tdullien

9 ms·

61.
by tdullien 2y ago
Or casual references to stomping live turtles.
62.
by tdullien 2y ago
"Somebody needed to get promoted"
63.
by tdullien 2y ago
Confidentiality agreements are usually part of employment agreements. I haven't checked expiry dates etc.; and tbh I'm not sure they'd apply here, but I'd rather check ;)
64.
by tdullien 2y ago
Xoogler here (2011-2018). At some point I proposed making it easy for people to "lie" to an app (if it asks for location, provide fake data etc.). This would preserve true customer choice about anonymity. The reaction to that idea
65.
by tdullien 2y ago
Google and Meta do in-production profiling. I think that tech is coming to everyone else slowly.
66.
by tdullien 2y ago
The issue is that in order to optimize for human readability you'll need a huge number of human evaluations of graphs?
67.
by tdullien 2y ago
My father had a brain hemorrhage 2 years ago and now dementia. Watching the only video I have of him pre accident, there is a very marked difference in eye movements.
68.
by tdullien 2y ago
Eric Schmidt does not have good insight into anything happening on the factory floor at Google post-2013/2014. I'd take anything he says with a huge grain of salt. I was at Google march 2011- dec 2018 and when I tried reading Schm
69.
by tdullien 2y ago
I've used it for years for vulnerability development (when you need to debug why a particular exploitation run went wrong etc.). It is indeed marvelous for many situations; except for concurrency bugs ;)
70.
Fast-takeoff AI Doom scenarios are implausible(addxorrol.blogspot.com)
1 points by tdullien 2y ago|0 comments
71.
by tdullien 2y ago
I think it is fair to insist on what you were initially promised, but also make sure both sides understand each others' motivations properly. If both parties work in good faith, usually finding a point of contention in a contract draft
72.
Some geometric intuition for single-layer ReLU networks(addxorrol.blogspot.com)
4 points by tdullien 2y ago|2 comments
73.
by tdullien 2y ago
I ran some experiments to firm up my geometric intuition about ReLU networks.
74.
by tdullien 2y ago
My question is: what could possibly be so important that you want to take it with you to have your hands full in an emergency potential life-or-death situation?
75.
by tdullien 2y ago
Frankly, if in a rare life-or-death situation you care about your hand luggage, I'd recommend carefully reflecting on life choices and priorities.
76.
by tdullien 2y ago
Well, I cannot help your inclination to take Eliezer's side, but I would like to point out all the anti-physics stuff in the original article: 1) Eliezer makes up a term ("diamondoid bacteria"), and the current scientific und
77.
by tdullien 2y ago
If you Google a few words you'll find the original source and context. Viruses don't replicate without host cells, there's no "nano factory" involved, and ... do tell me how to implement an accurate timekeeping mech
78.
by tdullien 2y ago
I was brought up by Keynesian folks in much the same spirit. I am not sure the author isn't conflating "libertarian economists" with "good parents".
79.
by tdullien 2y ago
"The nanomachinery builds diamondoid bacteria, that replicate with solar power and atmospheric CHON, maybe aggregate into some miniature rockets or jets so they can ride the jetstream to spread across the Earth's atmosphere, get i
80.
by tdullien 2y ago
A vast number if scenarios advocated by Yudkowski and adjacent people are anti-physics and rely on the world as we know it to operate under vastly different principles than we think right now. I think there's zero reason to think AIs w
81.
by tdullien 2y ago
Oh, I wasn't aware people seriously advocated that. I've heard people argue LLMs are approximate search engines and their results kernel-smoothed combinations of multiple terms in the index and the embedded prompt (and I can see s
82.
by tdullien 2y ago
I've worked on zero days for decades and friends of mine with great academic pedigrees and deep experience in both AI and security are uniformly frustrated with the inability of even cutting-edge LLMs to reason in the way required fo
83.
by tdullien 2y ago
I think there are more than enough people with limited domain understanding imagining that AI can do all sorts of things that it won't be able to do any time soon, if ever. There's enough Yudkowskis fabulating the impossible and s
84.
by tdullien 2y ago
Iirc at the time there was still "at fault divorce", and the divorce settlement was influenced by who was at fault. So a (likely overly charitable reading) of the entire thing is that Feynman agreed to be portrayed in a bad light
85.
by tdullien 2y ago
This is the right way to look at it.
86.
by tdullien 2y ago
How about a chronically unemployed mathematician that is flooded with job offers to work as a miner in a coal mine?
87.
by tdullien 2y ago
It's going to play poorly when C++ exceptions are thrown/caught.
88.
by tdullien 2y ago
Important to keep in mind that these returns ceased to be compounding quickly: they restarted from scratch 10bn each year to score 30%. Successful quant stratégies tend to hit capacity limits...
89.
by tdullien 2y ago
How does this compare to SAGE?
90.
by tdullien 2y ago
It's not even clear if weights are copyrightable in the first place, so no.

More ›