nmca

7 ms·

121.
by nmca 3y ago
GDS - government digital services - always seems shockingly sane, a d it shows in lots of online forms etc.
122.
by nmca 3y ago
I'll be ecstatic when the inevitable rust-backed python typechecker arrives, python type checking has been excruciatingly slow in the various places I've used it. Ruff is a great start though!
123.
by nmca 3y ago
Whilst a lovely achievement, this title definitely overstates current robot abilities. Here is some human parkour for contrast: https://m.youtube.com/watch?v=QHqAVaQqQWQ&t=106s
124.
by nmca 3y ago
I think this is a mandatory software update, no physical devices are moving, and that the article is written misleadingly, even if technically correct. Or am I the only person that doesn't understand this new definition of "recall
125.
by nmca 3y ago
Are there tools, processes or organisations that could make this easier? For example, I could imagine: * a blog of success stories * a guide on how to do things like coordinate breaking leases, getting leases aligned, etc * maybe advice on
126.
by nmca 3y ago
systems level - so if, for example, I can have two services or serverless functions, one of which can access the world and the other of which cannot see, for example, the pii tables. (I'm not a production customer, and I haven't t
127.
by nmca 3y ago
(on the off chance the devs read this) I can see pain in the future for the currently excellent KV ergonomics around access control --- is the solution just "implement it in userland and don't write bugs" or is there anything
128.
by nmca 3y ago
I find deno to be very exciting. Viable business model, great ergonomics, glorious lack of bullshit configuration busywork.
129.
by nmca 3y ago
I don't believe this story, despite much hands on experience with LLMs. (including sampling a shit-ton of poems, which was a major source of entertainment)
130.
by nmca 3y ago
This comment is silly because of course you need to turn the pages and at volume paying people to do so is prohibitive. On the software side though, progress marches on: https://facebookresearch.github.io/nougat/ is do
131.
by nmca 3y ago
I've had my last four phones get issues due to usb-c. The standardization is worth it, but goddman I wish they had standardized on a better physical port.
132.
by nmca 3y ago
This is nuanced and interesting, despite most likely being completely wrong - as others have mentioned, it's predicted on LLMs being at the top of the sigmoid and they just ain't. Also, this goal-post-moving did crack me up rather
133.
by nmca 3y ago
Maybe we're talking at cross purposes; what I mean is that PCA+KNN or other tricks mean that for high dim and large amounts of data you can get useful results; with the enlarged view of KNN as "anything where the last step is KNN&
134.
by nmca 3y ago
I dunno I've done a lot of high-dim knn that were quite pragmatically useful
135.
by nmca 3y ago
I wish we had real money prediction markets. Afaict it would be legal to run such a business in the UK.
136.
by nmca 3y ago
house prices evolve from price p0 to price p. The market gains (or loses) p/p0 in that period. If you get a mortgage, you're borrowing money to buy at p0; at any point in time with a price p and total repaid r your total value is
137.
by nmca 3y ago
Worth pointing out, imo, that if you think this problem is not real then you are asserting you understand AI better than Geoffrey Hinton, Yoshua Bengio and Ilya Sutskever. If you don't know who they are, then well I guess that makes se
138.
by nmca 3y ago
This wasn't recent
139.
by nmca 3y ago
I found the following line in a codebase once, which made me laugh: ``` interval = interval*1.6 # fiboroughly ```
140.
by nmca 3y ago
I too worked for Google once.
141.
by nmca 3y ago
I basically don't believe non-degenerate (psued)anonymization is possible, although that complicated af homomorphic encryption stuff makes me a little uncertain.
142.
by nmca 3y ago
This plainly isn't true, just to briefly articulate: * Hypothesis generation sped up * Candidate participant discovery sped up - including finding controls * large classes of analysis possible on just the data, including difference in
143.
by nmca 3y ago
Some folks would be materially harmed by medical record release; e.g. women who have had abortions in states where it is illegal, gays in very conservative areas, passing trans people, the functional severely depressed, passing folks with a
144.
by nmca 3y ago
sure but to engage in the relevant consequentialist reasoning correct you need to argue that this is more people, not just some people (I'm undecided on the whole thing but this argument as stated is bad)
145.
by nmca 3y ago
nobody with lots of experience with proprietary LMs is surprised
146.
by nmca 3y ago
It is.
147.
by nmca 3y ago
This was a great investigation of a trick one might not naively expect to be important & a theoretically justified fix. Nice work!
148.
by nmca 3y ago
caring about grammar is a mug's game, always has been and always will be.
149.
by nmca 3y ago
It's important that humans are still able to understand the actions of these systems for safety reasons; the area focused on the problem you identify is "scalable oversight", see for example: https://arxiv.org&#x2F
150.
by nmca 3y ago
In the not too distant future, GPT-like models will have read all of your emails, every message you've sent, the entire company slack, your calendar, all the company internal docs and all the code. Some more forward-looking orgs will t

More ›