jeffshek

8 ms·

31.
by jeffshek 7y ago
A lot of work from researchers are using *conda on Windows. ML research is happening across all fields, not just computer science.
32.
by jeffshek 7y ago
I know this isn’t tech related, but Kobe was an idol growing up. I never played basketball, but looked up to his work ethic in so many different areas. His willingness to push against determination, setbacks and overcome obstacles largely s
33.
by jeffshek 7y ago
Adoption speed is probably the only proxy that could be used. Measuring quality of pet life in suitable homes is rarely collected. I'm not even certain of how that could be objectively measured for children, let alone pets.
34.
by jeffshek 7y ago
Not if the underlying model was bad, no tweaking of hyperparameters can change that. It's safe to say, he probably did consider this (and it probably didn't work well enough).
35.
by jeffshek 7y ago
https://firstround.com/review/the-founders-guide-to-discipli... A really good podcast about how her strategy on efficient fundraising.
36.
by jeffshek 7y ago
More likely than not, its easier for a liar to get by in densely populated areas. Get caught in a lie, or marked as a liar in a small town can hurt social standing. However, I wonder if this changes now with social media - for better or wor
37.
by jeffshek 7y ago
Alpha is also a reference to the "learning rate" in machine learning models.
38.
by jeffshek 7y ago
If you can prevent being cynical and take that effort into doing actual work, your skillsets quickly compound. If the job sucks, take your improved skillset elsewhere. But do not sit idly and complain - it does nothing, hurts your career, y
39.
by jeffshek 7y ago
I don’t have enough experience with no-code tools either, but I don’t agree it’s motivation that prevents it from taking off. No-code is used mostly by non-technical founders who do seem quite motivated (enough to do it themselves). Enginee
40.
by jeffshek 7y ago
But fool you ten times, and you get Dropbox.
41.
by jeffshek 7y ago
It would be a bad thing to have only one clear winner - you now have to deal with antitrust laws. It’s in ones best interest to look like they have a close competitor, but in reality so distant it won’t catch up.
42.
by jeffshek 7y ago
Sense of social status, cheating artificially increases your rank. No one knows your cheating, so you get all the respect. Leaderboards are dopamine clickbait for trying to improve. Why do billionaires want to richer?
43.
by jeffshek 7y ago
Not quite, the term IC also comes with connotation you don’t want to be a manager track. Many software engineers (including myself) much prefer the IC route.
44.
by jeffshek 7y ago
It’s hard to make rational long term decisions if the CEO is trying to boost short-term profits for public markets.
45.
by jeffshek 7y ago
This captures my sentiment. It’s hard to see how the incumbents could be beaten — precisely because how effective they are around data, buying potential competitors (Instagram, YouTube) .... but this is precisely because we don’t know what&
46.
by jeffshek 7y ago
It would not be close to similarly efficient. They have completely different loss functions.
47.
by jeffshek 7y ago
Blog Post (Last Year) : https://blog.samaltman.com/how-to-be-successful
48.
by jeffshek 7y ago
I’m biased toward my use case - but if I have to move a bunch of terabytes for a ML data pipeline - I arguably am having stuff on GCP. To your point though, yes there is business risk of having it on one cloud.
49.
by jeffshek 7y ago
This is the light that shines through cynicism - the light that makes people brave enough to post new stuff.
50.
by jeffshek 7y ago
You need to watch three of them. He has a style and once it “clicks”, it’s rather interesting. I had a similar impression months ago, and after listening to a few - I was so happy I gave it a fair chance.
51.
by jeffshek 7y ago
I may be wrong, but I recall reading Azure grew quickly because they already had MS contracts in place with many enterprise companies. They simply said, we’ll also hold your hand in the entire cloud process. This was critical because by the
52.
by jeffshek 7y ago
GCP has a really strong ecosystem around AI. If you end up dealing with some type of ML problem with absurd memory requirements, TPUs can be useful there. A lot of recent Transformer architectures have these requirements. Also, a huge amoun
53.
by jeffshek 7y ago
One caveat though is global macro is notoriously difficult. Even when you think all signs point to a conclusion - think “peak oil hypothesis” something can come out of the moon that changes all your hypothesis, ruining your funds return eas
54.
by jeffshek 7y ago
I agree with so many points here. Generative AI is one of those things that are immediately applicable. Instead of problems that require 100% accuracy, generative models don’t require that. As long as the avatar looks close enough, it works
55.
by jeffshek 7y ago
I love PyTorch, but I’m not confident the claim that it is the most popular is close to true. The cited link, which brings up a lot of new research is in PyTorch simply doesn’t account for the amount of TensorFlow in production. Sure, a lot
56.
by jeffshek 7y ago
Another vote for Code Complete. But coming from someone who's bought and gifted a few books, I'm thinking instead of the quote ... "If you want to build a ship, don't drum up people to collect wood and don't assign
57.
A Guide to Keras Functional API(perfectlyrandom.org)
1 points by jeffshek 7y ago|0 comments
58.
by jeffshek 7y ago
https://www.indiehackers.com/ is a great resource around this. The Mom Test (Book) is a way to verify the thing you're building people will actually want. Providing market-fit can be hard. Startup School by YC is anoth
59.
Why Mike Posner Walked Across America(outsideonline.com)
1 points by jeffshek 7y ago|0 comments
60.
by jeffshek 7y ago
That's a little unfair no? Supporting TF's API is much easier than building TF for Google (and then releasing it to the public).

More ›