dylanbfox

8 ms·

1.
How RLHF Preference Model Tuning Works (and How Things May Go Wrong)(assemblyai.com)
95 points by dylanbfox 3y ago|9 comments
2.
How to run Stable Diffusion locally(assemblyai.com)
8 points by dylanbfox 4y ago|2 comments
3.
AssemblyAI (YC S17) Is Hiring Senior Research Engineers
1 points by dylanbfox 4y ago
4.
AssemblyAI (YC S17) is hiring senior engineers to build ML systems at scale
1 points by dylanbfox 4y ago
5.
Differentiable Programming – A Simple Introduction(assemblyai.com)
159 points by dylanbfox 4y ago|49 comments
6.
Is hybrid work the worst of both worlds?(economist.com)
3 points by dylanbfox 5y ago|0 comments
7.
Is light pollution a problem? [video](youtube.com)
1 points by dylanbfox 5y ago|0 comments
8.
by dylanbfox 5y ago
Looks interesting! Do you guys offer any sort of visibility tools/reports into customer usage of different endpoints, tracking of actual API requests (including payloads/etc) per request, etc? Is it possible to use you guys just f
9.
by dylanbfox 5y ago
Agree. The places I go on the web have become more and more centralized/limited. I think projects like this that help to surface and aggregate interesting content from the web (which is really what I come to HN to find) are great.
10.
Space pictures from the Voyager 1 and 2 missions(planetary.org)
1 points by dylanbfox 5y ago|0 comments
11.
by dylanbfox 5y ago
Nice work! The UI is really simple - and love not having to log in to use it. Have you thought about leveraging the ListenNotes API ( https://www.listennotes.com/api/ ) to automatically pull in the podcast episodes via s
12.
by dylanbfox 5y ago
I agree with all your points - but one thing I think about is: how do we fix what we have today? How do you fix the concrete jungles that most cities are today in the US. Or is it inevitable that more concrete will just be poured over time
13.
Text Segmentation – Approaches, Datasets, and Evaluation Metrics(assemblyai.com)
5 points by dylanbfox 5y ago|3 comments
14.
An Overview of Transducer Models for ASR(assemblyai.com)
8 points by dylanbfox 5y ago|0 comments
15.
by dylanbfox 5y ago
Hi there - OP here - thanks for reading! This blog is more of an intro to a few high level concepts (multi-GPU and multi-node training, fp32 vs fp16, buying hardware and dedicated machines vs AWS/GCP, etc) for startups that are early i
16.
by dylanbfox 5y ago
Author here. Thanks for your comments! In general - this is expensive stuff. Training big, accurate models just requires a lot of compute, and there is a "barrier to entry" wrt costs, even if you're able to get those costs do
17.
by dylanbfox 5y ago
Dylan from Assembly here. If you want to send me one of your audio files (my email is in my profile) I'd be happy to send you back the diarized results from our API. You can also signup for a free account and test from the dashboard wi
18.
by dylanbfox 5y ago
Great question. This is technically referred to as "Wake Word Detection". You run a really small model locally that is just processing 500ms (for example) of audio at a time through a light weight CNN or RNN. The idea here is that
19.
by dylanbfox 5y ago
Interesting. How do you guys manage spot interruptions when training on spot instances?
20.
by dylanbfox 5y ago
This is tricky. The de facto metric to evaluate an ASR model is Word Error Rate (WER). But results can vary widely depending on the pre-processing that's done (or not done) to transcription text before calculating a WER. For example if
21.
by dylanbfox 5y ago
> Salary costs are probably even higher than compute costs. Yes exactly. Managing that much compute requires many humans!
22.
by dylanbfox 5y ago
Dylan from Assembly here. Most of our customers have actually switched over to us from Google - this Launch HN from a YC startup that uses our API goes into a bit more detail if you're interested: https://news.ycombinator.co
23.
How to train large deep learning models as a startup(assemblyai.com)
273 points by dylanbfox 5y ago|81 comments
24.
by dylanbfox 5y ago
Interesting. It seems like in the "real world" WER is not really the metric that matters, it's more about "is this ASR system performing well to solve my use case" - which is better measured through task-specific me
25.
by dylanbfox 5y ago
> since less common words tend to be more important for meaning. Exactly. Errors with proper nouns are usually more problematic than errors with stop words, yet they're weighted equally in the WER calculation. Ie, deleting "Bob
26.
by dylanbfox 5y ago
Yes! Perplexity is a great idea. Although you could technically have a low perplexity prediction that is not similar to the ground truth transcription. CER is definitely more granular. There are papers that basically count Deletions, for ex
27.
Is Word Error Rate a Good Metric for Speech Recognition Models?(assemblyai.com)
29 points by dylanbfox 5y ago|30 comments
28.
The Roomba j7 has poop-detecting artificial intelligence onboard(arstechnica.com)
6 points by dylanbfox 5y ago|1 comments
29.
Faster than Light: Warp Drive (2013)(youtube.com)
1 points by dylanbfox 5y ago|0 comments
30.
How to Build a Burner Phone in Python(assemblyai.com)
4 points by dylanbfox 5y ago|0 comments

More ›