jedharris

9 ms·

31.
by jedharris 4y ago
Lots of excellent references showing LLMs have rich internal models such as color spaces. Good argument meaning doesn't require "grounding" in perceptions (with citations).
32.
Meaning without reference in large language models(arxiv.org)
1 points by jedharris 4y ago|2 comments
33.
Meaning without reference in large language models(arxiv.org)
2 points by jedharris 4y ago|1 comments
34.
by jedharris 4y ago
Lots of excellent references showing LLMs have internal models. Good precis of the philosophical argument that meaning doesn't require reference.
35.
by jedharris 4y ago
Probably in this case "Incentive Stock Options"
36.
by jedharris 4y ago
After a fair amount of searching I found what seems to be an equivalent powder at https://www.technoglowproducts.com/glow-in-the-dark-powder . The price is indeed better and the company offers interesting options for grain
37.
by jedharris 4y ago
From Wikipedia: The tetragonal disphenoid tetrahedral honeycomb is a space-filling tessellation (or honeycomb) in Euclidean 3-space made up of identical tetragonal disphenoidal cells. https://en.wikipedia.org/wiki/Tetra
38.
by jedharris 4y ago
The term "illusion" here is confusing -- see most of the comments. Consider a rainbow. It looks like a colored thing in the sky. If it is complete enough it seems to come down to earth at specific points. We even talk about "
39.
by jedharris 4y ago
Maximum Likelihood Estimation
40.
by jedharris 5y ago
Changing the laws or enforcement will take a while and software won't help directly. But extensive crowd-sourced measurement of externalities would be a big deal and software should help with that. Actual measurement (and definition) o
41.
by jedharris 6y ago
I'm sure the last few Fed chairs would like to know that they are guaranteed success in raising inflation! Now you just have to explain why they weren't even able to get it up to 2% even with a guarantee.
42.
by jedharris 6y ago
In effect we somewhat have this kind of system with Food Stamps and housing vouchers -- "currencies" that are restricted to certain types of expenditures. Also of course there are advocates for education vouchers. Etc. Unfortunate
43.
by jedharris 6y ago
Great point about TIPS! Effectively everyone who's arguing that we're in for catastrophic inflation doesn't trust the wisdom of the market as much as their own reasoning based on Economics 101 models. I wonder what that tells
44.
by jedharris 6y ago
You say that in the 1980s "most countries moved to fractional reserve banking rather than full reserve". This is at best confusing: in most developed countries fractional reserve banking, regulated by central banks, goes back to
45.
by jedharris 6y ago
The GFC does not count. The GFC was caused by (1) bad underwriting policies (banks, AIG) that generated a huge debt overhang and (2) bad and poorly understood dependencies between banks that led to breakdown of the normal flows of assets
46.
by jedharris 6y ago
Key observation. In recessions (liquidity crunches) the cake grows more slowly. Maybe if we increase liquidity a lot (i.e. print money) and give it to the bottom tiers who will spend it (increase velocity) then the cake will grow faster! We
47.
by jedharris 6y ago
You say "There is no magic or mystery here. This is not complicated. It is not new, unknown and to be determined." Clearly wrong. Specifically consider the velocity of money. If it falls as the supply increases there will be no in
48.
by jedharris 6y ago
As Noah says in the article. The question he's addressing is how to know when we are approaching the cliff. We all agree it is out there but we can have a much better society if we know how far. If we fall off the cliff of course tha
49.
by jedharris 6y ago
Recently the Fed hasn't been good at getting inflation to 2% (their published target) and economists haven't been good at predicting whether the Fed will succeed, or modeling why the Fed has failed. Noah (article) doesn't emp
50.
by jedharris 6y ago
This kind of behavior can kill the legal system pretty easily. Responding takes much more time and effort than committing the offenses in the first place, the sanctions have little effect, and the lawyer continues to practice. Probably he&#
51.
by jedharris 6y ago
Same topic as a year ago, but deserves much more examination than it got then.
52.
by jedharris 6y ago
This paper is a great overview of the state of the art in building flexible generalizable deep learning systems. Not easy to read, but if you puzzle through it you'll understand where the field is and where it is going.
53.
by jedharris 6y ago
This paper drives a stake through the heart of claims that explicit symbolic reasoning is needed for intelligent systems. It builds extra-symbolic (deep learning) systems that substantially outperform symbolic reasoning on tests specificall
54.
Object-based attention for spatio-temporal reasoning(arxiv.org)
2 points by jedharris 6y ago|2 comments
55.
by jedharris 6y ago
The paper doesn't help to answer: Why now? The model it proposes would have produced high polarization any time, but polarization is much higher now than 40 years ago... If the model could help us analyze the dynamics of polarization i
56.
by jedharris 6y ago
The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play. See < https://blog.janestreet.com/deep-learning-the-hardest-go-pro... . This covers developments since the
57.
by jedharris 6y ago
Related demos, repositories, papers and posts: More recent and extensive post: https://blog.janestreet.com/deep-learning-the-hardest-go-pro... Site to play against KataGo: https://online-go.com/player/5
58.
Much More Efficient Self-Play Learning in Go(blog.janestreet.com)
2 points by jedharris 6y ago|2 comments
59.
by jedharris 7y ago
Why didn't the investigator subpoena the details about the accounts sending the accusations? That would have shown right away that the accusations were not "credible" and the university could have just dropped the investigati
60.
by jedharris 7y ago
Could the university investigation subpoena the information behind the email accounts? If not, why not? This seems like a basic investigative move, and unless they are not allowed to do it, makes me doubt very much that they were serious ab

More ›