justinl33

6 ms·

31.
by justinl33 2y ago
speculative fiction is the best. Naomi Kritzer is a phenomenal writer
32.
by justinl33 2y ago
two tips that I've found that help take the load off support: - document everything obsessively; 80% of support tickets can be answered by 20% of your documentation - use subreddits and discords and let your users help one another
33.
by justinl33 2y ago
counterargument: toast is unobtrusive by design. it's not part of the main workflow, and not causing the user to context switch. +1 to screen readers
34.
by justinl33 2y ago
The S in LLM stands for safety.
35.
by justinl33 2y ago
What's more interesting is Dublin's high ranking - I wonder how much of that is driven by Ireland's corporate tax structure versus actual technical advantages.
36.
by justinl33 2y ago
"organizing the world's information"
37.
by justinl33 2y ago
"organizing the world's information" or 'maximizing revenue?' I don't know - somehow either argument justifies this
38.
by justinl33 2y ago
as someone who's worked on a project involving weather data for agriculture over developing regions, I can't explain how bad (sparse and inconsistent) the data is. Can I ask how you handled the regulatory aspect of launching from
39.
by justinl33 2y ago
I hear you. Imo it's still a tradeoff/balancing act between business outcomes and your own career
40.
by justinl33 2y ago
This reminds me of an experiment in grad school on spatial memory in rats. We noticed that during sleep their hippocampal place cells would fire in a way similar to those seen during maze navigation. I personally experience this all the tim
41.
by justinl33 2y ago
the biological grok
42.
by justinl33 2y ago
100%. I've seen junior PMs fast track their careers by literally just always framing their work in terms of business outcomes. One of my reports went from entry-level to senior in 2 years by always tying her projects back to user growt
43.
The case against work life balance (Shyam Sankar)(shyamsankar.com)
1 points by justinl33 2y ago|0 comments
44.
by justinl33 2y ago
> start over each day This reminds me of "spaced repetition" in learning theory. Drilling the same problem from scratch is a great way to get better at iterating through your rolodex of mental models, but so many people priorit
45.
by justinl33 2y ago
yeah Lyrebird health... been hearing some crazy as stuff about them - wouldn't be surprised if they were YC soon
46.
by justinl33 3y ago
Technical report here: https://openai.com/research/video-generation-models-as-world...
47.
by justinl33 3y ago
This will probably cost some downvotes, but can we start a thread explaining the architecture behind this for this interested in how it actually works?
48.
by justinl33 3y ago
Best way to optimize a python dict is to switch languages and use a dict there
49.
by justinl33 3y ago
> The point is, there's a truckload of fucked up shit on the net and AI isn't that good yet. I would beg to differ. AI is definitely good enough to identify fucked up things. Feed GPT-4 even a subtly f cked up question, or f
50.
by justinl33 3y ago
Am I the only one wondering how any half capable AI classifier would be unable to identify a literal beheading video or man having s*x with a fish? And especially at a cost that would justify using manual labellers? I get that these human a
51.
by justinl33 3y ago
Isn’t this a bit of an ‘oh no… anyways’ moment for HelloFresh? The delivery cost of the ads (spam) themselves was probably a factor of 10 greater; considering they sent literally 80 million emails
52.
Someone added dot plots to Matplotlib(github.com)
1 points by justinl33 3y ago|0 comments
53.
Show HN: Dotplotlib – hijacking Matplotlib to make better dot plots(github.com)
3 points by justinl33 3y ago|0 comments
54.
by justinl33 3y ago
This doesn't work; the door needs to swing to 140 degrees, once it swings past this range, it would never be able to round the corner to close.
55.
by justinl33 3y ago
Re: the refrigerator door. I’ve always wondered if just engineering a door that shuts itself is a better option here… like a servo that swings the door around enough to shut… what’s the overhead for either unnecessary piece of technology…
56.
by justinl33 3y ago
1. this is interesting. tokens are arbitrary identifiers... of text. It's called a 'text encoder' for a reason 2. it's a strong assertion but it is true. it's kind of inherent in the name; next word prediction . Wh
57.
by justinl33 3y ago
I disagree; 1. the parameters are there to make the model better at learning textual patterns, not arithmetic patterns. 2. Next token prediction is a terrible way to perform arithmetic. 3. Perhaps most importantly, the loss function does no
58.
by justinl33 3y ago
it's sort of a weakness inherent to LLM's... next word prediction isn't really supposed to be good at math. I doubt it will be ever 'cracked' with better LLM's, only multimodal ones with access to program execu
59.
by justinl33 3y ago
tldr; - 'sparse mixture of experts' means each layer contains 8 mini-neural-nets (experts), each trained to be good at certain data/tasks - a router network passes each token (word) to 2 experts which suit it best - since onl
60.
by justinl33 3y ago
First experience with the app was 10/10; other than not being able to type the letter ‘m’ in my first Doc. Little things like these can make the difference between instant churn and lifetime users.

More ›