective function of the AIs.

For it is they who will have the power to shape the world. Let us hope they are coded with care.

The lump of labour fallacy is itself a fallacy, as it would fail to be true if we had robotics and software that could do 100% of all economically relevant cognitive tasks. Economists who don't mention this limiting case when denouncing Luddism really do annoy me.

Still, in practice it appears to me to be a very good heuristic for the reasons mentioned above. Ironically, the Luddites are wrong for the "right" reasons, applying economic theory dogmatically to realms where it ceases to matter. And the Economists are right for the wrong reasons, ignoring a limiting case that is obviously possible and probably inevitable.

ZhuanXia··on Nursing Homes in the US Overmedicate People with Dementia (2018)
Reversal test, how do you “know” they want to live?
ZhuanXia··on Nursing Homes in the US Overmedicate People with Dementia (2018)
I hope when I get that degraded I will live in a society kind enough to euthanize me, rather than slowly torture, drug and humiliate me, while extending my demented, horror-show 'life' to the maximal possible extent in order milk my children, savings, and/or the state. Soylent green is a utopia compared to what we have now.

Dr. Scott Alexander writes well here on how we treat the demented: http://slatestarcodex.com/2013/07/17/who-by-very-slow-decay/

"And now every time I hear that phrase I want to scream. 21st century American hospitals do not need to “cultivate a culture of life”. We have enough life. We have life up the wazoo. We have more life than we know what to do with. We have life far beyond the point where it becomes a sick caricature of itself. We prolong life until it becomes a sickness, an abomination, a miserable and pathetic flight from death that saps out and mocks everything that made life desirable in the first place. 21st century American hospitals need to cultivate a culture of life the same way that Newcastle needs to cultivate a culture of coal, the same way a man who is burning to death needs to cultivate a culture of fire."

ZhuanXia··on Letting nurse practitioners be independent increases access to health care
Anything we can do to water down the Doctor’s guild is good. Their salaries are absurd and possible only because of regulatory capture.
ZhuanXia··on Ask HN: A New Decade. Any Predictions?
1. 100 trillion parameter neural net is trained at the very end of the decade. May or may not be useful but I think this is mostly a problem of memory bandwidth and is doable with a minimal number of die shrinks.

2. I will take the under on SelfDrivingCars cars. Way too much pessimism here. The entire first wave of startups( save for TESLA and Comma) are using the same DARPA challenge codebase. New players using reinforcement learning will crack the problem.

3. China will not become a democracy.

4. China will be a leader in software as well as hardware. A Chinese-made social network will be widely used in Europe.

5. Learning-based systems will be successfully applied to proof search. Very significant mathematical problems will fall to such systems before the end of the decade.

6.Quantum computers will not be capable of simulating any significant/useful chemistry in this decade.

7. A baby with 100+ edited alleles will be born and healthy, probably in China.

8. Cognitive genomics will come to the fore as it becomes clear several standard deviations increases in mental traits, including IQ, are possible with such edits. Lysenkoist ideologies may have trouble adapting, but by the end of the decade it will be clear to most which way the wind is blowing.

9. AGI has not happened yet, but the conspiracy of silence around the topic no longer prevails. The field. Most people in machine learning consider AGI the goal of the field, and say so without shame.

ZhuanXia··on Ask HN: A New Decade. Any Predictions?
Generally a good prior.
ZhuanXia··on Cerebras’s giant chip will smash deep learning’s speed barrier
Them shunning benchmarks is pretty lame.
ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Read it again. It clearly does.
ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
>That they are "excited" to account for at best 2.5% of high IQ through genetics suggests that genetics is not a large contributor to high intelligence.

Bit of a stretch there mate; that doesn't follow at all.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
I suppose I should have said it does not help much. You are right; it does help.
ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
>They conclude that high IQ (>170) is mostly not caused by genetic variants.

The abstract does not mention that; they claim:

"This study shows the utility of extreme trait selection for genetic study of intelligence and suggests that extremely high intelligence is continuous genetically with normal-range intelligence in the population."

I don't think anyone would seriously claim that high IQ is mostly not caused by genetic variants, given the prior knowledge of the heritability of IQ.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Yes. That is true. It will take hundreds of edits to increase IQ several standard deviations.

This is not a problem for embryo selection.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
the enriched population does not help as high IQ is not caused by rare variants. Rather it is the proportion of common variants that increase IQ to common variants that decrease it. This is why we need millions of samples.
ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
>Polygenic scores for IQ are correlated with other traits that are less desirable (schizophrenia for example).

Nope. IQ is negatively correlated with most mental illness. It is certainly not positively correlated with schizophrenia: https://www.gwern.net/iq

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Yes, do to their large population and slightly higher average quantitative abilities (at least if you trust their PISA scores), they have roughly 20x more 4 sigma people than America.

Chinese technological dominance seems likely even ignoring embryo selection. This will likely be amplified by better agglomeration in cities.

Were western institutions capable of thinking long-term, this would add to their desire for human enhancement.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Yes. For polygenic traits like IQ we need millions of samples tagged with IQ. 3000 is a really small sample.
ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Yes. I am a little worried about genetic engineering in the very very long term, as it seems to make eusocial organization possible, which I find pretty abhorrent.

In the short term though, it seems like a good idea. Regardless, in a multipolar world we don’t have much choice in the matter.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
We don’t need the causal variants for embryo selection, just the predictive variants. We have this for height now. The only reason we don’t have this for IQ now is everyone doing GWAS studies are too chicken to put an IQ test in with their other measures. As the cost of a sequence declines, this will not last.

