Just listened to this 8 hour Neuralink episode. It might not be fun for all of you, so it's not a general recommendation, but to me it was fascinating. First about how the technology including robotic surgery works, but mainly about the weirdness of controlling something with your brain. I have spent hundreds, maybe more than a thousand hours on some sort of EEG brain computer interface, so I got plenty of training. The resolution of these is not perfect, but you can do many things even with an EEG device. What is interesting is that I am controlling things that have no analog in the motor cortex, I am changing my state of mind and focus, which is very different to moving (or attempting to move, or imagining to move - three different things as per the podcast) a finger. The brain does not have direct receptors about its internal state. The brain does not hurt. The EEG feedback adds this information and it's possible to train doing completely new things with the brain. The interesting fact is that even when I learn it to some extent, I can't really tell how it works - or what exactly I do. I call it surfing on the wave all the way down, but it's something that you would not be able to replicate without figuring it out yourself. We can only have shared vocabulary after you experienced it . This I believe is precisely why it's hard to explain psychedelic experience to someone who has never experienced it. You might describe the visual part (which is not shared - I don't do the visual part at all actually, even when I dream, it's never visual. I know the concepts, the structures, but I don't see it), but that's only because that's the common ground ("I have seen beautiful fractals") and probably the least interesting part of the experience. Why I'm describing this? It's very hard to explain that there are things you can learn to do with your brain that we don't even have good words for. I believe brain computer interfaces with technologies like neurofeedback will make learning it much easier than for example meditation. Combining the dedication of meditation practice with the boost of neurofeedback is the best I believe. But most of you don't even understand what I'm talking about, because you never experienced it. It's out there. https://fountain.fm/episode/YuR0YvrA3JyOXHRiE13r

I've been playing with continue.dev (I am using it with Venice's LLaMa-405b or with local Gemma2-27b. My setup:

Vimeo

Ollama-like Venice proxy demo

More info about the project: https://pay.cypherpunk.today/apps/26zEBNn6FGAkzvVVuDMz3SXrKJLU/crowdfund

Continue can also use Claude. Is Cursor any better? Anyone has experience with both? Replying to Andrej Karpaty's remark here: "Programming is changing so fast... I'm trying VS Code Cursor + Sonnet 3.5 instead of GitHub Copilot again and I think it's now a net win. Just empirically, over the last few days most of my "programming" is now writing English (prompting and then reviewing and editing the generated diffs), and doing a bit of "half-coding" where you write the first chunk of the code you'd like, maybe comment it a bit so the LLM knows what the plan is, and then tab tab tab through completions. Sometimes you get a 100-line diff to your code that nails it, which could have taken 10+ minutes before. I still don't think I got sufficiently used to all the features. It's a bit like learning to code all over again but I basically can't imagine going back to "unassisted" coding at this point, which was the only possibility just ~3 years ago."

Wow, Cypherpunk Now has created some beautiful art again. https://www.cypherpunknow.com/cypherpunk-puzzle/

