yobbo

43 ms·

301.
by yobbo 3y ago
Guessing ... because of appreciating real estate and concentration of jobs, living in certain cities/areas became culturally synonymous with "being a winner" while staying in rural/industrial towns meant "being a lo
302.
by yobbo 3y ago
To put the "Marshall Plan" into a modern perspective, it was equivalent to $173 billion in 2023 dollars, over a period of four years. https://en.wikipedia.org/wiki/Marshall_Plan
303.
by yobbo 3y ago
Hm. I sense there is an expectation that "good ideas will win" if only we communicate them properly. But, when people are invested in <ideas that are orthogonal to your good ideas>, communication is not the right tool. Like
304.
by yobbo 3y ago
Maybe he thinks his own disappointments could have been avoided by better communication, and thus his advice is to "learn to communicate well, not just to code". It's not bad advice on its own, but I sense this is a naive vie
305.
by yobbo 3y ago
Also, the gradients of "softmax loss" and mean-square-error loss are the same. The network learns to optimize the same function, up until the output activation.
306.
by yobbo 3y ago
> They told their customers: "1000 Songs in Your Pocket". That is talking about specs, and it is exactly what the competition did too. Here in the unit "number of songs". Maybe most people at the time couldn't tr
307.
by yobbo 3y ago
It seems it scratches the same itch as a model railway.
308.
by yobbo 3y ago
> the belief that the amount of work is fixed. You would have to define work for this to be meaningful. Generally, demand for "work" is unlimited if the price is zero, but demand falls to zero as price increases. Humans are wil
309.
by yobbo 3y ago
> ... If no one knows about it or won't give it a chance, then it's as if it never happened. It is rational to stop somewhere between 0% and 100%. There is no guarantee you will find traction in the last 10% or the last 1%, if
310.
by yobbo 3y ago
If the position is such that it doesn't matter who is hired (meaning: there is usually no measurable difference in outcome for people in the role, and all candidates have basic qualifications), then nepotism isn't necessarily wron
311.
by yobbo 3y ago
Rather, "meritocracy" is a theatre that provides perfect alibi for nepotism. Referring to someone's CV is not in principle different from referral by a person. It's just another document that has been curated by people,
312.
by yobbo 3y ago
> I can explain it to non-experts in understandable way Ok? But you are required to motivate a decision that is based on terms, conditions, probabilities, consequences that can only be encoded at knowledge-level A. When projected down to
313.
by yobbo 3y ago
You could start reading on CMA-ES; which is something like a particle filter on the model parameters. So for 100 "particles", it means 100 resampled copies of the model, which are then evaluated to create something like a "sy
314.
by yobbo 3y ago
When gradients become auto-correlated, they are not zero mean. In that, case vₜ can grow large and uₜ becomes very small. Maybe try centering the g² term in vₜ like (gₜ-mₜ)². Also, when restarting training, m and v are probably reset to zer
315.
by yobbo 3y ago
Adam is somewhat analogous to an audio compressor on the gradient "signals". (Edit: eh sort of ...)
316.
by yobbo 3y ago
I have vague memories of reading about the coating on F-117. It was supposed to be multiple layers of epoxy paint with suspended ferrous particles, each layer cured under a different magnetic field to align them. To re-coat, the plane would
317.
by yobbo 3y ago
"Bayesian" in this context most likely means naive bayes; which assumes occurrences of all words are independent of each other. The "score" of each label ends up being something like the product of all relative word freq
318.
by yobbo 3y ago
It was a 750cc two-stroke running at WOT.
319.
by yobbo 3y ago
"Sunswift 7" doesn't look completely different. In height and volume, they could be very similar.
320.
by yobbo 3y ago
All agree, but it can save time for many trivial short-lived things.
321.
by yobbo 3y ago
To experiment with SGD and back-propagation with 4096x4096 32-bit matrices, you would need a machine with hundreds of megabytes of ram in the 90s. In terms of software, you would need to be comfortable with C/C++ or maybe Fortran to be
322.
by yobbo 3y ago
> you are not the optimal person to hold the risk that The view presented here assumes that the market prices risk (premium) arbitrarily correctly, and then argues the benefits of that. What is optimal depends on the premium and a subjec
323.
by yobbo 3y ago
> monopolies seem less ridiculous to regulators. No. Over-hiring is probably the same everywhere. When "times are good", everyone in the organisation wants to hire because their number of subordinates makes them seem more impor
324.
by yobbo 3y ago
> why should it be that certain unsupervised (naive statistical) compressions They seem to be investigating the intermediate representations of each layer in deep networks, and the degree to which those representations can linearly separ
325.
by yobbo 3y ago
Sounds functionally equivalent to a whitelist of domains you accept from?
326.
by yobbo 3y ago
"Quality" is not an indisputable property of the code. Someone's judgement about code quality can only have as much weight as their competence.
327.
by yobbo 3y ago
The second-gen 900 was a mish-mash of various Opel and Saab parts, but had Saab drivetrain, "wrap-around" wind-shield (sort of), rear-axle, etc. It developed into the first-gen 9-3. The second-gen 9-3 was GM Epsilon-something-base
328.
by yobbo 3y ago
More likely, the car had exhaust system leaks. (Ah you are referring to the two-stroke Saab 93 from 1950s!)
329.
by yobbo 3y ago
The first year for the 9000 was 1984. It was face-lifted in maybe 1991? First year of the 9-5, which could be seen as a successor, was 1997. It was not a "rebadged" GM but there might be GM-sourced components.
330.
by yobbo 3y ago
For newcomers, https://github.com/rust-lang/rustlings is a good way to start.

More ›