HN user

bagavi

22 karma
Posts0
Comments32
View on HN
No posts found.

Whisper is not a mature enough to be criticized at this level. They are trying to solve a very hard problem, give them constructive feedback.

IPhone didn't support 100s of languages in the first few years.

The author is either entitled or resentful; neither helps.

Why judge it by it's worst qualities and worst case behavior? This attitude is uninspiring and worse the luddistism.

Every new technology initially has its own shortcut comings.

Steve jobs in his inaugural iPhone launch, used four different phones because each phone was capable of only one feature.

Waymo and Cruise have burnt through money like no one else. They want to make their own cars, thier own Ai chips/lidar, their own self driving tech. Delusional at best, malicious at worst. They made a hard SW problem into hard HW+SW problem. They are zombies unless some magic happens in the world of nn (which is not unlikely tbh)

Whereas comma is taking the right approach of a nimble team, iterate fast and ship a working product (even if not L4-5), get cash flow, next milestone.

Simple explaination of the general idea using nearest neighbor search as example:

1. You want to calculate the l2-nearest neighbor to y from a set of x1, x2, ... xn. Where each point is a d-dimensional vector.

2. Naive approach will take O(nd) computations. For very large d, say 1e4, this is very expensive

Multi-arm bandits idea:

3. Estimate the distance of y to each point (xi) by sampling only along log(d) dimensions.

4. Keep the nearest n/2 points (throw away the rest)

5. Repeat step 3 and 4 until you have one point left

6. The total run time = nlog(d) + nlog(d)/2 + nlog(d)/4.... = O (nlog(d)log(n))

This is a huge improvement over O(nd) for large d (which is usually the case today)

This paper (https://ar5iv.org/abs/1805.08321) from Stanford pioneered this idea for nearest neighbors and many such household problems. It also theoretically proves that the above approach gives the right answer and empirically gets huge (100x) speed boost on large datasets.

The k-medoid paper is based off of this work :)

The banks don't want you to split the money because they will have to pay higher insurance to FDIC. This will eat into their margins. The business and the customers interests are not aligned in this case.

Until the law changes, you have to look out for yourself and not beg for bailout. Also, this is not complex since software takes care of it behind the scenes. Our company did it from day 1.

Also more than half of deposits by volume are FDIC insured so the system can handle the whole volume if the customers choose to do so

This is the real deal breaker for me. Dvc is super slow but it works with S3 (one of the greatest technologies built in last 15 years). At our company, we've written own (10x) faster version of dvc for commonly used features.

The alternative tool you are referring to is `Dud` I believe

Dvc is the best tool (I found) inspite of being dead slow and complex (trying to do many things).

What alternatives would you recommend?