Hacker News Reader

samcodeson Show HN: Visualizing Your Life in Weeks

I don’t think everyone agrees that this is a stick. There is space between the observation (my life is finite) and the emotional response (I have anxiety about my life being finite). I think the idea of these tools is to help people identify that space, and be able to experience the observation without the emotional response.

samcodeson ALBERT: A Lite BERT for Self-Supervised Learning of Language Representations

> I guess one solution would be to use a pretrained ALBERT and finetune to get the initial model and then use model distillation to get a smaller, faster model.

IME this is the way to go; take an ensemble of big, accurate models, then distill them down to the smallest model you can get away with. There are really good tricks in this paper

https://www.aclweb.org/anthology/D19-5632/

samcodeson Mathe Carlo

I wouldn’t call myself an “experienced MC tinkerer” but I liked this series on MC using Python https://towardsdatascience.com/monte-carlo-simulations-with-...

samcodeson Dive into Deep Learning

MXNet existed before AWS picked it, and it has a lot of strengths. I’d use it (especially with Gluon) over TF any day. But that said, PyTorch is usually easy to use on AWS... the preference for MXNet seems weak

samcodeson Dive into Deep Learning

Yeah IME expectations with ML are just the worst. Somehow, non-ML-educated stakeholders expect it to be predictable, like they pretend traditional software engineering is... but also to be magical in scope.

samcodeson DJ Steve Aoki quietly opens a cloud only pizza restaurant in Oakland

Calling to criminalize anything you even mildly oppose is reckless at best. You think if you noticed that two pizza places shared a kitchen, you’d feel the correct response was calling the cops?

samcodeson Machine Learning Reproducibility Checklist [pdf]

Eliezer Yudkowsky wrote a lot about (many things, including) refining the scientific method in Rationality [0]. IIRC, his point is that the scientific method puts no constraints on what hypotheses you test. He thinks a more orderly search of hypothesis space is preferable.

[0] https://www.readthesequences.com/

samcodeson A Mathematical Model Rescued My Book About Math

Get into ML, it's just linear algebra, taking gradients, and arithmetic!

samcodeson Twine is an open-source tool for telling interactive, nonlinear stories

You either hand some of this over to a ML model to handle, and lose some control / deterministic-ness, or you write a ton of code to handle the combinatorial explosion of potential parses of the sentence.

samcodeson BMW shares AI tools used in production

This is impressive only because they are in automotive. I worked in that industry for years, they are generally at least a decade behind the rest of the software world

samcodeson Show HN: Coerce – Actor runtime for Rust using async/await and tokio channels

Not sure on rust compatibility, but have you considered Cap’n Proto? https://capnproto.org/

samcodeson Having Kids

I have a 4 year old and a two month old. We’re in a major metropolis, and most couples seem to choose between 0 and 1 kid, but my wife and I have both always said we wanted 3 or more. This was a nice perspective to hear. For the first 2 years of #1’s life, I couldn’t imagine having a second. But the feeling of “that was it, no more kids” only lasted like a month with #2.

One thing I’m noticing - I expected the joy to grow linearly with number of kids, but I think it actually grows linearly in number of relationships in the family. The interactions between them are so amazing to watch. I’ve always been really terrified to die, and now I definitely don’t want to, because I want to be there for my kids... but if I know that they will have each other’s back, I’m much more okay with not existing one day.

samcodeson Learning at work is work, and we must make space for it

“using JIRA and tracking sprint efficiency wrong” is the working definition of agile at most companies AFAICT.

This makes me sad. I had a great experience with what I now think of as “authentic agile” at my second job, now I doubt I’ll ever get to do things that way ever again.

samcodeson Docker-slim: Minify your Docker container image without changing anything

There is a stage during the analysis where you interact with the container. So you can do something to trigger the resource usage that you want it to pick up on.

samcodeson Amazon SageMaker Autopilot

Since this produces a notebook with python code, you might be able to tweak it so the final model works in tensor flow.js. But depending on model size / hardware requirements, it might be better to just make a network call.

samcodeson Ask HN: Which charities do you donate to?

https://www.givewell.org/

In case utilitarianism is the correct moral framework.

https://www.splcenter.org/

An attempt to slow the US' descent into fascism.

https://www.actionagainsthunger.org/

It really fucks with my head that people are starving.

samcodeson Ask HN: How do I understand the results of my job search?

