gradys

8 ms·

121.
by gradys 6y ago
I'm super interested in this area, with my own vaporware attempt at building it (fully abandoned, unlike yours). I'm an ML engineer focused on NLP applications. Contact info in my profile if you ever want to chat, e.g. about diffe
122.
by gradys 6y ago
IMO, the ClojureScript SPA ecosystem is a stronger competitor in its domain than Clojure on the backend. Re-frame in particular is a gem. It's as though someone tried the React/Redux stack, thought long and hard about actions and
123.
by gradys 6y ago
I worked on project very much like this last summer, a transformer language model applied to code completion. You'd be surprised how easy it is to get a model that performs as well as what you see in the video. And it's even easie
124.
by gradys 6y ago
I'd put it differently. This is going to take your job, just like an assembly programmer from the 70s might consider Python to have basically taken their job. In software, the job is constantly eating itself and transforming. It's
125.
by gradys 6y ago
Curious what analysis you're referencing here. While I don't doubt people have shown that various transformer models have certain limitations, I'm pretty bullish on transformer models in general. Here's a post exploring
126.
by gradys 6y ago
I'm late to this party, but this is fantastic. I've popped it into the Firefox Stylus addon. Consider me a happy user. Thanks so much!
127.
by gradys 7y ago
> Tech companies unnecessarily demand butts in chairs in the Bay Area, foisting the externalities onto the commons (infrastructure and housing). I'm not so sure it's unnecessary. I (and my employer) find it very useful that I h
128.
by gradys 7y ago
The highest level of severity is reserved for anything that allows users to uninstall the crapware they ship with their phones.
129.
by gradys 7y ago
T5 is another architecture from Google that set the SOTA on GLUE and SuperGLUE this year: https://arxiv.org/abs/1910.10683 It uses a Transformer and a BERT-style masked language model loss, but is fairly different in o
130.
by gradys 7y ago
> The census’ annual American Communities Survey defines a home as vacant if there is either no occupant or a temporary occupant—temporary meaning “people who will be there for two months or less.” So this is counting AirBnBs and other s
131.
by gradys 7y ago
The flip side is that if companies are making less profit, there might be more consumer surplus, leading to more savings (which can also be reinvested in the economy) or a higher standard of living.
132.
by gradys 7y ago
Parameters are what we update during training. One parameter is one scalar among the learned variables of the model. Audio-visual inputs do sometimes end up having 1B+ individual values, but there isn't necessarily a 1:1 relationship b
133.
by gradys 7y ago
For anyone looking for a quick summary of what a lottery ticket is, how they're found, and some implications, here's what I remember from a I talk I saw about lottery tickets at EMNLP 2019: - The core hypothesis is that within &qu
134.
by gradys 7y ago
I believe Zuckerberg's answer to that question was that such an ad would be banned. It's an explicit exception in the policy that permits lying in political ads, an exception that bans lies in service of voter suppression. To the
135.
by gradys 7y ago
If someone has hacked your computer such that they can control your camera, how worried are you about being fingerprinted?
136.
by gradys 7y ago
This feels hollow. Can't this be said about any benchmark? It seems natural and proper that as one benchmark becomes saturated, we introduce harder benchmarks. I don't think anyone in the field thinks that once we match human perf
137.
by gradys 7y ago
Totally with you on supoorting ranked-choice voting. It seems like a clear and obvious upgrade over what we're doing now in ways that can be mathematically verified. Question though - why prefer the Bourda Count over a Condorcet method
138.
by gradys 7y ago
Thanks for the excellent and detailed reply! Are there any papers you recommend im particular? One of the harder parts of getting into a new domain of science is that it's hard to tell the good/important papers from the others.
139.
by gradys 7y ago
People who work in or near genetic engineering: CRISPR seems to solve the problem of editing DNA to contain arbitrary human-designed data. Are there major caveats or limitations to this that people outside the field don't recognize? e.
140.
by gradys 7y ago
My understanding of their rationale for the disclosure policy is that while this model might not necessarily be so incredibly dangerous, 1) There was a non-zero chance that their releasing it would do more harm than good by making things li
141.
by gradys 7y ago
Here's Krafcik reacting to how what he said there was quoted and interpreted: "Yeah some context missing here, but it did make for a fun headline.( said the same thing about my own driving.) The point is that autonomous driving, l
142.
by gradys 7y ago
Could you describe some potential legislation you'd be happy with? How would you define "properly"? I'm especially interested in how this would work in places like the United States where there are strong free speech law
143.
by gradys 7y ago
Why would it lead to those things? Building more housing seems like it would lead to fewer homeless people in the long run.
144.
by gradys 7y ago
IANAL, but I took a commerical law class in college. I feel like I remember there being a doctrine that if there's a written contract, agreements outside of the contract don't matter. (Which is not to say that verbal contracts are
145.
by gradys 7y ago
The analogy Go : C++ :: Go's governance : C++'s governance works pretty well. Maybe C++ is a good example of this, and maybe that is a good way to build a language like C++, but it might be a bad way to make a language lik
146.
by gradys 7y ago
Apply s/car/phone and it was all true in the mid-2000s as well. Maybe they saw an angle you're not.
147.
by gradys 7y ago
> They could always find more money by charging higher fares... It's not clear that they'd actually end up with more money as a company or more money for drivers this way. I'd expect they could pay many fewer drivers somew
148.
by gradys 7y ago
My dad is a real estate agent (primarily for residential properties), and at least when I was growing up, it was 3% each for the buyer and seller's agent.
149.
by gradys 7y ago
Framing it as "human psychological experiments" evokes famous psych experiments like the marshmallow test and the Stanford Prison Experiment. Tech companies run these kinds of experiments occasionally (I believe Facebook is known
150.
by gradys 8y ago
See also Eliezer Yudkowsky's excellent response to that article: https://intelligence.org/2017/12/06/chollet/

More ›