rajansaini

A successor to OpenLLaMA could make this possible. Imagine being able to deploy your own LLM.

Read the InstructGPT paper and see alpaca. You just need instruction fine-tuning.

Unlike what the other commenters are saying, RLHF, while powerful, isn't the only way to get an LLM to follow instructions.

Seconded! If you make it easy/fun to share I (personally) would totally do so :)

Could it be useful to use these to charge long term batteries to be used millennia from now? So that future generations living centuries from now can have easy power when the fossil fuels run out?

I am surprised that Algorithms to Live By has not been mentioned. (Disclaimer: I am still reading it.) That book is both accessible and practical for any layman. However, just enough hints at variations of the discussed algorithms are given that even a non-technical reader might be motivated to dive further into theoretical CS and read a paper or two.

You should check out the VOLT paper, I think it would work well. It's a new technique for splitting up a vocabulary into subwords while minimizing entropy. These subwords could then be mixed and matched, maybe by a neural model, for better results.

I'm only a student, but I personally really liked https://jeffe.cs.illinois.edu/teaching/algorithms/

This is better for more advanced topics, like dynamic programming (well, advanced for me, anyway). I started out taking over an hour to solve the first problem in the problem sets, but they build on each other slowly, so I was soon able to see solutions within minutes. It took me a weekend to go through chapters I was struggling with, and I did really well on my coding interview the next day.

Location: Southern California

Remote: Yes

Willing to relocate: No

Technologies: Python, C

Email: [email protected]

Looking to do something math-heavy part-time. Interested in applying math from ML to other domains (am currently obsessing on a maritime shipping optimizer, if that's any indication of my interests). Controls and robotics are very interesting, too. Computer vision/embedded background, currently interning at Amazon. Looking to do some freelance work from January to June 2022.

Bonus points if you need neural networks sped up (pruning, sparsifying, quantization, etc.); that's been a past research focus!

In my experience anyone can be conditioned to love anything, and inspiring a love for learning something is the only stable long-term solution.

I think pyrtl was designed to enable abstraction in hardware design; it emits VHDL. Would be curious to know if anyone else has used this tool.

Those are very interesting empirical results. This lecture explains the deeper vs shallow tradeoff theoretically: https://www.youtube.com/watch?v=qpuLxXrHQB4. He's an amazing lecturer; wish I didn't need subtitles!

(If you're too lazy to watch, it turns out that there exist functions that a shallow network can never approximate)

> I wish we'd see an app that teaches financial literacy and responsibility

I agree with you 100%. I'm not sure how feasible it is to build a business around it, though, since monetization techniques are likely to turn people off or harm the product. Bogleheads is an awesome existing resource, in case anyone's interested

That's really sick, UI is really pretty! As an idea for a follow-up project, I wonder how difficult it would be to make it non-deterministic; i.e. play a different variation each measure.

Sick project! When's the IPO? :D

Hopefully someone can provide more info, but apparently there are some bacteria in geothermal vents near Baja California that seem to be able to handle venus-like conditions. No idea what the details are though

Why is that out of curiosity?

Seconded!

Didn't Yann LeCun do important work on CNNs there as well?

I'm not at all familiar with this domain, but wouldn't this kind of farm be incredibly vulnerable to a foreign threat?

I'd like to play the foil to enrich the discussion.

Our instincts force us to seek validation and approval from others as individuals. I wonder if the success of blogs, Twitter, and academic research stems from incentivizing people to maximize that self-centered sense of ownership. For example, being first author on a well-cited paper is a powerful motivation to produce better research, which in-turn benefits the community and strengthens the institution.

Current Location: Irvine, CA and Goleta, CA

Remote: Yes

Relocation: Yes, within US/Canada

Technologies: python (TensorFlow/keras, numpy, pandas, scikit-learn, PIL, opencv, librosa, jack), C++ (TensorFlow, ImageMagick, uWebSockets), Java (Swing, sockets), web languages (React + TypeScript + XState)

Resume/CV: https://drive.google.com/file/d/1OeozjtrIiMQcsSZ0xa09uix3Lpy...

GitHub: github.com/rajansaini691

Email: [email protected]

I am a third-year undergraduate with a strong research interest in embedded applications of ML and vision/audio problems. I spent this summer helping a research company speed up/improve accuracy of their object detection models and the summer before that doing embedded work for a physics lab.

Seeking an internship for 2021.

I didn't read the full paper carefully, but does this mean P != NP? :)

Intuitively, it seems like voice signatures, body language (like walking habits), and height/weight would play a larger role in helping humans identify a masked person than exposed facial features.

Because the network behind Apple's facial recognition software cannot have access to this kind of data (well, maybe voice, but that doesn't seem secure), I'm not sure this is a fair comparison.

Would love to be refuted, however.

I think today's technology does stifle creativity as well, though. Music was meant to be spontaneous and improvised, which is very easy to do with older synths and acoustic instruments. Production in a modern DAW is anything but that, and it's way less tactile.

That's a very interesting perspective. There have been many articles complaining about the "hype" behind ML, but I too wonder if DNNs could assist with controls, especially when it comes to reacting to sensor data. After all, it's just matrix math.

I want to respond to your point about violin-playing, because I think it will open up an interesting discussion. Anecdotally, I was always motivated to practice my instruments by the same exploratory drive discussed in the article. I was taught to ask, "what if I did XXX this way?", i.e., what happens to my sound if I play long tones, what techniques do I need to play piece Y, how do I use scale Z to improvise?

Growing up, I found that those who ran on pure extrinsic factors with all discipline and no enjoyment either stopped or were stunted by locally optimal decisions. On the other hand, those who went on to study in university and play professionally seemed to derive the same obsessive enjoyment from their craft as I would a video game.

Do you think these data points further prove your point about emotional maturity? Or is there something more universal, that true success requires a fierce passion?

Whoa thanks!! I'll definitely take a look! Have you happened to come across anything dealing with optimal initial conditions that was particularly noteworthy?

This is gold to me, thank you.

I apologize, what do you mean by proxies?

Do you have any recommendations for a less-abstracted view of history (focusing on individuals rather than their emergent structures)? Or branches of philosophy that are particularly illuminating in this context?