Yeah, for me and anyone I know who tracks it, out conversion rate (without referrals, which convert much better) is 100 applications to 1 offer.

samcodeson Ask HN: How do I understand the results of my job search?

Hard disagree. Open source contributions are definitely something EM look at. And a good EM knows to look at blog posts. “Can this person explain technical topics to a technical audience” is a hugely important factor.

samcodeson Simple-twitter: A bare-bones Twitter clone implemented in a single file

Yeah, another comment said this is “unreadable” if you don’t do FP... I have done a little lisp + Elixir, but 99% of my coding has been in C-descendant languages and this was definitely readable.

samcodeson Show HN: EasyDB – A One-Click Ephemeral Database

Hard disagree. There’s a button, I pushed it, I read the code... and said “ohhhh that’s clever”

samcodeson YC, not Lambda School, is unbundling education

I took this as a completely serious and reasonable proposal. For context, I have taught (middle school through University), am on the board of an education non-profit, have an advanced degree and plan on accumulating more of them.

What is unreasonable about what the author says? You seem put-off by the way they break down college into 3 principal components: signalling, training, and fun... But other than an emotional reaction to that framing, can you say why it's wrong? I can think of one piece left out, but I will explain why it's reasonable to leave it out.

As part of marketing 4 year liberal arts degrees, schools play up a fourth component, which was more of a factor in the past: non-economic human-development (i.e. learning certain topics enriches your life and adds meaning). I do think this is a real component, but I think, in an effort to maximize profit, universities have completely thrown this away in practice (while making it a cornerstone of their pitch). And if we separate economically usefully human development from non-economically useful human development, the later is (by definition) a luxury good. This luxury good can be obtained for free on the internet, so paying $100-500k for it is just signalling. Hence it can be ignored.

Don't get me wrong, I want to live in a world where everyone's motivation is non-economic human-development... But that is not the late-stage capitalism dystopia we find ourselves in - and I think the author does a good job describing the state of things.

samcodeson Deep Learning in the Real World: Dealing with Non-Differentiable Loss Functions

Reading about how Julia does automatic differentiation helped me to understand this a lot:

http://www.juliadiff.org/

samcodeson Brain approaches tricky tasks in a surprisingly simple way

I love when people publish companion blog posts with their academic papers.

samcodeson Ask HN: What are unintended consequences of new tech you've noticed?

I do this a lot. It recently began to bother me. I started a list of things I have to do before I dick around on my phone: practice Japanese and meditate. I don't always remember, but when I do, I'm happier.

samcodeson Ask HN: Agriculture startups doing interesting work?

None of the jobs are explicitly remote - any heuristics for identifying which roles are remote friendly?

samcodeson Boeing Engineers Lost Control of the Company

Regulatory capture is a common feature of policies promoted by the deregulate crowd

samcodeson The State of Machine Learning Frameworks

Julia is a fantastic language. If I did original research I would use it, but I'm in industry and focus on productionalizing things.

samcodeson N8n.io – Workflow automation alternative to Zapier

Just registering my disagreement with this position. I think the author is right, most people are going to look at the difference between this license and true FOSS and say "whatever".

samcodeson New Query Language for Graph Databases to Become International Standard

I have used Gremlex (Gremlin in Elixir) for querying a DB that supports Gremlin (Neptune) and found it really pleasant.

https://github.com/Revmaker/gremlex

samcodeson How to be concise on multifaceted, complex topics like “experience” or “culture”

To run with this, I assume your potential customers aren't looking explicitly for "VR training". They're probably looking for "training that takes less time and/or increases $success_metric" - can you take this angle? Do you have data that show something like "patients treated by EMS workers trained with $our_product have $improved_success_metrics"?

Or maybe it takes less time? Something like "Spend less time on training. Our VE training takes 50% of the time as traditional training while improving $test_outcome"?