formalsystem

I found open offices to be an incredibly alienating experience, felt about as personal as a a cubicle since everyone is pretty much using headphones and "deep working" and you can't really make much noise since you're disturbing an entire org if you are.

My favorite experience was working on a team where the open office was essentially a large office for about 6 people only. That felt great since there was a more communal vibe and I had sufficient personal space. We had a couch, a table with a chessboard and some people brought snacks over.

It seemed to me like when I went to grad school that a lot of students were there for the freedom to explore their interests. They then quickly realize that their freedom is subject to a bunch of constraints like the availability of grants or the specific interests of your advisor.

And if you're going to be that constrained may as well get a job and make some money while you're doing this.

Your salary as a PhD student is a joke but you're still an expensive asset as far as your advisor is concerned because of your tuition which pays for stuff like maintaining the campus, allowing you to take classes, your office etc..

However, I feel that most of those resources are useless as far as a student is concerned. You just need to pay for rent and maintain a relatively comfortable lifestyle. So something like $2,000 a month would be more than sufficient for a student and is a tiny cost for an investor looking to have interesting research published either as a paper or open source project.

The only real working relationship in a graduate program is the relationship between faculty and student so maybe it's worth it for some of the wealthier faculty to cut out the middlemen.

Besides the obvious applications like search and voice assistants here's a few really good ones

* https://www.grammarly.com/ is pretty interesting. Corrects your grammar and not just your syntax

* Swype keyboard which uses your spelling corrections as a way to better figure out how you use a swype keyboard

* https://flexibits.com/fantastical and many other calendar apps do NLP stuff to pull out meeting info. Microsoft, Google and Facebook do this kind of stuff pretty well (Disclosure: I worked on such features at Microsoft)

Happy to give more examples if people are interested in working in this space.

I had no idea VS Code even had tracking..

Does anyone know what exactly is being tracked?

Feelsgood that we're moving towards a single modifiable repo of history! What could go wrong?

How does this work exactly? I notice it's a partnership with Visa, are payments converted to dollars or is the payment in Bitcoin directly?

Also does Visa or Coinbase then make a server side log of all the transactions?

Sure but the output would still leak data. E.g: You give the RNN the phrase "Chase Bank" and it outputs "Sell the stock".

Encrypting the data secures the data loading part but it's not like your model didn't learn anything.

So let's say I'm in a regime where I'd like to train something like a language model RNN on a private text dataset. If I train the RNN directly on the data then I'm essentially leaking private data to the output. Do you know of any approaches to generate fake data using GANs, train a language model on that fake data and still get a good classifier that does well on test data while still quantifying how much private data is being leaked?

Related: do GANs work well as a data augmentation technique and can their exact contribution to how much a model could potentially be improved be quantified.

EDIT: Added some clarifications

Why not just make a new client? As others have mentioned, I don't see this landing well with Discord.

Just because there are a bunch of group chat apps already, doesn't mean people won't switch to a better one.

When I was a teenager I used to play chess competitively and when I mention this to people, they look at me all impressed. When I mention that I now play Dota competitively, I get an "oh I hear that game is really addictive" or "I'm not a video game person". And since I've played both for quite some time now I can promise you that playing any interesting game is not a passive experience and is in fact pretty brain melting if you get into it.

"I don't like games" to me now sounds a lot like when people say "I don't like music" or "I don't like books". You may not like specific genres but not liking an entire medium is often just a sign of miseducation.

I would challenge the most dedicated of programmers to build an efficient factory in Factorio or an adventure buff to see what it feels like to beat a monster like in Sekiro or Dark Souls. The Witness, Baba is you, Crusader Kings 2 etc.

Please play any of those games and tell me that gaming is a passive activity.

Does cloud infrastructure have to be centralized for the foreseeable future? I'd imagine there's a bunch of people even on HN working on ways to decentralize compute, something like you run a shell command and part of your computers resources are now allocated for anyone part of that ecosystem to use.

Thank you for those articles, would you happen to also have books you can recommend that informed your thoughts on government policy?

Yeah I've been researching Rust a bunch for my own projects. Mostly around doing efficient reinforcement learning and physics engines so I've been finding lots of cool repos as I research.

Not really, I'm just a fan of the ideas in Rust and it's nice to see codebases that do interesting stuff.

Does this work typecheck matrix operations?

Genuine question: Is this restructure for the purpose of taking government military contracts? I don't see how investors would be getting 100x returns otherwise and my understanding was that salaries for employees was competitive with big tech companies. Curious where Open AI feels like there's money to be made.

1. We use a couple of tricks for this, there's ideas from human preferences. There's modeling a reward function to maximize fun instead of pure performance, there's handicaps and a host of other tricks. Different tricks work better for different games but game developers can easily configure them

2. It can, imitation learning turns RL into a supervised learning problem though from my experience it's not really needed. It's more of a if you already have it then you can make the training a bit faster

3. Yes all the RL part becomes tuning a config file where the game developer needs to describe the task it wants to be accomplished

Regarding technical details, the infra is agnostic to the specific deep learning library you'd like to use and the algorithms are chosen based on the types of inputs (continuous vs discrete), size of the space (the end all of strategy games or a casual puzzle game) and type of game (full information vs partial information).

