HNHacker News
TopNewBestAskShowJobs

hzay

648 karma · joined March 8, 2011

submissionscomments
hzay··on Dengue rates plunged after release of lab-altered mosquitoes
Currently down with dengue. Wish they'd put these good mosquitos where I live!
hzay··on Most demographers now predict that human population will plateau
Women who've had kids live longer & have fewer diseases btw. Breastfeeding reduces risk of breast cancer. It's not such a bad deal actually.
hzay··on America does not have a good food culture
"Authentic X-country food" in US is not authentic. The mildest andhra chicken curry is unbearable for the average American palate. What US has is the appearance of diversity.
hzay··on The myth of the myth of learning styles
Mixing and matching methods is best, I feel. It's like ensembling models. The benefits/inefficiencies of your favorite method get counterbalanced by the benefitd/inefficiencies of your 3rd favorite method.

For ex, if you really like deep diving into theory, you can do that for 2 weeks, then maybe a week of practical coding. And vice versa. I feel in this way, you learn at different focal lengths (when you point the lens of your mind at the task at hand), and that helps build a more thorough learning rather than choosing your favorite focal length and staying there.

hzay··on Advanced NLP with SpaCy
Regression is unsolved, for a start. Re LLMs, they are expensive to finetune and their inference times & memory footprints are poor compared to smaller models.
hzay··on How we know languages like proto-Indo-European existed [video]
Dravidian languages are extremely not lost.
hzay··on Anti-AI Hype LLM Reading List
Nice list, many are resources I used to get up to speed on this. I'm currently competing on kaggle on an NLP competition and I feel this is honestly great way to see the limitations of LMs and LLMs. Kaggle's community is like nothing else out there. They share incredible resources and experiments constantly.
hzay··on Understand (1991)
The protagonist is super intelligent but didn't foresee this? Also, what is the idea that destroys him?
hzay··on Understand (1991)
Please, am I the only one who found the last few pages of this story as meaningless word salad? I enjoyed Flowers for Algernon so much more, it moved me and OK, made me cry. This one started with the same premise but had to force myself to finish it. However I did enjoy Tower of Babylon by this author.
hzay··on Neural networks in the 1990s
Yes, but the tweet is talking about single layer networks!
hzay··on Prison chess clubs helping rehabilitate inmates
> Even if you only let the inmate keep 20%, the inmate is getting skills, money, the system is getting extra money, the company getting the data entry is getting cheap labor.

The inmate is also getting dignity and practice at regular life, which might convert a % of them to non-criminal life (I forgot the word for it).

hzay··on Archaeologists find 3000-year-old sword, exceptionally well preserved
My knowledge of bronze vs iron comes purely from playing dwarf fortress. Are they fairly comparable in the damage they can cause?
hzay··on Working quickly is more important than it seems (2015)
Eh. Anyone who observes small children will see that they learn very slowly, but constantly. And they are extremely good at it - better than nearly all adults - at a minimum they pick up language, gross motor skills like walking, running, climbing, any number of fine motor skills, within 2-3 years.

They don't work quickly, they display no haste or concern for efficiency. In fact, they're quite inefficient.

hzay··on In praise of blowing up your life
> it will make you a broken soul with broken social fabric, but with more experience.

I didn't expect this :( Why are you this way? Reading your comment until that point, I thought you'd figured life out.

hzay··on OpenAI's plans according to sama
Success invites haters.
hzay··on Meta AI announces Massive Multilingual Speech code, models for 1000+ languages
That statement makes no sense.

My point was that languages evolve, even though we write books and make dictionaries with fixed-in-time vocabulary lists. Similarly, langauges will evolve even if LLMs, for mysterious reasons, have a fixed-in-time vocabulary. I was responding to "and it locked their vocabulary in time, disallowing any further language blending".

hzay··on Meta AI announces Massive Multilingual Speech code, models for 1000+ languages
Imagine if someone said this about dictionaries.
hzay··on Salvador Dalí is the most faked artist?
I think that commenter is quite aware of the popular stereotype of the "greedy businessman", and is countering exactly that.
hzay··on The Little Book of Deep Learning [pdf]
Andrew Ng's courses are great too btw
hzay··on The Wealth Gap in Science: How Your Parents's Income Affects Your Career
Spot on. Culture is a taboo topic for Americans, although they're as influenced by it as it anyone else. The only kind of "culture" they can acknowledge (generalizing here but once I saw this, I couldn't unsee it) is one that is created with purpose, for a short term, like in software companies. Not the organic, multi-generation, undirected type.
hzay··on It doesn’t take much to make machine-learning algorithms go awry
Not really - citation is for citeable knowledge. What about unciteable stuff? Can you describe how you know what a dog looks like?
hzay··on It doesn’t take much to make machine-learning algorithms go awry
Can humans answer that? Isn't that a totally random question to use to determine whether something is "fundamentally defective"?
hzay··on Bilingualism affords no general cognitive advantages: Study
Only in US there's this strange obsession with counting how many languages one knows. If I count, everyone in my family is at least bilingual, and most are >= trilingual. It's not unusual for someone to speak 4 languages (2 regional languages, nationally common language, english).

If you really want to look into effects on cognition, it's better to count language families rather than individual languages. Two dravidian languages might be so similar (one to one mapping of linguistic concepts) that cognitively it might not change anything.

In india, the poorest people often speak the most number of languages and their cognition is affected by a number of other factors related to poverty. I'm not sure how one can properly control for such factors.

hzay··on Terence Tao on GPT-4
Imagine Terrence Tao manually entering data into spreadsheets!!
hzay··on Notes from a Sun Tzu Skeptic (2019)
This is a modern way of thinking. Ancient wisdom never comes with citations. But I'm happy to be proved wrong.
hzay··on Why human societies developed so little during 300k years
I read only Grimm's story, and there's no linenmaking in it?
hzay··on Yann LeCun and Andrew Ng: Why the 6-Month AI Pause Is a Bad Idea [video]
Can you explain a bit more? I'm taking Andrew's courses and he's so inspiring, but I want to know if he's been right or wrong in the past about AI.
hzay··on The roots of Japan's Indian curry
I appreciate your account but

> Also, most Indian Restaurants in India, if not owned, are pretty much run (cooks, etc.) by Nepalese.

Nope. Never seen a Nepalese cook in Chennai or bangalore?

hzay··on I Saw the Face of God in a Semiconductor Factory
The "point" is the meandering. The medium is really the message here.

Here's an example sentence - "But the nanoscale work of chipmaking is monotone only if your ears aren’t sharp enough to hear the symphony." See that nano, mono rhyme? It's intentional. It's OK if you personally don't like this sort of thing, but it's fun for some:)

hzay··on I Saw the Face of God in a Semiconductor Factory
Not sure why everyone is complaining this is badly written - it's prose, written for dramatic effect with TSMC as the idk, plot? It should be read for beauty and word-taste, not just the story.

If AI can produce this, that's only good news since that means we can have more fiction-like nonfiction (for these who want it).

← PreviousPage 2 of 7Next →