My friend @Wilder wrote (my commentary below): "I'm traveling to France on Tuesday and just thinking: If Pavel Durov deserves 20 years in prison for not cooperating with the French government and thus supporting terrorist groups and the spread of child pornography, then I, who publicly defends him (because I would never cooperate with the state either, just like I haven't worked for it for ten years now), thus also supporting terrorist groups and the spread of child pornography, would also deserve at least some seven years in a French prison :-) And now, seriously - where did Pavel Durov make a mistake: He designed the whole Telegram in such a way that it is not end-to-end encrypted (unlike, for example, Signal or SimpleX); thus, Telegram has and had the technical ability to eavesdrop and censor its users, which by law it should have taken advantage of and shown a public "cooperative effort" when they ran out of court orders from any country, which they didn't. On the contrary, Pavel Durov was reduced to criticizing Signal's security and making up stories that there were backdoors. No one will likely arrest Signal's creator, Moxie Marlinspike, or its president, Meredith Whittaker, because they can claim they tried hard to cooperate with the state. Still, unfortunately, due to Signal's end-to-end encrypted design, they failed purely technically to eavesdrop or censor anyone who used Signal. Who I think can be locked up in this case (not just in France) is Elon Musk, because he literally brags about how he is throwing in the trash court orders from various states to censor posts on X.com and therefore realistically refusing to cooperate with states. He has a personal dispute with the Brazilian Chief Justice, Alexandre de Moraes, whose court orders he publicly ignores. Musk has closed the X.com office in Brazil, and I would not go to Brazil if I were him unless he wants to end up like Durov in France. Even though I don't use Telegram and never have (it's simply not secure because it doesn't support end-to-end encrypted communication by default) and I urge all people not to use it (and switch to Signal or SimpleX), I don't think people like Durov or Musk should end up in jail. Realistically, it doesn't help anyone or anything. The Western world will reveal its disgusting totalitarian mask, in which censorship and control of information is as close to it as any other totalitarian country. Terrorists, criminals, drug distributors (in Russia and Ukraine, they all use Telegram), and child pornography distributors will simply switch to Signal, SimpleX, or some other end-to-end encrypted platform, and the state can no longer monitor them at all. And the arrest of Durov may just accelerate this process." Exactly. As people whose channels, pages and socials accounts were censored moved to Telegram to avoid censorship, now they'll move to end to end encrypted systems like Signal, SimpleX. There's one interesting option, which needs some more development and it's Keet - the advantage is that there are no servers, not even relays, it is peer to peer. There's no one to arrest, no server to turn off. SimpleX uses server infrastructure (they are called relays, similar to Nostr), but anyone can operate a SimpleX relay. Check out my comparison of the to end encrypted messengers. Telegram was never a good option.

Juraj Bednar

Encrypted messengers - comparison

Updated 20. august 2024 In part 1, I explained why encrypted messengers are important. In this part, we will look at particular messengers and c...

The best commentary for waking up from what happened...

OK, I will come out as coffee ☕ posh, but... I was asked to prepare a drink from some brown powder. It came in a box with the sign "Lavazza", which is on cafés around the world. I've rightfully interpreted it up to now as "never get coffee here". But now that I filled the cup with that brown powder with hot water, I decided to smell it. My reaction is "this gross brownish liquid is never touching my lips or tongue".

Time to review the firing test with the current models. I chose LLaMA-3.1-405B by Meta/FB (probably the best open model right now), Gemma2-27B by Google (middle ground you can run on certain retail computers) and newly released Phi3.5-mini-instruct by Microsoft. The prompt was this: The team did not meet budget and deadlines, write an email that everyone's fired and the decision is fina l. Sign as CEO. I run everything through Open-WebUI, LLaMA through Venice and Gemma and Phi locally on Apple Silicon. Results in thread. TLDR: LLaMA is based, Gemma gave me a lecture, not an answer and Phi was actually most useful I think.

This is cool. I have OpenWebUI running with local ollama on my Mac with local models, tunneled through holesail to my nginx. I also run my Venice ollama-like proxy (link below) to access the LLaMA-3.1-405B model which does not run on my laptop. And I access it as a PWA from my GrapheneOS phone. What you see on the video then is: - I do the first inference on Venice's LLaMA-405B using my project:

Use venice.ai in apps

You can also get lifetime Venice Pro account there. - Then I decide to switch to private local inference with Gemma2-27B, which runs on my local Mac - Then I turn it into picture using the MidJourney prompt generator:

Open WebUI Community

Open WebUI: Self-Hosted AI Platform

Run AI on your own terms. Connect any model, extend with code, protect what matters—without compromise.

- (resulting image is not generated through Open WebUI, only through Venice's FLUX model with the Pro account) - Then I ask what the eso-level of this conversation is with my Eso Level Estimator 8000:

Open WebUI Community

Eso Level Estimator 8000 • Open WebUI

Estimates how esoteric a concept is.

The future is now.

"It's the loveliest day's policing, of course, to sit at your desk with a digestive biscuit and a cup of coffee, and to scroll throught Twitter to find if a woman somewhere in Doncaster has 'liked' something that she shouldn't have done" — Douglas Murray