HNHacker News
TopNewBestAskShowJobs

KhoomeiK

1,299 karma · joined August 7, 2018

research @ openai

https://rpandey.tech

submissionscomments
KhoomeiK··on I am sick and tired of hearing tech companies complain about developer shortages
Totally agree and I thought about mentioning this in my comment. Yeah, part of the blame for this problem is on us as developers for creating a culture that encourages constant company hopping.
KhoomeiK··on I am sick and tired of hearing tech companies complain about developer shortages
I honestly don't think pay is an issue in tech, but most of the others the author brings up are valid points. The crux of the issue in my opinion is the fact that NO ONE HIRES JUNIOR DEVELOPERS.

Companies don't want to take on the burden of training junior devs and then complain when there aren't enough senior devs to go around. Obviously it makes sense from a business perspective, but taking a step back, training junior talent in this field should be a top priority for everyone.

KhoomeiK··on Show HN: Lofi.cafe
Definitely better than going to YouTube multiple times a day to search "lofi" and clicking on one of the newly uploaded compilations!
KhoomeiK··on Show HN: NNext.net – A Firebase-like managed vector storage for ML applications
What I've done for most of my research projects is just pickle a PyTorch dataset object that contains all my embeddings. The pkl file can then just be uploaded anywhere and becomes plug and play with any Torch model.

What advantages would this bring for a user like me? I guess it might make more sense for people working closer to production?

KhoomeiK··on Graph theory, graph convolutional networks, knowledge graphs
Recently came across this beautiful paper [1] about embedding graphs in hyperbolic space. Graph embedding algorithms for euclidean space are flawed because you essentially need to add an additional dimension for every node. Of course there are techniques you can use to reduce the number of dimensions by graph clustering etc, but fundamentally the dimensions scale with the size of your graph.

Hyperbolic space, on the other hand, enables you to embed hierarchical information in a single dimension due to the curvature itself (yes this is hand-wavey mainly because I still barely understand it, read the paper). This is very conducive to embedding graphs, and especially structured graphs (or trees).

[1] https://arxiv.org/abs/2005.00545

KhoomeiK··on This is How Google will Collapse (2017)
Google will fail with a 508 Resource Limit is Reached error?
KhoomeiK··on When parties reverse their stance, supporters immediately switch their opinions
I've been noticing this shift recently taking place with US party opinions on foreign policy. Republicans have historically been more anti-Russia and hawkish, but during the Trump presidency, Democrats started calling out Trump for getting close with Russia. The Republican voter base has also suddenly turned into a bunch of doves, getting behind Trump's calls to pull out of Syria and Afghanistan.
KhoomeiK··on How to overcome Phone Addiction
While it's certainly true that LSD is the primary reason I had ego semi-dissolution in this experience, I'm saying that phones serve to reinforce our understanding of ourselves by "interrupting" our raw conscious experience with memories and data that we correlate with our self. On other trips where I had my phone, or when I was at home in an environment I've known all my life, I had virtually no ego dissolution at all.
KhoomeiK··on How to overcome Phone Addiction
Lol I'm not proposing an LSD-based solution, but I do have a relevant anecdote from my first trip.

We were on the beach sitting on blankets, and my phone somehow got lost in the folds of the blanket while I was coming up. For the next ~4 hours I basically forgot "who" I was, being deprived of a phone that constantly shows me my friends, notes, and pictures. Eventually, I did find my phone and it was the weirdest experience ever to look at my phone and slowly start "remembering" who I am.

KhoomeiK··on How to overcome Phone Addiction
> Your online and offline behavior gets more integrated and you don’t know what you did in the digital reality or the material reality.

Not sure if it's just me, but this is getting very hard to separate during quarantine. I often remember conversations that happened over text as having happened in person. Perhaps my mind is accounting for the lack of face-to-face interaction.

KhoomeiK··on Archaeologists uncover ancient street food shop in Pompeii
Yeah, it could be a modern phenomenon and even now, a very small percentage of people have an understanding of the grand scheme of history. Even without knowing the dates and exact order of events though, it still seems plausible for Persians to have an emotional connection to historical legends that they pretty much grew up with.

That 5000 number is interesting though. Do they feel a connection with the Elamites, Sumerians, and Akkadians? That would be kind of surprising. It would make sense for them to connect with Xerxes and the Achaemenids or Medians though, despite the difference in religion. Sorta related, I've heard that there's a growing Zoroastrian movement among the "rebellious" Persian youth.

KhoomeiK··on Archaeologists uncover ancient street food shop in Pompeii
The way the art has been preserved makes you sorta realize that "ancient times" weren't that long ago.

2000 / 25 = ~80 generations

It's not some uncountable number with a million zeros. Assuming everyone knew one of their grandparents, it's more like 40 interpersonal "links".

I honestly feel like if Europe didn't have the Early Middle Ages (~500-1000AD), we would feel pretty connected to all the history that happened before in the same way that we feel pretty connected to Copernicus developing Heliocentrism or Columbus arriving in the New World.

