HNHacker News
TopNewBestAskShowJobs

KhoomeiK

1,299 karma · joined August 7, 2018

research @ openai

https://rpandey.tech

submissionscomments
KhoomeiK··on Knowledge Representation in Sanskrit and Artificial Intelligence
Linguistics: Astadhyayi of Panini, the first generative grammar and arguably kickstarted the Western structural linguistic tradition via Saussure

Logic: Navya Nyāya developed formal logical approaches and proof paradigms 600 years before the West

Poetics: Pingala mathematically defined poetic meters with combinatorics (DeBruijn Sequences, Pascals triangle, etc) over 2000 years ago and the Sanskrit poetic tradition is widely recognized as one of the most aesthetically sophisticated. Check out the Sisupala Vadha of Magha for crazy poetry stuff.

Astronomy: Gupta-era astronomers developed the base 10 positional numeral system to make calculations easier, and as a result, Indian calendar systems have accounted for the precession of the Earth for over a thousand years

Rhetoric: the Nyāya school developed a formal theory of debate that was used in intellectual circles across the subcontinent for centuries. It enabled debaters to identify argumentative strategies and logical fallacies.

Mineralogy and Geography: there's a lot of Puranic literature about this but I'm personally not too familiar

Statecraft: Kautilya's Arthasastra is the most famous example but there are countless other treatises over the ages discussing politics and warfare

Each of these examples I gave are but a small sample of the work done by Sanskrit intellectuals over the past few millennia. Don't blame your ancestors for your own ignorance.

KhoomeiK··on Knowledge Representation in Sanskrit and Artificial Intelligence
> primarily used for religious purposes

*intellectual purposes. The constant religion-washing of Indian intellectual history is colonial-era propaganda, for lack of a better term. It has gone so far that now the vast majority of people think there is nothing more to India's intellectual past than Hinduism. A very substantive portion of Sanskrit literature, if not the majority, deals with non-religious topics such as linguistics, logic, poetics, astronomy, rhetoric, mineralogy, statecraft, geography, and more.

I agree with your main point though.

KhoomeiK··on Ask HN: Who wants to be hired? (December 2021)

  Location: Silicon Valley / San Francisco Bay Area (CA), Pittsburgh (PA)
  Remote: Open to it
  Willing to relocate: Open to it
  Technologies: Python, PyTorch, TensorFlow, Keras, HuggingFace, SpaCy, NLTK, NumPy, Pandas, Sci-kit Learn, Flask, JavaScript, Express, React, React Native, PostgreSQL, SQLite
  Résumé/CV: https://resume.rpandey.tech
  Email: '%s.%s@%s.com' % ('rohan', 'pandey', 'gmail')
I'm an undergrad at Carnegie Mellon who's won over 10 hackathons with my full-stack experience and published machine learning research papers at top venues like NeurIPS and EMNLP.

My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Genomics but I also enjoy building impactful products!

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

https://rpandey.tech

KhoomeiK··on Ask HN: What are your 2022 predictions?
Figure 3.8 in the GPT-3 paper (SuperGLUE Performance) pretty clearly shows a plateau in performance, with the increase in performance from 13B to 175B parameters (13.5x params, +6 score) not being significantly different from the increase between 6.7B to 13B (1.9x params, +4 score).
KhoomeiK··on Ask HN: What are your 2022 predictions?
Sorta predictions, sorta hopes...

GPT-4 or another huge language model like it will be released, and we will see diminishing returns continue to plateau.

Graph-based approaches in NLP will continue to grow in significance, with KGs, ontologies, and semantic graphs seeing growing adoption for their interpretability.

Model robustness in ML will probably remain a significant obstacle to productionization, holding back use-cases that need high reliability e.g. self-driving and front-facing chatbots.

Reinforcement Learning may see increased research focus for applications in NLP and user interaction.

