formalsystem

Yes you're correct, I'm just giving the intuition that I found helpful to understand the Principle of Least Action.

Is anyone aware of any work for how to apply differential privacy to language models?

So the main question I have is let's say I'm working with sensitive data like emails or doctors notes. How can I train an ML model that would still learn something useful without leaking private data.

When I say "leak", an example would be I train an RNN on some company data email data and when I feed the RNN "$AMZN" the network would say SELL.

How can I quantify how much the model has learnt and how much privacy has been leaked.

This is a great book but a bit dense first. At a high level it goes through physics with an optimization viewpoint, as in find the actions that minimize a system's energy to figure out how a system will evolve.

I would strongly suggest you learn Lagrangian and Hamiltonian Mechanics from this book first [1] since it comes with many more illustrations and simple arguments that'll make reading SICM much easier. If you don't have time to read a whole book and want to get the main idea I've written a blog post about Lagragian mechanics myself [2] which has made it to the front page of Hacker News before. The great thing about SICM is that it's a physics textbook where the formulas are replaced by code [3] which you means you can play around with your assumptions to gain intuition for how everything works.

IMO I believe in introductory physics we overemphasize formalism over intuition and playing around with simulators is a truer way to explore physics since most physical laws were derived via experimentation not derivation. Another book that really drives this point home is [4]

[1] https://www.amazon.com/Jakob-Schwichtenberg/dp/1096195380/re...

[2] https://blog.usejournal.com/how-to-turn-physics-into-an-opti...

[3] https://github.com/hnarayanan/sicm

[4] https://natureofcode.com/

I had setup Pi-hole and it was blocking something like 66% of my internet traffic which amounted to ads and spyware.

A couple of days later, it just mysteriously stopped working and I couldn't for the life of me figure out why. My theory is that AT&T detected my blocking and blocked my internet as a result. Is that too much of a tin foil hat theory? How can I know for sure?

I don't know too much about computer networking, only the basics.

Yeah definitely not my expectation to make a living out of it, I'm just doing this for fun but streaming into a void isn't as fun as streaming to a handful of people at a time. On Twitter for e.g it's easy to get started with 0 followers since a few insightful replies to posts by popular people can bootstrap you. A popular streamer bootstrapping me on Twitch has a lot more friction and is a lot less likely to happen.

I hear you, I'm sure some streamers got to where they are with sketchy policies but for the most part the streamers I follow are either Dota 2 pros that have spent years of their life getting good at the game and also stream very consistently (almost everyday) while still being entertaining for 3-5 hours at a time, they also have dedicated editors which will do shorter clips with the occasional meme on youtube if you weren't able to catch the stream live.

For programming related streams I follow Jonathan Blow because he has a unique viewpoint around programming and also because I trust him because he's shipped multiple succesful games.

Be succesful IRL seems to be one of the best ways to get a popular stream.

Also great advice, Youtube is better for well edited content and I have been teaching myself Da Vinci Resolve and have gotten a lot better at it over the past few weeks. Once that takes off, it may be easier to cross pollinate my Twitch channel.

I have a Desktop PC that I've built, AMA if you want to get started. I have a 32 core Ryzen CPU so I'm really well equipped.

This is great advice, contribute to various Discord and Slack communities and once you establish trust start plugging your stream and even have various people from the community as guests. I guess key is to give give give before you expect people to reward you with their attention.

Thank you, I really appreciate that. https://www.twitch.tv/videos/606535356

I'm slowly getting better at it too and am planning on streaming some Julia Differential Equations stuff tomorrow PST.

Yeah reach out whenever you like, my email should be in my profile

twitch is one of the platforms where I feel like I've struggled the most to get viewers. just playing my main game like Dota wasn't too productive since I'm not a pro or super hot so my value proposition there isn't too compelling. Instead I've been trying to focus more on streaming myself programming and learning new scientific or game programming libraries and have been enjoying it quite a bit. My viewer count is veeery slowly increasing but almost noone is subscribing, many of my friends have expressed interest in watching me so my plan is to start letting them know when I'm about to stream to seed some viewers. Also I've realized that the more specific your brand is on social media the more effective, I've been looking at branding myself as a strategy game buff/developer so will be streaming niche strategy games and Unity game development every weekend. I'll re-asses after a month to see if this plan was effective.