I think any game with strategy is going to have to use reinforcement learning to have any hopes of being understood by its players. Concerns around cheesing strategies can be mitigated for real time games using constraints and cheesing in turn based strategy games is a symptom that maybe the game you're working on isn't that strategic.

Ultimately Chess, Go, Dota, Starcraft are compelling games to watch and play because the skill cap is so high. No hand coded AI for such a game could ever prepare you to enjoy playing any of those games professionally or get to a point where you can understand professional play. High level strategy in those games would never even be discovered had we not had a large population of players playing those games for long hours.

Odds are most strategy games won't be as successful as any of the above games but the newcomers could still have compelling strategic elements that gamers could explore had they been able to find other players that are slightly better than they are so they learn but at the same time aren't frustrated by too big a skill gap. A small multiplayer community is why the strategy game market is pretty bleak except for a few large players.

Reinforcement learning can provide a better experience to your players because you the game developer don't need be the world expert at the strategies of your game to create a top level AI. You just make sure the rules of the game are there and self play will take care of the rest. You can constrain the AI such that it's enjoyable by both your newest and most devoted players. You can discover OP strategies and better guide your design process. Excel is great for balancing but if you want to make a game complex enough then balancing will be a tremendous challenge without a larger team.

If you're interested in leveraging reinforcement learning for your game, I run yuri.ai which helps game developers do exactly that. So reach out to me at mark@yuri.ai if I can help!

> Our hope is that this kind of machine learning approach can strengthen the business case for wind power and drive further adoption of carbon-free energy on electric grids worldwide. Researchers and practitioners across the energy industry are developing novel ideas for how society can make the most of variable power sources like solar and wind. We’re eager to join them in exploring general availability of these cloud-based machine learning strategies.

Would be nice to know a bit more detail? Which power grids was this approach tested on? How does this approach work exactly? Are we talking about DNN on time series? The graphs also really need labels to inform.

Happy to collaborate on it if you like, shouldn't be too hard

I'll check it out thank you for the note!

You'd be entirely right not to trust it as it is in an industrial setting. There's been some research around safe exploration that would add additional terms to the reward function to do things like punish flailing around and such but I haven't experimented with those techniques myself.

The analytical solution wouldn't involve any training, you'd solve a system of equations where position of the finger is equal to where the goal is constrained by the equations which tell you how an arm moves.

The advantage of the RL approach is that it doesn't need to know how an arm moves but then involves some training.

What's nice about the move is that the reinforcement learning algorithm doesn't fundamentally change. The action and state spaces will be larger since a joint has more degrees of freedom in 3D than it does in 2D so learning may take longer so we'll need to also increase the size of the replay buffer and the episode length.

I'm planning on writing a followup post on 3D so stay tuned!

I really enjoyed the section on symbolic data in SICP https://xuanji.appspot.com/isicp/2-3-symbolic.html - gives lots of good insight to start thinking about how automatic differentiation is implemented in the deep learning libraries.

The book is overall excellent. I learned way more from it (and had way more fun) than I did ever reading something like the Gang of four or its various explanations.

Best book to learn programming patterns in general if you ask me

Bad blood: couldn't stop flipping the pages, reads like satire. Seeing the names of the most powerful people in the US and how they enabled Theranos' rise to prominence made me realize how much we fetishize geniuses and how little popular experts know. Books like these remind me not to pay too much attention to how people frame their success

Leonardo da Vinci by Isaacson: I went into this book expecting to be blown away by how much smarter Da Vinci was than me. Instead I was blown away by how human and hard working he was and it taught me valuable lessons on how to structure my time and life to work on things I care about. This book was instrumental in my decision to quit my job

Modern Robotics Mechanics planning and control: Because it gave me the mathematical foundation to understand how robots work. It's dense but brief so well worth the attempt. Robots was something I've been wanting to get into since I was a kid and seeing a concise treatment like this one with good intuition, mathematical formulas and code reminds me that these days I can learn anything I want

Thanks for coming on OP. Am wondering if you have any plans on sharing the code as well?

SEEKING WORK Location: San Diego; Remote; Consulting

Experienced ML engineer and product designer. I've built ML infrastructure that has interfaced with millions of people around the world. I've worked as a researcher and have a strong mathematical background but also understand what it means to prioritize things and measure success.

Skills: Machine Learning, Computer Vision, Robotics, Business Intelligence, Product Design

Languages & Frameworks: Tensorflow, Keras, Pytorch, Unity, C#, Python, SQL

LinkedIn: https://www.linkedin.com/in/marksaroufim/

Contact: marksaroufim at the gmail dot com

Background:

* Applied Scientist & PM at Microsoft * Data Scientist at NASA's JPL * Visiting scholar at UC Berkeley * M.S in ML from UC San Diego

Open source contributions: https://github.com/msaroufim.