I wonder whether civilizations with seemingly more continuous histories like China feel more connection to the rest of their history. India's history was sort of interrupted by Islamic invasions starting from the 1200s. The Middle East, despite its tremendously long history, experienced a significant shift in culture with the rise of Christianity and then Islam.

If Chinese people don't feel that historical connection, on the other hand, it might mean that our disconnect with ancient history might more just be a product of Modernity and the rapid change of the last couple centuries.

KhoomeiK··on Ask HN: What Lived Up to the Hype?
It's pretty hard to get your hands on a hard copy because of how large it is (usually published in multiple volumes for academia). The most accessible translation is Ganguli's, which you can find here [1].

[1] https://www.mahabharataonline.com/translation/

KhoomeiK··on Ask HN: What's the best paper you've read in 2020?
Discovering Symbolic Models from Deep Learning with Inductive Biases [1] trains graph neural nets on astrophysical phenomena and then performs symbolic regression to generate algebraic formulae to elegantly model the phenomena in a classical physics framework. It's largely gone under the radar but has pretty interesting implications for NLP and language theory in my opinion.

Direct Feedback Alignment Scales to Modern Deep Learning Tasks and Architectures [2] applies DFA, an approach to training neural nets without backprop, to modern architectures like the Transformer. It does surprisingly well and is a step in the right direction for biologically plausible neural nets as well as potentially significant efficiency gains.

Hopfield Networks is All You Need [3] analyzes the Transformer architecture as the classical Hopfield Network. This one got a lot of buzz on HN so I won't talk about it too much, but it's part of a slew of other analyses of the Transformer that basically show how generalizable the attention mechanism is. It also sorta confirms many researchers' inkling that Transformers are likely just memorizing patterns in their training corpus.

Edit: Adding a few interesting older NLP papers that I came across this year.

StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding [4]

Do Syntax Trees Help Pre-trained Transformers Extract Information? [5]

Learning to Compose Neural Networks for Question Answering [6]

Parsing with Compositional Vector Grammars [7]

[1] https://arxiv.org/abs/2006.11287

[2] https://arxiv.org/abs/2006.12878

[3] https://arxiv.org/abs/2008.02217

[4] https://arxiv.org/abs/1908.04577

[5] https://arxiv.org/abs/2008.09084

[6] https://arxiv.org/abs/1601.01705

[7] https://www.aclweb.org/anthology/P13-1045/

KhoomeiK··on Siteswap
The article was honestly a bit confusing for me with no juggling knowledge, but this video I found helped a lot with a visual demo of how the numbers are actually used [1].

[1] https://www.youtube.com/watch?v=NsIlQDhMKro

KhoomeiK··on Ask HN: Who wants to be hired? (December 2020)
Location: San Francisco Bay Area / San Jose / Cupertino

Remote: Open to it

Willing to relocate: Open to it

Technologies: PyTorch, TensorFlow, Keras, Sci-kit Learn, NumPy, Pandas, Flask, Express, React, React Native, PostgreSQL, SQLite, Python, JavaScript

Résumé/CV: http://resume.rpandey.tech

Email: rohan [dot] pandey [at] gmail [dot] com

Website: https://rpandey.tech

GitHub: http://github.rpandey.tech

LinkedIn: http://linkedin.rpandey.tech

I've won 10 hackathons with my Fullstack expertise, am publishing an undergrad Machine Learning research paper, and am using DL for NLP + Computational Neuroscience work in my current internship. My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also just love building impactful products!

Open to internships, part-time, and some full-time positions.

KhoomeiK··on YouTube-dl has received a DMCA takedown from RIAA
Git-on-the-Blockchain?
KhoomeiK··on Show HN: Turn your website console into a chat that connects website visitors
Was awesome until the loop spam began. I think your server just crashed too. Never thought I'd say this, but please add rate limiting. :)
KhoomeiK··on Cat Gap
No. From later in the article:

> During or just prior to this "cat gap," numerous caniform species evolve catlike features indicative of hypercarnivory, such as reduced snouts, somewhat enlarged canines, and fairly extreme reduction of their crushing molars.

They propose that Felids were pushed out of their hypercarnivorous niche by Canids as evidenced by changes in Canid dental morphology from this period.

KhoomeiK··on Ask HN: Who wants to be hired? (October 2020)
Location: San Francisco Bay Area / San Jose / Cupertino

Remote: Open to it

Willing to relocate: Open to it

Technologies: Flask, Express, React, React Native, PostgreSQL, SQLite, PyTorch, TensorFlow, Keras, Sci-kit Learn, NumPy, Pandas, Python, JavaScript

Résumé/CV: http://resume.rpandey.tech

Email: rohan [dot] pandey [at] gmail [dot] com

Website: https://rpandey.tech

GitHub: http://github.rpandey.tech

LinkedIn: http://linkedin.rpandey.tech

I've won 10 hackathons with my Fullstack expertise, am publishing an undergrad ML research paper, and am currently working on DL for NLP in my internships. My areas of research interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also just love building impactful products! Open to internships, part-time, and some full-time positions.