I really appreciate this project as going from 0-1 viewers on Twitch where the 1 isn't your friend is challenging.

I went out to grab a burger last night and there was a huge line of Uber Eats drivers and they were waiting for what seemed like a long time especially considering how little they're paid.

Is there any reason why an OSS Uber competitor wouldn't be tractable? Drivers and passengers can pay a small booking fee which would go towards maintaining server costs. Client app can be really bare bones and I believe people would be willing to use it if it's way cheaper. So many scammy people talking about decentralized computing but this seems like a low hanging fruit which gives gig workers their autonomy back.

Same question for all gig worker apps.

Thank you for this, I just bought the book. Any books you'd recommend to go over the prerequisites you mentioned? I read this [1] ages ago but I honestly don't remember much at this point.

[1] https://www.amazon.com/How-Read-Financial-Report-Wringing/dp...

I'd be curious to read a study as to productivity tools actually do help people be more productive.

Anecdotally, I used to use many different tools from EverNote, WunderList etc.. but over the past 2 years my process has been use Google Drive as a journal and write down my top priorities on a post-it.

Post-it naturally forces me to prioritize a small number of tasks and my journal is to help me document everything I learn. I don't see ticking off low leverage items as a value add to anyone even though the dopamine rush does feel good.

What's the evaluation criteria for this work? How do I know if a piece of computer generated music is good or bad in general? What effect does human involvement have on the evaluation?

start a fund with your ML approach - evaluate how much money you've made after 1m,6m,1y,2y,10y etc..

So many papers by "smart" people predicting the past

Does anyone know of a good distributed systems resource? It's one of the rare areas of CS I have a lot of trouble finding good tutorials and textbooks. Is it just something you pick up with pure practice?

I don't know what we've built in the US but it certainly doesn't feel like we've built cities for people. In California, most of the outside real estate is dedicated to roads with the monotony occasionally interrupted by a parking lot with the same cookie cutter businesses. It's very rare that I'll see a mom and pop shop here and for the most part the outside world with billboards, malls and fast food chains is a distribution channel for big businesses.

Anyone familiar with the history of how PDFs became such a widespread format in the first place? I get that it looks nice but not being able to edit it by default just seems weird to me.

I wrote an e-textbook that'll teach you all the Math and Machine Learning you need to simulate robots and start building robots at home. A couple of chapters have made it to the front page of Hacker News.

http://robotoverlordmanual.com/

Taleb elaborates on this idea quite a bit. Opinions are irrelevant, what matters is your position. Opinions are useless without a downside.

For example, don't tell me that you think Microsoft is a good or bad company. Tell me how many shares of Microsoft you own or how you're shorting Microsoft.

I think it's totally OK, normal and beneficial to want to experiment with different applications.

The main issue is after a few years you may feel like you have nothing to show for it and the best solution to that feeling that I've found is to document what you've learnt in a blog post. Over time you'll end up with a large personal knowledge base on which you can draw on when you want to and actally can finish a complex project. You'll also attract the right kind of people that can help you by putting your thoughts down in writing.

The idea is you want to slowly build yourself to be the kind of person that can finish a complex project in a weekend and the best way to do that is to constantly be in the high payoff space you get when you're learning something totally new.

The author's previous book Game Programming Patterns is 1000x better than the Gang of Four book.

Really looking forward to going through this.

Video editing (Adobe Premiere Pro + After Effects) and Video streaming (OBS + Twitch).

I figure it's gonna be a while before we get back to normal. People will continue to crave attention and video is a more scalable way to disseminate your knowledge vs traditional meetings or Zoom calls.

I'm specifically looking at doing more analytical video game streams and educational Machine Learning content for the advanced n00bs.

Who we view as charismatic will change quite a bit in the near future since we can endlessly edit and improve our message before we share it with others.