KhoomeiK··on In the 17th century, Leibniz dreamed of a machine that could calculate ideas
> if he could find a way to symbolically represent these fundamental concepts and develop a method by which to combine them logically, then he would be able to generate new thoughts on demand

Reminds me of some recent neurosymbolic work, like "DreamCoder: Growing generalizable, interpretable knowledge with wake-sleep Bayesian program learning" [1]

[1] https://arxiv.org/abs/2006.08381

KhoomeiK··on Brain implant translates paralyzed man's thoughts into text with 94% accuracy
Previous threads:

https://news.ycombinator.com/item?id=27134049

https://news.ycombinator.com/item?id=27157369

KhoomeiK··on Ask HN: Who wants to be hired? (November 2021)

  Location: Silicon Valley / San Francisco Bay Area (CA), Pittsburgh (PA)
  Remote: Open to it
  Willing to relocate: Open to it
  Technologies: Python, PyTorch, TensorFlow, Keras, HuggingFace, SpaCy, NLTK, NumPy, Pandas, Sci-kit Learn, Flask, JavaScript, Express, React, React Native, PostgreSQL, SQLite
  Résumé/CV: https://resume.rpandey.tech
  Email: '%s.%s@%s.com' % ('rohan', 'pandey', 'gmail')
I'm an undergrad at Carnegie Mellon who's won over 10 hackathons with my full-stack experience and in the process of publishing a few machine learning research papers.

My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Genomics but I also enjoy building impactful products!

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

https://rpandey.tech

KhoomeiK··on Ask HN: Who wants to be hired? (October 2021)

  Location: Silicon Valley / San Francisco Bay Area (CA), Pittsburgh (PA)
  Remote: Open to it
  Willing to relocate: Open to it
  Technologies: Python, PyTorch, TensorFlow, Keras, HuggingFace, SpaCy, NLTK, NumPy, Pandas, Sci-kit Learn, Flask, JavaScript, Express, React, React Native, PostgreSQL, SQLite
  Résumé/CV: https://resume.rpandey.tech
  Email: '%s.%s@%s.com' % ('rohan', 'pandey', 'gmail')
I'm an undergrad at Carnegie Mellon who's won over 10 hackathons with my full-stack experience and in the process of publishing a few machine learning research papers.

My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also enjoy building impactful products!

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

https://rpandey.tech

KhoomeiK··on Ludwig Wittgenstein: A Mind on Fire
If we're talking about philosophy of language, Buddha himself didn't have much formal reasoning in that, did he?
KhoomeiK··on Has AI found a new foundation?
> GPT-3 does what it says it does on the sticker. It continues text based on what it reads.

The problem is that this isn't what it "says on the sticker" to the vast majority of people, especially outside of academia. What we think of as a simple probabilistic language model trained on a huge corpus of English, the general public thinks of as "an artificial intelligence" that has logical reasoning capabilities etc. Using the term "foundation models for AI" intentionally seems to obscure this fact by implying a direct link between language models and AI, when in reality, an LM is nothing more than a faculty for syntactically (and vaguely semantically) acceptable language generation.

KhoomeiK··on Ask HN: Who wants to be hired? (September 2021)

  Location: Silicon Valley / San Francisco Bay Area (CA), Pittsburgh (PA)
  Remote: Open to it
  Willing to relocate: Open to it
  Technologies: Python, PyTorch, TensorFlow, Keras, HuggingFace, SpaCy, NLTK, NumPy, Pandas, Sci-kit Learn, Flask, JavaScript, Express, React, React Native, PostgreSQL, SQLite
  Résumé/CV: https://resume.rpandey.tech
  Email: '%s.%s@%s.com' % ('rohan', 'pandey', 'gmail')
I'm an undergrad at Carnegie Mellon who's won over 10 hackathons with my full-stack experience and in the process of publishing a few machine learning research papers.

My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also enjoy building impactful products!

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

https://rpandey.tech

KhoomeiK··on Ask HN: Who wants to be hired? (August 2021)

  Location: Silicon Valley, San Francisco Bay Area
  Remote: Open to it
  Willing to relocate: Open to it
  Technologies: Python, PyTorch, TensorFlow, Keras, HuggingFace, SpaCy, NLTK, NumPy, Pandas, Sci-kit Learn, Flask, JavaScript, Express, React, React Native, PostgreSQL, SQLite
  Résumé/CV: https://resume.rpandey.tech
  Email: '%s.%s@%s.com' % ('rohan', 'pandey', 'gmail')
I'm an undergrad at Carnegie Mellon who's won over 10 hackathons with my full-stack experience and in the process of publishing a few machine learning research papers.

