Month: January 2023

Finally got around (eyes on) to checking out the Neuralink conference by Musk. Three hours long. Highly recommended for all engineers. Every minute of it showcases things next to which all sorts of ChatGPT and other mind-boggling gadgets simply look like child’s play. If these guys succeed, they’re going to rake in all the money in the world. There’s also a similar video about Tesla, but that one is simpler and more down-to-earth.

In short, for those out of the loop. Neuralink is developing a robot that drills a hole in the skull, positions a special sewing machine next to the hole, which drives about fifty electrodes into your brain. Then, the hole is sealed with an electronic device with a remotely rechargeable battery, and you get the ability to control a computer using that part of your brain (still a work in progress, but that’s the goal, and there are working prototypes). Like typing text or pressing buttons in an interface. And in the future (so far, just distant plans), the reverse is also possible: understanding what’s happening in the outer world in ways other than our five senses.

And there are a lot of technical challenges in this process, and the video talks about them and how they were overcome.

Their website:

Unfortunately, Boston Dynamics and SpaceX don’t make such wonderful technology videos.

I wonder, what value does reading Shakespeare in translation hold if all that remains of Shakespeare in his tragedies is merely the script? Translating a sonnet into Russian, let’s say by Marshak, is very liberal. It essentially starts from scratch, a composition stretched over a framework of dialogue. Furthermore, Marshak’s translation is very beautiful, but here I would give Marshak 9 points for the beauty of the translation and 1 point to Shakespeare for writing texts that are difficult to translate 🙂

I need to find some time to try reading something in the original. I imagine an interactive book where you click on a phrase and a brief comment appears, click again and a detailed one comes up, with context and details. I wonder if anyone has made such a thing?

I play “Elegy” by Nobuyuki Tsuji (a Japanese composer, blind from birth). Originally, Nobuyuki dedicated it to the victims of the earthquake and tsunami, but…

Good Rules by Kieran Drew

I recently set up Stable Diffusion locally with the default model and openjourney – which is an unlimited prompt-based image generator. And here’s a thought. The prompt should be formulated as if the computer crafted it, looking at the final image. Computers recognize patterns in images that have previously appeared in the training dataset. Thus, it can’t properly draw a three-legged cat unless the dataset included three-legged animals. Likewise, it would see the cat in a three-legged cat, not the three-legged aspect. In essence, abstract thinking is severely limited to what it has been trained on.

Yet, it’s possible to train using algorithmically generated objects. For example, if you define three-leggedness in modern animation programs, you could generate thousands of diverse animals on such a skeleton, in various poses. It would be interesting to develop a system based on this. In theory, this approach could extend beyond just animals. Implementing a CAPTCHA based on ai-generated results could lead people to filter and tag results for free.

So, today I finally snapped and decided to create a book with AI) took me 40 minutes. Text by ChatGPT. Pictures by Midjourney. Russian translation by DeepL. Layout is my own. Click on the title, then 12 pages follow.

Just a heads up – it’s not a cheerful one!

So today I snapped and decided to make a book using AI) it took 40 minutes. The text is from ChatGPT. The images are from Midjourney. The layout is mine. Click on the title, and then there are 12 pages.

Just a heads up—it’s not a cheerful one!

A small new batch of my photos from various times

Sixth painting. There were grapes, now there’s wine!

I believe that the next big breakthrough will be based on the combination of generative AI with AI-driven depth map reconstruction from a single photograph. In fact, it already exists — MiDaS, and several others. Why this is interesting — essentially, it allows for integrating objects into the environment on the photograph in such a way that shadows, palette, and lighting are taken into account. Currently, this is challenging because, figuratively speaking, the AI does not know that the surface of the table in the photo is unevenly lit not just by chance, but because it is angled towards the light source in such a way, and that tree over there creates a shadow. With a depth map, this begins to make sense.

I don’t yet understand how to implement this right now, but it feels like it’s very much the near future. NVIDIA demonstrates recreating 3D from several photos — this is photogrammetry through AI, much faster and at first glance very accurate.