Obviously, this is not an issue if there is state involvement.

ZhuanXia··on Chinese scientist who edited twin babies' genes jailed for 3 years
Interesting. I still roll to disbelieve on China shunning human enhancement. The gains are just too great, the fruit too low hanging. Several standard deviation increases in polygenic traits are on the table, including the most important economic trait, intellectual ability: https://www.gwern.net/Embryo-selection

Given China’s almost unique state capacity to think long term, history of eugenic policies in sport, an ironically less Lysenkoist population than America, and a cultural focus on educational achievement and competition, this seems overdetermined.

In fact, animal breeders are getting very keen on iterated embryo selection. So the technology will get investment even in environments where it is illegal for humans.

We live in interesting times. Even a 2 standard deviation increase would allow a small population to produce the vast majority of genius-level intellects. 2 standard deviations is on the way low end of what iterated embryo selection offers. Should we find the causal variants, direct editing offers vastly more.

A world-wide ban on this technology is not on the table. Plan accordingly.

ZhuanXia··on The promise and peril of the new science of social genomics
To eliminate Mendelian disease. To increase lifespan, health, intellectual ability, and productivity? This is quite possible with minimal technological advancement. Why exactly is this bad? Because we saw a Star Trek episode?

Shunning these technologies will be seen, in time, as akin to the shunning of inoculations in the Victorian era.

When you find yourself cheering on, however indirectly, Mendelian disease consider perhaps that you are one of the baddies.

ZhuanXia··on The promise and peril of the new science of social genomics
The gains from engineering human capital are so large they will be plucked. Screaming “eugenics” is not going to change that. One slur does not an argument make.

Gains of several standard deviations in mental traits, including intelligence, are looking very plausible with minimal technological progress. We are well past the speculative phase.

As we live in a competitive technological landscape and China, Korea and India do not share our taboos against genetic enhancement, we will be forced to embrace this, too.

Tar and feathering human enhancement as “eugenics” worked when human enhancement was not possible. We no longer live in that world. The signal is no longer cheap. Whatever nations fail to embrace embryo selection, iterated embryo selection, and eventually genetic engineering for intelligence will find themselves as children, competing in a technological landscape they are too dull to understand.

Like Lysenkoism of old, our modern mental Lysenkoism will become unsustainable. We would be wise to discard it now.

ZhuanXia··on New approaches shed light on the magnitude of sex differences in personality
>The danger is twofold. First, people forget that these are population level trends and that it is bad reasoning to automatically assume every individual in a given group conforms with these trends.

Yet it is equally bad reasoning to ignore the prior as it is to refrain from updating it. One form of foolishness will find you unpersoned, the other such mental handicap allows one to signal great virtue. As it takes great moral restraint to distrust your lying eyes.

We are in interesting ideological waters. The great taboo of our age is quantifying human capability, precisely when our tools have never been sharper.

Cognitive genomics is joining this strange fray. Extremely politically incorrect truths are beginning to stumble reluctantly out of the field. The confirmation of a dysgenic trend in Iceland is one of the milder examples:http://www.pnas.org/content/114/5/E727. Much more controversial truths are yet to come.

One wonders how much truth our mental Lysenkoism will survive. It has resisted all reason so far, so one must bet it will continue.

ZhuanXia··on World Chess Champion Magnus Carlsen Is Now World Fantasy Premier League #1
I often wonder about this in the context of musicians and actors. You would have to be a fool to pick such careers, as the prospects are so dim. Yet because of this dimness the selection pressure is very high on the pool of contenders. So our best actors and musicians are fool/geniuses.

Which implies the most innately talented potential musicians and actors are working at a hedge fund or Google DeepMind. And perhaps the odd personalities of musicians and actors is more to do with the bottlenecked selection than some relationship between talent and impulsivity: Only the egotistical and impulsive would step into such fray.

And so we are left not with the best but with the best of a bad lot!

ZhuanXia··on Robots in Finance Could Wipe Out Some of Its Highest-Paying Jobs
>$70k per device

That's pretty steep compared to the price of the labour.

ZhuanXia··on What Would the Father of Cybernetics Think About A.I. Today?
Worth noting that this is no longer true:

“The world’s best chess players are neither computers nor humans but humans working together with computers.”

ZhuanXia··on A Look at Cerebras Wafer-Scale Engine: Half Square Foot Silicon Chip
Interesting things I’ve learned about this chip after a little sluthing.

CEO implies in the FYI podcast that it can handle model sizes of a max of ~4 billion parameters per wafer. Respectable but not as large as I assumed would be possible when I first read of the scale of the chip.

CEO claims model parallelism will actually work well with these devices. Would be intriguing to know the limits of this.

Based on the cooling requirements, at least a GHz clock speed appears likely. If we take the heretical position that 1 parameter is >= 1 synapse, 4 billion parameters is about the size of a bee brain. 1GHz is about 5 million times faster than a bee brain.

It would take about 20000 such chips to simulate a human-sized network. Not economical and presumably the model parallelism would break long before this. But it is interesting to note that under the not implausible 1 parameter is >= 1 synapse assumption, we are only a few orders of magnitude away from human-sized networks training 5 million times real-time.

ZhuanXia··on As L.A. ports automate, some workers are cheering on the robots
Rents are by their nature exclusionary; it is hardly surprising they became hereditary rights. They are called guilds after all.
← PreviousPage 2 of 6Next →