Role: Machine Learning Scientist, Software Engineer Location: San Diego Willing to Relocate: Yes (US, UK) - US citizen

Technologies: Machine Learning, Python, Tensorflow, Pytorch, Julia, C#, Unity, AWS, Node, SQL

Skills: math, optimization, physics, graphics, design

Resume: https://www.overleaf.com/read/wkbttymdcmqt

Personal site: robotoverlordmanual.com

Email: marksaroufim@gmail.com

Hi I’m Mark, I’m an Applied ML Scientist and Product Designer. I'm the founder of yuri.ai where my goal is to make it really easy for game developers to balance their games using Reinforcement Learning. I’m looking for a job because the Lebanese banking system is collapsing and I’ll soon need income to support my parents.

How I can help you:

I can write top notch documentation and can explain anything to anyone: My book robotoverlordmanual.com is a visual and accessible robotics, ML and math textbook with over 28000 monthly viewers. I’m very comfortable writing and speaking.

I can manage your most complex projects: I was the BI lead when Microsoft was selling its display ads business to AOL, I made sure Outlook AI efforts were compliant. I’ve worked on projects with 100+ stakeholders and have bootstrapped projects where I was the first engineer to 10 engineers.

I can turn your research into a product: I’ve done this with Yuri, I’ve done this at Microsoft when I was working on a next gen email ranker and a part of speech tagger and I’ve done this at NASA’s Jet Propulsion Laboratory when I was setting up their computer security anomaly detection pipeline from scratch

I can setup your entire BI infrastructure and measure what matters: I’ve done this for more than 10 teams at Microsoft. I will help you find and measure the metrics that are most indicative of your product success.

I will also get along with your dev team: I have extensive science and development experience and can tell the difference between realistic work and sci-fi. I have extensive experience with tooling and research in BI, ML, RL and game dev.

Please ask me questions!

I love this. I've been messing around with Premiere Pro and Audacity for the past couple of days trying to get more into making video. Video puppet looks way easier to debug and collaborate on since scrolling back in forth in your video looking for stuff gets very tedious very quickly.

Is there any way I can add my own voice and then still write the words that I want my voice to say?

Role: Machine Learning Scientist, Software Engineer Location: San Diego

Willing to Relocate: Yes (US, UK) - US citizen

Technologies: Machine Learning, Python, Tensorflow, Pytorch, Julia, C#, Unity, AWS, Node, SQL

Skills: math, optimization, physics, graphics, design

Resume: https://www.overleaf.com/read/wkbttymdcmqt

Personal site: robotoverlordmanual.com

Email: marksaroufim@gmail.com

Hi I’m Mark, I’m an Applied ML Scientist and Product Designer. I'm the founder of yuri.ai where my goal is to make it really easy for game developers to balance their games using Reinforcement Learning. I’m looking for a job because the Lebanese banking system is collapsing and I’ll soon need income to support my parents.

How I can help you:

I can write top notch documentation and can explain anything to anyone: My book robotoverlordmanual.com is a visual and accessible robotics, ML and math textbook with over 28000 monthly viewers. I’m very comfortable writing and speaking.

I can manage your most complex projects: I was the BI lead when Microsoft was selling its display ads business to AOL, I made sure Outlook AI efforts were compliant. I’ve worked on projects with 100+ stakeholders and have bootstrapped projects where I was the first engineer to 10 engineers.

I can turn your research into a product: I’ve done this with Yuri, I’ve done this at Microsoft when I was working on a next gen email ranker and a part of speech tagger and I’ve done this at NASA’s Jet Propulsion Laboratory when I was setting up their computer security anomaly detection pipeline from scratch

I can setup your entire BI infrastructure and measure what matters: I’ve done this for more than 10 teams at Microsoft. I will help you find and measure the metrics that are most indicative of your product success.

I will also get along with your dev team: I have extensive science and development experience and can tell the difference between realistic work and sci-fi. I have extensive experience with tooling and research in BI, ML, RL and game dev.

Please ask me questions!

Viewing Linear Algebra as the study of linear operators instead of matrices makes everything so much simpler.

Of course AB != BA

Composition makes sense

Inverse makes sense

This is the book that helped me get it http://linear.axler.net/