nborwankar

6 ms·

1.
by nborwankar 2y ago
Not sure if this article provides enough definitive info - happened to see it yesterday. https://www.newscientist.com/article/mg26335104-500-the-brai...
2.
by nborwankar 2y ago
Seems ironic to be working for a company without integrity, drawing a salary ie your livelihood from such an organization while believing somehow your personal integrity is not already impacted by doing so. The two seem incompatible to an o
3.
by nborwankar 2y ago
Can you share what kinds of problems were conducive to hyperbolic embeddings in your experience. Also, separately, are you saying companies are using these in practice but don’t talk about them because of the advantage they give? Or am I re
4.
by nborwankar 2y ago
They have vacuum-cup like pads on their feet?
5.
by nborwankar 2y ago
+1 for Mathematica’s documentation system. It is the best of any software doc, OSS/free or closed/commercial. By far.
6.
by nborwankar 2y ago
This is related to a classic relational database problem called “bill of materials problem” - suggest you Google that. Also look for book on Graphs and networks in SQL by Joe Celko where he goes into modeling graphs in SQL tables. Naive fo
7.
by nborwankar 2y ago
Gives a whole new meaning to “manufactured consent“.
8.
by nborwankar 2y ago
If it could support the pgvector extension it would be a super fast vector database with all the power of Pg - the relational aspect brings the ability to add and query using rich domain specific metadata usually contained in relational dat
9.
by nborwankar 2y ago
+1 on fastmail
10.
by nborwankar 2y ago
I didn’t take a year fully off but last year was slow in my consulting work so I did a deep dive on the emerging LLM area. One recommendation - get a beefy Mac laptop so you can run LLM’s locally. I got an M2 with 96G RAM. Makes a huge diff
11.
by nborwankar 2y ago
Slightly unconnected but I’ve found a rather disturbing “dark pattern” in Amazon sales of food items in CA. Background: CA requires a warning on food that has some threshold level of carcinogenic or harmful substance - esp lead and arsenic.
12.
by nborwankar 2y ago
Programmers Introduction to Mathematics by Jeremy Kun should get you started. Look on Amazon. Then you can decide if you want to do a formal course and which one. I would also strongly suggest you buy yourself a personal copy of Mathemat
13.
by nborwankar 2y ago
Here’s another one - it’s older but has some interesting charts and graphs. https://arxiv.org/abs/2303.18223
14.
by nborwankar 2y ago
That’s not quite correct. Private data especially in the enterprise cannot use public LLM’s like GPT-4 or 5 or N. Use cases needing data privacy have to use an internally implemented LLM application. In Currently, RAG is a concrete and pra
15.
by nborwankar 2y ago
See Large Language Models - a Survey. Minaee et al https://arxiv.org/abs/2402.06196
16.
by nborwankar 2y ago
If the robots are cloud connected we are doomed the robot apocalypse could be triggered by cyber warfare. Or by malicious human owners. If they are autonomous there better be an emergency “wake word” that instantly disables the robot. In t
17.
by nborwankar 3y ago
The secular core of Buddhist thought has parallels in the Sankhya system of thought in the set of knowledge systems collectively called Hindu metaphysics. Buddhism and Hinduism had active and passionately debated exchanges of ideas leading
18.
by nborwankar 3y ago
The question "Can you hear the shape of a drum?" was first posed by mathematician Mark Kac in a paper published in 1966. It was proven in the 80’s that there are manifolds for which you can’t.
19.
by nborwankar 3y ago
“Integer arithmetic is all you need” ? NVIDIA stock arrow up or down?
20.
by nborwankar 3y ago
Re RAG aren’t you ignoring the fact that no one wants to put confidential company data into such LLM’s. Private RAG infrastructure remains a need for the same reason that privacy of data of all sorts remains a need. Huge context solves the
21.
by nborwankar 3y ago
You do get some money back for a trade-in depending on how old the machine is.
22.
by nborwankar 3y ago
Llama and GPT are auto-regressive decoder only architectures which for pure translation jobs are not the optimal architectures. Training seq2seq models or encoder/decoder models on datasets of sentence pairs designed for translation wi
23.
by nborwankar 3y ago
Got 502 Bad Gateway
24.
by nborwankar 3y ago
Here’s a thread from biology stackexchange about this which also refers to CETI. https://biology.stackexchange.com/questions/108499/can-whale... .
25.
by nborwankar 3y ago
Meta comment about the comments saying “I didn’t find what I was looking for” - core fact is it was using Wikipedia as a knowledge base. So if your topic is not represented well enough in Wikipedia it’s not going to return good results. Sec
26.
by nborwankar 3y ago
You don’t have to look far. What happens when everyone can write a letter to the editor and have it instantaneously published? What happens when everyone has a handheld camera and can publish huge amounts of visual crud? Do we really savor
27.
by nborwankar 3y ago
I have a fundamental disagreement about the premise that abstraction from CPU instructions should be the basis of a language model. I posit that the starting point should be lambda calculus which is formally equivalent to Turing machine bas
28.
by nborwankar 3y ago
Shouldn’t this be called Regularized SoftMax? Adding 1 in the denominator looks a lot like a regularization in other ML contexts.
29.
by nborwankar 3y ago
“Everyone is ignorant but in different things” Feynman.
30.
by nborwankar 3y ago
CPU only?

More ›