bitforger

Pretty cool.

I once worked on AI Dungeon and we had a similar idea to parse the story so far into a graph, so that we could manage long-term memory outside of the context window (which was only 2048 tokens).

Coreference is hard. ("he took the sword"... who is he?) Updating the graph is also hard. (As the story progresses, new facts contradict old facts. Jenny was dating Tom, but now she's dating Mike.)

And knowing what to do with the knowledge graph is hard too, especially if you don't know the schema up front. The only thing we could think to use it for was... programmatically turning relevant sections back into text and prepending it to the context window. (There were easier ways to get a similar effect.)

Arithmetic by Lockhart is also a gem.

yes

I think we are mostly on the same page. Pain and suffering are different things, and often fully embracing pain and letting go of aversion decreases the amount of suffering in that moment. So yes, it's very important that practice is motivated by wise consideration of the true nature of suffering, and perhaps I should have qualified that.

However, given that qualification, I still believe that practice should be a smooth descent of the gradient of suffering. You may increase your conscious awareness of pain or unwholesome states, but you should always feel less suffering after you do that. This is because letting go to the reality of the present moment is the opposite of clinging. So you should always feel better* when you practice than when you don't.

I will admit, though, that perhaps my perspective is colored by my experience and this is not true for all people.

*by feeling better we mean less suffering (dukkha), not more pleasure and less pain

Yes, I hope that most readers take away this sentiment as well.

I will say that some of Sasha's criticisms don't vibe well with my understanding, though. For example, we don't renounce pleasure because Buddha tells us we should, but because we directly observe that clinging to desire inherently feels bad. Our subconscious mind, when confronted with this info, naturally drops the object like a hot coal.

Telling newbies that traditional Buddhism is about giving up everything that makes you happy is a mischaracterization imo.

I also think his understanding of clinging is more narrow than what's described in the Abhidhamma. The interludes of TMI[1] give a pretty approachable introduction to this model of conscious experience.

[1] https://a.co/d/04tc3dJ

I've been curious about this in my own practice, but I haven't yet been able to find an example of increasing suffering as having a positive impact on my insight.

Do you remember the book or teaching that that analogy came from? Ajan Chah comes highly recommended from others as well.

Edit: oh I guess the exception is clinging to the Dhamma, like the sutta about the raft.

May you and all readers be well, happy, and healthy. :)

It makes me sad that many people will use this as an opportunity to write off Buddhist practices. Please don't! It has personally helped me greatly. Just remember:

1. The goal is to end suffering, so if a practice is making you suffer stop doing it.

2. Do what works for you. The only truth is what you can directly experience in the laboratory of your life. All the other teachings are just suggestions.

And if you'd like a concise overview of Theravada Buddhism (which is somewhat easier to grok without the added teachings of Mahayana, Vajrayana, Zen, etc.) I highly recommend [1] and [2], the second of which can be read in a day or two.

[1] https://a.co/d/iiAtDs5 [2]: https://a.co/d/asUIQUR

I think you can get the effect you're looking for by using the previous cell as an init image and only repainting the character.

As for consistency of character details, I think that will depend on how many images you use to train dreambooth etc. and how varied those images are.[1]

[1] https://www.youtube.com/watch?v=W4Mcuh38wyM

I haven't used a feed reader in a long time, but I had a brief period when I was obsessed with Fraidycat. Worth a look if you're interested in a different approach to keeping up with people.

https://fraidyc.at/

Sure

Suppose Twitter / FB were forced to comply with this law and remove censorship.

Could they do that only for accounts that are verified as being legal US citizens and continue censoring unverified accounts?

In that case I wouldn't be opposed to this, as long as it doesn't unleash a huge wave of bots.

I think this depends on if you're a constructivist, no?

We do this with Pinecone, but we use CLIP embeddings of images, and they work incredibly well. It's kind of crazy how easy it is to get semantic search of images these days.

CLIP also does caption embeddings, so you can lookup images via both images and captions.

Hey! Author here. I've noticed on certain mobile devices the contrast can be weird, not sure why.

I updated the default theme's text color to be 10% darker, lmk if that looks better to you.

I also just tested "X is Really Y" and the same thing happened. We could probably enumerate the adverbs and get a good sense of what the list looks like, if you're not afraid of getting banned :3

Wow! Super cool results.

This is just begging to be text-guided with the lyrics. Then you could put in the lyrics and have the model generate a clip it thinks would go with the lyrics. :)

Nope!

Also very interested in this. AFAIK, the best alternative to DALLE-type generation is CLIP-Guided generation (such as Disco Diffusion [1] and MidJourney[2]) which can take anywhere from 1 - 20 minutes on an RTX A5000.

[1] https://github.com/alembics/disco-diffusion

[2] https://twitter.com/midjourney?t=-kKC5UE-gjIkMvAb709SyQ&s=09

If you want to try out some of these models, AI Art Studio[1] has Pixray, VQGAN, and Disco Diffusion available.

[1] https://ai-art.latitude.io/

Cool! I hadn't heard of Overtone before, but I love the idea of live coding music.

People might also be interested in a similar project called Sonic Pi[1], which is Ruby based.

[1] https://sonic-pi.net/

Interesting how you still need to have some intuitive sense of what's going on under the hood. (You can't say "make Zelda," you have to ask for an array of symbols and manipulate them.)

In that sense it feels like this is still programming, but at a higher level of abstraction with a weird fuzzy compiler. Now we can go from natural language -> JavaScript -> assembly etc. rather than just the last two.

Mediocre programmers use APIs, while good programmers know what's behind the curtain and can debug them. I suspect this will stay the same, no matter how many layers of abstraction we add.

I believe they're currently using AI21's 178B Jumbo model for Dragon. Since they're completely off of OpenAI now, the content filter is much more lax.

A personal favorite example of notation improving my ability to think is Named Tensor Notation, which has gotten some attention in the machine learning community.

https://namedtensor.github.io/

Some diehard mathematicians have said it's a crime against linear algebra, but I think it's much more useful for talking about things like neural networks than conventional notation.

Reminds me of The Daily episode a few days ago about what's happening in Japan. https://www.nytimes.com/2021/05/05/podcasts/the-daily/japan-...

That is a good friend, you should keep them

Can you share your copy of the Strand book? A cursory Google search makes it seem somewhat elusive...

If I had a dollar for every caching bug I've encountered, I'd be a rich man.

I wrote a blog post a while back about whether methods like these will be necessary in the future:

http://mitchgordon.me/machine/learning/2020/01/13/do-we-real...

A alternative to this has recently been published at ICML that claims to be faster. The website and tutorial video are very nice, too.

https://linear-transformers.com/