HNHacker News
TopNewBestAskShowJobs

stuartaxelowen

701 karma · joined March 26, 2015

stuart at sao dot dev
submissionscomments
stuartaxelowen··on MLOps is mostly data engineering
Feature stores are essentially materialized views (aside from any realtime feature resolution needed). I think it's a good thing that there is specialized effort being taken here, though: features stores are an abstraction that could be useful in other domains also, and this surge in interest is an opportunity for us to make better tools.
stuartaxelowen··on Pause Giant AI Experiments: An Open Letter
93 points in 4 hours... Why did this go off of the front page?
stuartaxelowen··on 5M item limit for Google Drive: File unable to generate or upload due to 403
https://www.google.com/drive/ 's title does call it a "File Sharing Platform".
stuartaxelowen··on Experimental library for scraping websites using OpenAI's GPT API
In my experience, the hard part is not extracting data from websites, but observing and implementing the actual structure of the site - e.g. iTunes categories have apps, which have reviews, etc, and making your scraper intelligent enough to make use of that structure to gather the freshest data efficiently.

There is definitely a place for LLMs in solving this problem: in taking over for the human in interpreting the business goals/data to gather along with the available data on the web, but my experiments have shown that this is a significant problem due to limited LLM context length and difficulty distilling messy data. But, very excited to keep pushing, and seeing where things go :)

Note: I build https://www.thoughtvector.io/pointscrape/ to solve very-large-scale web-data gathering problems like these.

stuartaxelowen··on Lowering a dumbbell 6 times a day offers significant gains in strength
You don’t have to do the strict movement to raise the dumbbell, and can do less work by employing other muscles. For instance, using your legs and back to give the weight momentum to make it easier to raise it to the target position.
stuartaxelowen··on How do dolphins choose their name?
Incredible (but maybe unsurprising) that nature is maximizing information density and designing efficient codes (a’la information theory) depending on their noise environment!
stuartaxelowen··on Launch HN: Nyckel (YC W22) – Train and deploy ML classifiers in minutes
Check out Spacy, it provides PoS tagging (among other things): https://spacy.io
stuartaxelowen··on Launch HN: Humanloop (YC S20) – A platform to annotate, train and deploy NLP
Do you allow for on-premise inference?
stuartaxelowen··on Transcribed police scanners in real-time
Thanks!
stuartaxelowen··on Transcribed police scanners in real-time
Very cool! Is this yours, OP? I built a text classifier platform (https://www.taggit.io/) and I'd love to explore building something on top of this text stream to notify people of important happenings around the protests in real time.
stuartaxelowen··on Ask HN: What's your quarantine side project?
I've always been fascinated at looking at public feedback, for products and companies. I've been working on a dashboard that lets you explore recent iOS app reviews:

https://www.thoughtvector.io/vertext/

There's all kinds of fun stuff you can learn, like apparently Instagram just ratcheted up the number of ads shown and people don't like it. You can see different text topics people mention, and select time ranges in the volume and rating over time charts to filter.

You can look at other iOS apps too, just by adding the app ID in the search params. This page shows DoorDash's reviews:

https://www.thoughtvector.io/vertext/?app_id=719972451

It's still very much beta-level in terms of all the things I want in there, but it's been a fun distraction so far!

stuartaxelowen··on Shirt Without Stripes
This feels like a perfect example of how contextual natural language is!
stuartaxelowen··on People kept working, became healthier while on basic income: report
Is care of a sick family member "produced" in the same sense?
stuartaxelowen··on Python overtakes JavaScript as most questioned language on StackOverflow
It's much better to just use a table of phrases/keywords, even if it takes a little more effort. Phrases are specific and have clear intent - using only the most common words leaves out critical context!
stuartaxelowen··on Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts
To fine tune the model to the specific customer needs. For example, learning the parameters of they user's face.
stuartaxelowen··on Technical Debt Is Like Tetris
Check out Mari Kondo's show on Netflix! It is a guide on managing organizational debt in your home!
stuartaxelowen··on Disinformation and ‘fake news’: Final Report published
Why?
stuartaxelowen··on Star topics that interest you
There are also topics like "NLP".
stuartaxelowen··on Automatic categorization of text is a core tool now
What changed was pretrained word embeddings like fasttext and GloVe. These embeddings drastically reduce the complexity of text classification problems, and reduce the amount of data you need to produce a high quality text classifier a great deal, and in that 2 year span, understanding of how to leverage them spread far enough to reach applications like this.

The world has been sleeping on leveraging NLP, given the incredible amount corporations spend paying people to read and summarize text (answering questions like "How many people are having a bad search experience?", etc). The recent leaps in NLP make for many opportunities, and many products are coming to market that allow even non-technical folks to train and utilize text classifiers. If you're thinking about getting into NLP, now is definitely the time to do it.

Source: founded taggit.io

stuartaxelowen··on Bye Bye Mongo, Hello Postgres
RE 5: I'd argue that with a relational database (barring extreme scale cases) is the perfect place for your business logic. Constraints, unique indexes, etc, all ensure that your data is always consistent with itself and with the business logic you've encoded.
stuartaxelowen··on The Problem with Feedback
I think the author has it wrong, what the author sees as a "barrage of angry or ecstatic reviews" is actually a rich array of personal experiences waiting to be interpreted.

Traditionally, this interpretation has been done wholly by humans, and was a long and laborious task. However, we can now do much of the manual work with NLP [1] - discovering topics of concern and measuring the prevalence and sentiment. This lets us ask the dataset useful questions, like "What issue is angering customers most?", and "What features should we be talking more about?"

It's a move towards a much, much wider sampling of customer experiences (compared to focus groups), and I think it should be celebrated for empowering companies to make products/services that consumers really, really like.

[1] https://taggit.io

stuartaxelowen··on The Problem with Feedback
Not so - the business is looking for real ROI, and thus will not be satisfied with topics that are "in the noise". Also, you create the measurement that establishes a problem before you apply the treatment (create new feature X or fix bug Y), and then verify after treatment that the desired affect was achieved.
stuartaxelowen··on Show HN: All-pair similarity search on millions of sets in Python and on laptop
Data science is about drawing conclusions from data, not the size of it. Most data science happens on datasets numbered in the hundreds rather than those in the millions.
stuartaxelowen··on Python Data Visualization 2018: Why So Many Libraries?
I love vega-lite, thank you so much for your hard work!
stuartaxelowen··on End-to-end implementation of a machine learning pipeline (2017)
And if you need the index:

    for idx, genre in enumerate (list_of_genres)
stuartaxelowen··on The highest paid workers in Silicon Valley are product managers (2016)
Most companies still have a very hard time building a good machine learning pipeline.
stuartaxelowen··on Challenge – AI Initiative
First impression: your site claims that $15/hr is a living wage in a major city, when in Seattle you'll be spending at least half that on an apartment anywhere in the metro area.
stuartaxelowen··on Tao Tajima – WebGL and GreenSock Animation
This is breathtaking. This is the first good usage of scroll manipulation I've ever seen, and it has blown my socks off.
stuartaxelowen··on Kitty – a fast, featureful, GPU based terminal emulator
Why use the GPU? Is terminal redraw latency really this big of a problem?
stuartaxelowen··on Opera launches as a snap for Linux users
My account literally has my old legal name, and can't be changed
← PreviousPage 2 of 7Next →