kroolik

8 ms·

61.
by kroolik 4y ago
What makes me so negative is that it has a tremendous productivity boost regardless the intentions. ChatGPT saves you hours daily generating content, but it makes it as easy to generate convicing content on any subject, at scale. It may be
62.
by kroolik 4y ago
I'm really negative about the impact AI will have on the society. We have already been drowning in fake news and polarizing information. Now, the likes of DALL-E and DeepFake can generate convincing fake graphics. Chatgpt and the likes
63.
by kroolik 4y ago
My experience has been that the answers are very convincing, but not necessarily true. I would be careful asking gpt questions about abstract knowledge, less about linguistic structure.
64.
by kroolik 4y ago
What arguments have you used to finally convince them?
65.
by kroolik 4y ago
Good point!
66.
by kroolik 4y ago
Isn't your first paragraph all about not knowing what argument types the function accepts just looking at its declaration, as it defeats the purpose of using a dynamic language?
67.
by kroolik 4y ago
I think the example is a bit grey. In my opinion, function should list its dependencies and allow changing them. Having said that I dont believe the `is_ci` decision should happen in the function. The decision should happen at the entrypoin
68.
by kroolik 4y ago
On the other hand, big codebases without type hints rely on their engineers remembering types of every argument. Or the functions being overly defensive. Each to their own liking, I prefer knowing what argument types a function accepts so I
69.
by kroolik 4y ago
This is purely subjective. First of all, I prefer putting type hints everywhere. PyCharm helps reminding one in a lot of cases. If there is a bug and somebody passes invalid type to my function, I would just fix it and move on. Very often b
70.
by kroolik 4y ago
It's really annoying. Esp. For a person that uses golang occasionally, like me. Thankfully, Goland learned how to autocomplete these inside format strings.
71.
by kroolik 4y ago
Doesn't pip's hash checking mode solve this issue? Freeze your requirements with hashes. Pypi already provides hashes for sdists and wheels. See https://pip.pypa.io/en/stable/topics/secure-installs&#
72.
by kroolik 4y ago
Naming is hard ;) But thanks for the nitpick
73.
by kroolik 4y ago
You can pronounce it with the cardinals 1st Cache invalidation 3rd Concurrency 2nd Naming things 4th Off-by-one errors
74.
by kroolik 4y ago
Any reason the article has sections appearing out of no where, that don't fit the surrounding parts? For example: > large, heterogenous datasets. When all the rows look similar there will be fewer candidate plans and less chance of
75.
by kroolik 4y ago
What about just throwing every thing into docker?
76.
by kroolik 4y ago
Indeed it might not be enough, and there is no one size that fits them all ;) For inference times of 50ms I wouldn't bother complicating the setup that much. Again, it all depends on your requirement and trade offs available.
77.
by kroolik 4y ago
Depending on how much time it takes to query your model. If it takes significant portion of your time budget, you could just put a reverse proxy in front of your models and make it route the traffic. I would architect your workers around ho
78.
by kroolik 4y ago
This is a very good question, with a lot of different answers depending on your use-case. One approach is to translate the synchronous call into an async call plus polling on the webapp side. You push onto a queue, with the callback queue i
79.
by kroolik 4y ago
Would you describe getting response from the model within few milliseconds cpu or memory intensive? What I assumed as to the characteristics of inference is a process that takes multiple seconds to minutes. In a system that requires a respo
80.
by kroolik 4y ago
Exactly this. Prime example of the difference is whether you accept disruption of inference when you deploy a new version of your webapp. Very high chances you don't, thus you will start implementing a queueing mechanism without realiz
81.
by kroolik 4y ago
I have a feeling the root cause is not in FastAPI or Flask, but in the architecture of the system itself. Why? You are doing the inference in the same request, which is synchronous from the perspective of the caller. The request can be memo
82.
by kroolik 5y ago
Just make sure your BCP plan includes other provides. HN is full of stories where peoples' accounts are blocked with no reason and without means of effective contact.
83.
by kroolik 5y ago
> I'm guessing infra-costs are pretty low - the site hasn't changed much in 15 years Not necessarily ;) IMHO what matters is their actual traffic. For each user they are going to send an email in the future. If they have 10 dai
84.
by kroolik 5y ago
Then it might, indeed, be unrelated. Thanks for clarification!
85.
by kroolik 5y ago
Still, that engineer (if its engineer's fault) is extremely unlikely to make that issue again. IMHO, the problem is systemic in that why the system allows such errors (if its human error) to happen. Given Google's scale I think a
86.
by kroolik 5y ago
Why would you fire an engineer you have just spent millions to train?
87.
by kroolik 5y ago
Nothing wrong with it. People make errors, distributed systems aren't easy. More frequent changes - more likely a bug was introduced. My post is just a speculation, let's wait for the actual technical details doc from Google.
88.
by kroolik 5y ago
There were outages around the same time last year. Somebody in the HN thread commented back then that the employees evaluation and promotion window ends around december/eoy, thus more releases are made. https://en.m.wikipedi
89.
by kroolik 5y ago
> Some VP will ask "what weekday do users comment the most". This is why people invented read models. Because the data model that works for the app and makes it easy to work with there might not necessarily work for analytics.
90.
by kroolik 5y ago
What you did, in fact, was creating read models specific to use case. This is completely normal and allows one to optimize for different users: app's schema can be different than the one for analytics.

More ›