KhoomeiK··on Social Cooling (2017)
> and refraining from imposing their thoughts and opinions

This isn't what social cooling results in though. Thoughts and opinions are imposed, it's just that their imposition is monopolized and becomes implicit. Dirty laundry will still be aired in the town square, but it'll be the King's and everyone will be forced to smell it.

KhoomeiK··on Over 90% of Indian techies in the US are upper-caste Indians
Brahmins are ~5% of India's total population and another 15% are also considered "forward caste", mainly from warrior or mercantile subcastes. So yeah pretty significant overrepresentation.
KhoomeiK··on Talking is throwing fictional worlds at one another
Shameless plug for my essay [1] that discusses some similar ideas to this article. I focus more in-depth on what the consequences of compositional grammar are for ML-based NLP and why we should be looking to theories of psycholinguistics and philosophy of language for inspiration as a research community.

[1] https://rohan.bearblog.dev/humans-spoke-vectors/

KhoomeiK··on Ask HN: Who wants to be hired? (September 2020)
Location: San Francisco Bay Area / San Jose / Cupertino

Remote: Open to it

Willing to relocate: Open to it

Technologies: Flask, Express, React, React Native, PostgreSQL, SQLite, PyTorch, TensorFlow, Keras, Sci-kit Learn, NumPy, Pandas, Python, JavaScript

Résumé/CV: http://resume.rpandey.tech

Email: rohan [dot] pandey [at] gmail [dot] com

Website: https://rpandey.tech

GitHub: http://github.rpandey.tech

LinkedIn: http://linkedin.rpandey.tech

I've won 10 hackathons with my Fullstack expertise, am publishing an undergrad ML research paper, and am currently working on DL for NLP in my internships. My areas of research interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also just love building impactful products! Open to internships, part-time, and some full-time positions.

KhoomeiK··on Ask HN: Who wants to be hired? (August 2020)
Location: San Francisco Bay Area / San Jose / Cupertino

Remote: Open to it

Willing to relocate: Open to it

Technologies: Flask, Express, React, React Native, PostgreSQL, SQLite, PyTorch, TensorFlow, Keras, Sci-kit Learn, NumPy, Pandas, Python, JavaScript

Résumé/CV: http://resume.rpandey.tech

Email: rohan [dot] pandey [at] gmail [dot] com

Website: https://rpandey.tech

GitHub: http://github.rpandey.tech

LinkedIn: http://linkedin.rpandey.tech

I've won 10 hackathons with my Fullstack expertise, am publishing an undergrad ML research paper, and am currently working on DL for NLP in my internships. My areas of research interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also just love building impactful products! Open to internships, part-time, and some full-time positions.

KhoomeiK··on Lincos language
I've always been interested in constructed languages, but reading through this article just now made me wonder about curriculum learning for NLP models. Could better generalizable language models be achieved through curriculum learning of this sort, where simple mathematics and logic are introduced before anything else? The curriculum learning papers I've seen so far are mostly for specific tasks, like introducing simple questions for QA tasks before more complicated multi-hop reasoning.
KhoomeiK··on Scientists Discovered a Gigantic Structure in a Hidden Zone of Space
> At the largest scales it is mostly purely gravitational to my knowledge (plus dark energy which is also modeled gravitationally).

I think what OP might've been alluding to is the idea that phenomena like dark energy could have a more elegant explanation that is the result of emergent behavior at large scales. Just like magnetism was originally seen almost like magic, but it was later realized that it's simply an emergent property of the electric force i.e. electromagnetism.

KhoomeiK··on Facebook and CMU Introduce TaBERT for Understanding Tabular Data Queries
Previous discussion: https://news.ycombinator.com/item?id=23725829
KhoomeiK··on How Rice Farming Shaped Culture in Southern China (2018)
I thought about it, but the abstract does a pretty bad job of stating that hypothesis, which I think is the most interesting finding in the paper. It's definitely a bit sensationalist but I don't think you would've read the paper if that were the title. :)
KhoomeiK··on Polynesian DNA suggests epic voyage to South America 800 years ago
This is a super interesting read! In my previous conversations with Prof. Fehren-Schmitz he was highly skeptical of any Pre-Columbian contact based on his findings in [1]. He says the issue is pretty strongly politicized, because there's a modern Polynesian immigrant community in Peru that would "benefit" if admixture is found, while the Polynesians in Polynesia are a lot more skeptical due to nationalist tendencies. I might be getting this backward, not completely sure, but his point was basically that it's super hard to study this specific issue because of the surprising amount of politicization there is surrounding it.

[1] https://www.cell.com/current-biology/pdfExtended/S0960-9822(...

← PreviousPage 5 of 9Next →