johnsutor

6 ms·

31.
Scaling Exponents Across Parameterizations and Optimizers(arxiv.org)
2 points by johnsutor 2y ago|0 comments
32.
by johnsutor 2y ago
I know there are murmurs that synthetic data (i.e. using rendering software with 3D models) was used to train some generative models, including OpenAI Sora; seems like it's the only plausible way right now to get the insane amounts of
33.
by johnsutor 2y ago
BERT isn't dead for smaller tasks (think NER, Sentiment Analysis) where low latency is needed.
34.
Dola Decoding by Contrasting Layers Improves Factuality in Large Language Models(arxiv.org)
58 points by johnsutor 2y ago|43 comments
35.
Conventional Commits(conventionalcommits.org)
1 points by johnsutor 2y ago|0 comments
36.
by johnsutor 2y ago
How does this differ from Ragas? https://docs.ragas.io/en/latest/index.html
37.
Predicting Neural Network Accuracy from Weights (2021)(arxiv.org)
1 points by johnsutor 2y ago|0 comments
38.
CodeSignal's 2024 University Ranking Report(codesignal.com)
1 points by johnsutor 2y ago|0 comments
39.
by johnsutor 2y ago
https://archive.is/UnVyG
40.
by johnsutor 2y ago
Last weekend, had a listening event at the Angel Orensanz Foundation, where they took the original photos for the 36th Chambers album release. Was a cool experience.
41.
Flow Computing raises $4.3M to improve CPU performance by 100X(venturebeat.com)
2 points by johnsutor 2y ago|0 comments
42.
Study finds 268% higher failure rates for Agile software projects(theregister.com)
221 points by johnsutor 2y ago|190 comments
43.
Grokfast: Accelerated Grokking by Amplifying Slow Gradients(arxiv.org)
117 points by johnsutor 2y ago|39 comments
44.
PyTorch Hyperparameter Optimization on TPUs(dsuess.github.io)
1 points by johnsutor 2y ago|0 comments
45.
Show HN: Recreate Images with Emojis(replicate.com)
1 points by johnsutor 2y ago|0 comments
46.
by johnsutor 2y ago
Long industrial technology, Short pre-industrial technology
47.
by johnsutor 2y ago
They're only trained up to a certain point in time, so adding RAG should hypothetically allow such LLMs to access the most up-to-date information.
48.
NYC takes first step towards unleashing robotaxis on city roads(popsci.com)
1 points by johnsutor 3y ago|0 comments
49.
by johnsutor 3y ago
I've got a dust allergy and swear by the Levoit Core air purifier. It's crazy how much dust accumulates in such a short period of time (I live in a building with central air that doesn't have filters on the vents). I also try
50.
by johnsutor 3y ago
Main thing I would be concerned about is safety (I’m an avid Volvo fan) but it appears to be (relatively) safe as well: https://microlino-car.com/en/service/faq/how-safe-is-the-mic...
51.
The Internet Speculative Fiction Database(isfdb.org)
4 points by johnsutor 3y ago|0 comments
52.
by johnsutor 3y ago
Was more into their YouTube / Video content to be honest
53.
by johnsutor 3y ago
This is the premise behind some more recent self-supervised learning papers (see https://arxiv.org/abs/2303.03307 ). Turns out, it's a solid analog for classification tasks.
54.
by johnsutor 3y ago
Or maybe it's becoming sentient and wants to make us think it's spewing random words as a decoy /s
55.
by johnsutor 3y ago
I wonder how LlamaParse compares head to head with https://unstructured.io
56.
by johnsutor 3y ago
See the topic of GLMs: https://en.wikipedia.org/wiki/Generalized_linear_model
57.
by johnsutor 3y ago
Honestly, looks more useful. Tensorflow embedding projector is pretty limited except for quick nifty visualizations, but it doesn't really inform you much about different clusters of points or why different clusters or hierarchies emer
58.
by johnsutor 3y ago
I think it's definitely coming, with more recent advances in unsupervised learning and multimodal learning (along the lines of https://github.com/microsoft/unilm or https://github.com/Alpha-VLLM&#x
59.
by johnsutor 3y ago
I'm hopeful that this will get better as time goes on with advances in unsupervised and multimodal learning, that will enable us mere pions to achieve our niche use cases by fine-tuning a larger model with minimal compute.
60.
DataDreamer: Prompt. Generate Synthetic Data. Train & Align Models.(github.com)
14 points by johnsutor 3y ago|0 comments

More ›