My areas of interest include Deep Learning, Natural Language Processing, Neurotechnology, and Bioinformatics but I also enjoy building impactful products!

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

https://rpandey.tech

KhoomeiK··on Mammals dream about the world they are about to experience before they are born
Reminds me of recent Reinforcement Learning work on learning from internal world model "dreams".

https://worldmodels.github.io/

KhoomeiK··on Electromagnetism is a property of spacetime itself, study finds
Aesthetics, harmony, and simplicity are the guiding principles of mathematical conjecture. This piece is simply hypothesizing about a link while calling for further empirical research.
KhoomeiK··on Mysterious DNA sequences, known as ‘Borgs,’ recovered from California mud
I think you might be severely underestimating the level of parallelization happening in living beings. There are 100 trillion bacteria in every single human. Extrapolated to the number of bacteria on the planet, that's a whole lot of computational ability.
KhoomeiK··on Interpreting 9/11 commemorative Afghan carpets (2011)
The way in which the design reverts to the norm is so profound in a way, even though it's the product of "mistakes". Entropy, evolution, and economies follow parallel patterns.
KhoomeiK··on A call to minimize distraction and respect users’ attention (2013)
> So I never could figure out why our marketing team was all about engagement other than they kept hearing that word said about other digital products and, so, naturally it applies to ours as well and we must do what everyone else is doing!

"Cargo cult marketing"?

KhoomeiK··on What did ancient languages sound like?
Yes but it's still present in Vedic, literally in the second word of rgveda 1.1.1.
KhoomeiK··on What did ancient languages sound like?
I've listened to rgveda samhita patha of what I believe is the shakala shakha which is predominant in north india but even they seem to mispronounce the vocalic "r" as "ri".
KhoomeiK··on What did ancient languages sound like?
Lithuanian in general is considered one of the most conservative IE languages. There isn't really any clear reason that historical linguistics has produced for whether languages change quickly or slowly afaik.
KhoomeiK··on What did ancient languages sound like?
I agree with the sentiment but ironically your transliteration of the name of that language reveals that it certainly has undergone at least some sound changes. Why do people in the north pronounce the vocalic "r" as "ri" while in the south they use "ru"? Why do people in the north drop word final schwa but in the south they add a nasal bilabial?

While it's very easy to pronounce Samskrta in a way similar to what was likely used in classical times, there have definitely been sound changes to an extent.

KhoomeiK··on Thinking Like Transformers
I believe [1] is the work they're mainly referencing for their RNN equivalent/inspiration.

[1] https://nlp.stanford.edu/~johnhew/rnns-hierarchy.html

KhoomeiK··on Ask HN: What things that tech recruiters do, annoy you the most as an engineer?
Reaching out to me about Senior SWE positions when I'm literally an undergrad
KhoomeiK··on Dog ejected from car during Sunday crash found on sheep farm, herding sheep
You don't need hundreds of thousands of years to have meaningful shifts in allelic distribution. 10,000 years of agriculture is more than enough to select for height, lactase persistence, and more [1]. I don't think it's a far shot to conjecture a relationship between certain modes of civilization and cognitive features that they select for. I can't point to conclusive empirical evidence for this but my problem is just with people making the study of these issues taboo and denying the scientific possibility.

[1] https://www.sciencedaily.com/releases/2015/11/151123202631.h...

KhoomeiK··on Dog ejected from car during Sunday crash found on sheep farm, herding sheep
It's impossible that cross-cultural variations in societal structure or environment imparted a selection bias acting upon certain cognitive traits that would explain the observed disparities we see in modern data?
KhoomeiK··on Dog ejected from car during Sunday crash found on sheep farm, herding sheep
It blows me away that people can see things like this and still deny any genetic basis for certain cognitive features in humans.
KhoomeiK··on Scottish Café
Great metaphor for how to create an innovative, collaborative culture.
KhoomeiK··on An Introduction to Knowledge Graphs
Cool! I'm an incoming undergrad transfer to CMU this fall interested in graph-based NLP. Mind if I email you in the next couple days?
KhoomeiK··on An Introduction to Knowledge Graphs
Recent work suggests it's possible to generate knowledge graphs from large corpi of text encoded with a language model: https://arxiv.org/abs/2010.11967
← PreviousPage 4 of 9Next →