HackerTrans
TopNewTrendsCommentsPastAskShowJobs

zodzedzi

no profile record

Submissions

Ask HN: What's a good Linux OS and setup to build a dev “network” on my laptop?

15 points·by zodzedzi·3 anni fa·14 comments

Ask HN: How do GPTs grok high-level concepts, beyond word-level transformers?

2 points·by zodzedzi·3 anni fa·10 comments

comments

zodzedzi
·3 anni fa·discuss
For anyone else who was also seeing just garbled characters in the title, and since the first few paragraphs of the linked article don't explain what that is:

GF2P8AFFINEQB is one of the AVX-512 CPU instructions that performs an affine transformation (essentially AX+b) on a Galois field - finite number fields used in coding theory and cryptography.

https://en.wikipedia.org/wiki/AVX-512#GFNI
zodzedzi
·3 anni fa·discuss
Off-topic question: I've been seeing this "si" URL parameter pop up a lot in shared youtube links lately. Which app exactly pulled the URL for you?
zodzedzi
·3 anni fa·discuss
> Bilinguals and monolinguals remembered English competitor words that overlapped phonologically with a spoken English target better than control objects without name overlap. High Spanish proficiency also enhanced memory for Spanish competitors that overlapped across languages. We conclude that linguistic diversity partly accounts for differences in higher cognitive functions...

This conclusion sounds like quite the leap.

Even if the two observations they generated turned out to be 100% ironclad true, generalizing to "speakers of different languages" as a title and "linguistic diversity" from observing just two languages seems like a big jump.
zodzedzi
·3 anni fa·discuss
> ... we had to build our own physics algorithm for very specific problems, and ended up selling the software ...

Like a simulation algorithm? Can you elaborate on what kind of algo and the problem it solved?
zodzedzi
·3 anni fa·discuss
> ... heapsort, which is kind of slow, but prevents adversaries from smashing your stack.

How does heapsort protect against a stack attack?
zodzedzi
·3 anni fa·discuss
Unrelated, but this style he draws the timelines in: is it hand-written/hand-drawn, or software-generated?

It certainly has the overall feel and appeal of being done by hand, but I'm not sure.

If it's software, does anyone know which software, what's the name of this style, etc?
zodzedzi
·3 anni fa·discuss
Uber gets to introduce driverless cars finally (an old promise of theirs) without the costs of owning or developing hardware, and use that as leverage over drivers.

I don't think Uber has anything to fear from Waymo:

(1) It will take years for Waymo to ramp up to "independence" scale. How many cars do they have now, and how many would they need? How long will it take them to negotiate new regulations with every city and state? 5 years?

(2) This deal is probably not exclusive. Uber can strike a similar deal with Cruise as well. Uber becomes the Amazon of driving services, a platform gating access, with all the data.

(3) Having a big money company behind them is good. And if Waymo acquires Uber in 3 years, it's not necessarily a bad thing for Uber.
zodzedzi
·3 anni fa·discuss
Many comments saying Waymo could easily spin up an Uber-replacement app are missing one key point: integrating with Uber allows Waymo to have a slow rollout / soft-launch.

They can start adding support one city at a time. And for the Uber user, the Waymo option only pops up for you if the ride you requested is within the Waymo range and they have cars available.

This way they can also collect tons of data about how users respond to the offers, affinity to driverless cars per region, price elasticity, etc. And then dial the supply up or down as they wish. They can even start covering a city with just two cars if they wanted to, and then build popularity and word of mouth.

On the other hand if they started with their own app, the lack of car coverage in most areas (due to low car supply, pending regulations, etc) would quickly frustrate users who would then switch to another app, so user retention would be a nightmare.

Not to mention side-stepping all the customer-facing operations of running such a business, which Alphabet does not have an affinity for.
zodzedzi
·3 anni fa·discuss
> Generally, the models are compressing their understanding of all text, and in doing so, they're learning high order concepts

Are these higher order concepts accessible to us? E.g. can we list those learned concepts?

(Re-reading the paper you linked now...)