sdenton4

59 ms·

961.
by sdenton4 4y ago
But there's a massive industry of film shot in Canada. Start watching the credits and the supporting actors are very often Canadian, with projects shot in Vancouver or Toronto and 'local' hires to fill out the cast.
962.
by sdenton4 4y ago
I find the intuition extremely helpful - without the intuition, it's hard to build a gut feeling for the interrelation and relative importance of the bits in the jargon soup for any given question - apijove vs ascending node, and so on
963.
by sdenton4 4y ago
I think a closer-to-real example is the problem of playing a collection of blackjack tables. All the games are the same, but sometimes the state of the decks (ie, which cards have been discarded) will lead to better odds of winning. If you
964.
by sdenton4 4y ago
I think this is a bit over confident, though... For example, AlphaFold just fundamentally isn't a language model, but is fundamentally useful. We'll still need these models that Do Stuff in many areas, and that will still involve
965.
by sdenton4 4y ago
It's still problematic. The HEAR benchmark is a great eye-opener. They have basically three classes of audio tasks, and find very different models excel in each, with the best overall models being kinda-mindless ensembles of the ones t
966.
by sdenton4 4y ago
Sadly a lot of foundational ML research works for single-label image classification and not much else. ImageNet is a niche problem and way too much ML research is over-indexed on it. If you can make your problem look like ImageNet, you&#x27
967.
by sdenton4 4y ago
Any stateful function can be made stateless by turning the state into an argument to the function and the updated state into an additional return value: y := f(x) becomes y, z' := f(x, z). The usual implementation for LLMs are stateles
968.
by sdenton4 4y ago
Why is that a prerequisite for intelligence? Suppose we could upload + fully simulate a human mind. Now suppose that we chose to spin it up, run it briefly, then throw away all the state every time we wanted to interact with it. Would the
969.
by sdenton4 4y ago
There are piles of intermediate hidden activations involved in encoding the conversation and producing the next outputs, of course. "Internal" state also gets a bit weird as a concept when we can look at any activations we want.
970.
by sdenton4 4y ago
The same way we reacted to the incredible deluge of spam phone calls in the US: we'll move to platforms that have the best signal-to-noise ratios.
971.
by sdenton4 4y ago
ICE engines only manage to turn ~15% of the stored energy in gasoline into actual work. A bit of googling suggests that jet engines are about 35% efficient. Stored electricity is much more efficiently turned into mechanical work... Electric
972.
by sdenton4 4y ago
The pickles drive me crazy. Tensorflow saved models are a great way to solve the problem... Save the computation graph and weights, and drop all the crusty code dependencies. I think ONNX models are similar. I expect there should be a Jax e
973.
by sdenton4 4y ago
Why should I need to? Isn't God on the blockchain? (j/k)
974.
by sdenton4 4y ago
It's not strictly a question of trying Everything and seeing what sticks. There's a hypothesis in the parent comment - better handling of the time domain will lead to better modeling - which is actually fairly independent of archi
975.
by sdenton4 4y ago
Meh, people have been trying to ape biological networks for decades and they consistently produce poor functioning versions of the leading edge... Why wouldn't we have already seen spiking networks contribute something by now?
976.
by sdenton4 4y ago
garage : startup :: mom's kitchen : novel junkfood
977.
by sdenton4 4y ago
A random nxn matrix is full rank... So it's kinda the default: any amount of noise in the embedding is going to result in full-rank transformations. So it's really less-than-full rank which would require an explanation - ie, why d
978.
by sdenton4 4y ago
I would go with the police, health care 'system' or the curiously large proportion of the population who think qanon > democracy, but you do you, I guess.
979.
by sdenton4 4y ago
Curiously, I just bought an hdd for the first time in years, after noticing I could get 22TB of storage for $300. My use case is penalty a bit particular, though... It's really just for localized copies of large datasets that exist els
980.
by sdenton4 4y ago
And in a thousand years historians will look back at baseball, a well-known sports game with ritual overtones. The team known roughly as 'the bloody foot-mittens' is known to have been cursed by the gods for over half a century.
981.
by sdenton4 4y ago
Since 'gets anything wrong, ever' is the current goalpost for agi (per the Gary Marcus methodology), we have to judge human intelligence by the same stick. Since the author of this article misunderstood the gpt release process, th
982.
by sdenton4 4y ago
All of the UX interface have little up/down thumb icons... that's where the boolean feedback comes from. If people stop using that, sentiment analysis on the human responses will likely go a long way.
983.
by sdenton4 4y ago
That's the point of having multiple realizations of the same underlying model. The (depthwise) convolutional realization is extremely efficient for training, and the RNN is extremely efficient for inference. The scaling in both of thes
984.
by sdenton4 4y ago
Go read about S4, from these authors. It's about having a learnable state-space model which can be efficiently implemented as either an RNN or (very long) convolution, according to the needs of train or inference.
985.
by sdenton4 4y ago
There's enormous turnover is teaching due to too large class sizes and general bullshit; I imagine some combination of removing bullshit and raising pay will make it easy for many people with credentials to come back. One can also take
986.
by sdenton4 4y ago
Seems like the trick is to use the LLM to produce a query - then the data you get back is simply the result of the query, and you can examine the query to check that it's appropriate.
987.
by sdenton4 4y ago
I mean, if we're worried about ai taking all the jobs and concentration of wealth, just raise taxes on the billionaires and hire more teachers... Getting class sizes to 10:1 or 5:1 is quite the jobs program, and will lead to amazing lo
988.
by sdenton4 4y ago
Here's some incontrovertible evidence of the effectiveness of the Superman position. https://youtu.be/99K4gpJZWdw
989.
by sdenton4 4y ago
And Alphaville! https://en.m.wikipedia.org/wiki/Alphaville_(film)
990.
by sdenton4 4y ago
Compare: Omit needless words.

More ›