tehsauce

9 ms·

61.
by tehsauce 3y ago
I’ll save yall some time. If you are familiar with concatenating strings in python, langchain is a library which does just that, only a little less cleanly. Hope this helps!
62.
by tehsauce 3y ago
yes but it’s very easy to install if you have npm
63.
by tehsauce 3y ago
Just fyi, if you use a standard knn library like faiss and pretty much any embedding + language model raw from huggingface or an API, it will require ~15 lines of code to do what you describe. I’m not sure how much shorter langchain made yo
64.
by tehsauce 3y ago
Very excited to see how far minecraft servers can be pushed! If anyone is curious why you’d want to have 1000 players in minecraft, this video documents a pretty amazing example scenario: https://youtu.be/zv-TS_mEHE4
65.
by tehsauce 3y ago
Hi! I really appreciate folks like you conducting and publishing real research. There have been a ton of companies recently which have been very rosily promoting their new models. My criticism was only to push back on overly optimistic mark
66.
by tehsauce 3y ago
Their main claim of “faster” unfortunately is false. > Running on an Nvidia A100 GPU, Paella took 0.5 seconds to produce a 256x256-pixel image in eight steps, while Stable Diffusion took 3.2 seconds Using the latest methods (torch 2.0 co
67.
by tehsauce 3y ago
Usually labeling text/images, answering simple questions, ranking or scoring.
68.
by tehsauce 3y ago
Overall a very interesting result. I’m not sure why they compare to a list of results for the MBPP benchmark but don’t seem to include these results which are much better: https://paperswithcode.com/sota/code-generation
69.
by tehsauce 3y ago
Correct except they did this for gpt2 not gpt3
70.
by tehsauce 3y ago
Nope! That surely doesn't apply to humans, why would it necessarily apply to learning machines?
71.
by tehsauce 3y ago
No one has trained a LLM of the open source level quality with just 3 gpus. Fine tuning sure, but pretraining the even the smaller models takes more than that.
72.
by tehsauce 3y ago
Seems in similar spirit to the “perceiver” architecture from deepmind a couple years ago: https://arxiv.org/abs/2107.14795
73.
by tehsauce 3y ago
Why is it the worst way though?
74.
by tehsauce 3y ago
Earlier discussion: https://news.ycombinator.com/item?id=36085936
75.
by tehsauce 3y ago
This is very cool despite the most important caveat: “Note that we do not directly compare with prior methods that take Minecraft screen pixels as input and output low-level controls [54–56]. It would not be an apple-to-apple comparison, be
76.
by tehsauce 3y ago
When I finally got access to mojo, I was very unimpressed to say the least. For how much marketing and attention they’ve gotten already and the size of their team, the product is in very early stages. It’s not clear how their product is dif
77.
by tehsauce 3y ago
This is a great in depth and sober analysis.
78.
by tehsauce 3y ago
If you want to do this today you can also use the torch c++ api! It’s whats pytorch binds to under the hood.
79.
by tehsauce 3y ago
On vast.ai I rent 3090s for $0.12 an hour. Nothing comes close in price.
80.
by tehsauce 3y ago
What reason is there to learn a new query language when I can program a LLM with any existing language?
81.
Modern language models refute Chomsky’s approach to language [pdf](lingbuzz.net)
5 points by tehsauce 3y ago|1 comments
82.
by tehsauce 3y ago
They're paid with the subscription money based on how many members watch! If you only watch a small number of videos or don't mind ads then it might not be worth 4x for you. However I personally find advertising deeply disturbing
83.
by tehsauce 3y ago
Yes, I would happily pay more for the ability to remove shorts.
84.
by tehsauce 3y ago
+1 on this. YT premium is easily worth 4x its price, it's an amazing deal. It's the largest media catalog in the world, and unlimited ad-free access for $12 is a steal. There is no digital subscription that comes close in value.
85.
by tehsauce 3y ago
Your computer can’t really do any useful crypto mining, so I wouldn’t be too worried.
86.
by tehsauce 3y ago
Vast.ai Nobody has better prices.
87.
by tehsauce 3y ago
Should have the label 2022
88.
by tehsauce 3y ago
“Complex numbers” are rather poorly named. They are more naturally understood as simply a vector which has a magnitude, can be rotated and scaled. As geometric objects they are much more intuitive. The subject geometry algebra takes a grea
89.
by tehsauce 3y ago
In the example from the article, copilot produces identical comments, not just a functionally identical implementation. So in this case your hypothesis is false. But thanks for trying to stand up against the open source community for micros
90.
by tehsauce 3y ago
Non-determinism is due to the implementation rather than the fundamental method. In principle a language model can be executed deterministically with any temperature you want.

More ›