Laan Tungir

I was thinking that a fun project would be to create a nostr account that regularly published the btc price or block height "encrypted" as maybe poetry, or recipes, or something along those lines.

We both decided to implement! I had the llm do a python script first, and then wanted something to run on a html page that was easy for people to see as a demo. I just picked a model to run the test. Didn't put too much thought into it.

nostr:npub1vadcfln4ugt2h9ruwsuwu5vu5am4xaka7pw6m7axy79aqyhp6u5q9knuu7 inspired me from his post about using LLM's for encryption. Here is an example web page that will download GPT-2, and then encrypt a secret message, and decrypt it. The encrypted message looks like regular text, but hidden in the text is a secret message. Clearnet: https://laantungir.net/client/llm-steganography.html Tor: http://kn2jam4kyz6s5wacyozo3do3d2zsvh4to45uh7xliulwdust6zjuvnad.onion/client/llm-steganography.html FIPS: http://npub1crpldvy49ef8z34wlacwujnfudy4nd7k96aqdx5wgn6ckztz7z8q9t59ud.fips/client/llm-steganography.html nostr:nevent1qvzqqqqqqypzqe6msnl8tcsk4w28capcaegeefmh2dmdmuza4ha6vfut6qfwr4egqqstxmgsd0egtq57gj8mghckthsv79dhegh7xh97hwtwve6l56g4s9sh48wkc

It's different from what I was thinking. Here is a Chinese LLM generated example for irony. ## A Concrete, Step-by-Step Example To make this tangible, let's use a **toy model with a small vocabulary of 11 words** so you can see every number. The real scheme works identically, just with vocabularies of ~50,000 tokens and floating-point probabilities. ### Setup **The model:** A tiny "LLM" with a fixed vocabulary of 11 words: `{apple, date, banana, cherry, pie, juice, tart, sauce, for, with, and}` At each step, the model assigns a probability to **all 11 words** based on the context. The probabilities change as the context grows, and they always sum to 1.0. (A real LLM does the same thing, just with ~50,000 tokens instead of 11.) **The shared context/prompt:** `"I like to eat"` — both Alice and Bob have this. It is not secret, just shared. **The shared secret key:** Used to seed a pseudorandom number generator (PRNG). The key produces the random stream: `0.15, 0.62, 0.40, ...` (uniform random numbers between 0 and 1). **The secret message Alice wants to send:** The bits `0 1 1` (3 bits — in reality this would be a longer message, but let's keep it tiny). --- ### Encoding (Alice's Side) #### Step 1: Alice runs the model Alice feeds the shared context `"I like to eat"` into the model. The model outputs a probability distribution over all 11 words. After "I like to eat", fruits are the most natural continuation, so they get the highest probabilities: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.20 | [0.00, 0.20) | | date | 0.15 | [0.20, 0.35) | | banana | 0.10 | [0.35, 0.45) | | cherry | 0.05 | [0.45, 0.50) | | pie | 0.20 | [0.50, 0.70) | | juice | 0.12 | [0.70, 0.82) | | tart | 0.08 | [0.82, 0.90) | | sauce | 0.05 | [0.90, 0.95) | | for | 0.03 | [0.95, 0.98) | | with | 0.015 | [0.98, 0.995) | | and | 0.005 | [0.995, 1.00) | This is just normal LLM behavior — the model is predicting what word comes next. Notice the probabilities sum to 1.0, and the intervals partition [0, 1). #### Step 2: Alice encodes her first secret bit Alice's first secret bit is `0`. She interprets this as: "look in the **first half** of the range [0, 1), i.e., [0.00, 0.50)." The words whose intervals fall within [0.00, 0.50) are: - apple: [0.00, 0.20) - date: [0.20, 0.35) - banana: [0.35, 0.45) - cherry: [0.45, 0.50) Alice uses her next random number from the shared key, `0.15`, to pick within [0.00, 0.50). The value `0.15` falls in [0.00, 0.20), which is **apple**. **Alice outputs: "apple"** Text so far: `"I like to eat apple"` #### Step 3: Alice runs the model again Now the context is `"I like to eat apple"`. The model outputs a new distribution over all 11 words. After "apple", food preparations like pie and tart become more likely, while the other fruits become less likely: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.15 | [0.00, 0.15) | | date | 0.10 | [0.15, 0.25) | | banana | 0.08 | [0.25, 0.33) | | cherry | 0.07 | [0.33, 0.40) | | for | 0.06 | [0.40, 0.46) | | with | 0.04 | [0.46, 0.50) | | pie | 0.20 | [0.50, 0.70) | | tart | 0.16 | [0.70, 0.86) | | juice | 0.08 | [0.86, 0.94) | | sauce | 0.04 | [0.94, 0.98) | | and | 0.02 | [0.98, 1.00) | Alice's next secret bit is `1`. This means "look in the **second half** [0.50, 1.00)." The words whose intervals fall within [0.50, 1.00) are: - pie: [0.50, 0.70) - tart: [0.70, 0.86) - juice: [0.86, 0.94) - sauce: [0.94, 0.98) - and: [0.98, 1.00) Alice uses her next random number, `0.62`, to pick within [0.50, 1.00). She scales it into that range: `0.50 + 0.62 × 0.50 = 0.81`. The value `0.81` falls in [0.70, 0.86), which is **tart**. **Alice outputs: "tart"** Text so far: `"I like to eat apple tart"` #### Step 4: Alice runs the model again Context: `"I like to eat apple tart"`. Model output — after "apple tart", connectors and prepositions like "with" and "and" become the most likely continuations, while the fruits and preparations drop in probability: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.10 | [0.00, 0.10) | | date | 0.08 | [0.10, 0.18) | | banana | 0.07 | [0.18, 0.25) | | cherry | 0.06 | [0.25, 0.31) | | pie | 0.08 | [0.31, 0.39) | | juice | 0.06 | [0.39, 0.45) | | for | 0.05 | [0.45, 0.50) | | tart | 0.06 | [0.50, 0.56) | | sauce | 0.04 | [0.56, 0.60) | | with | 0.18 | [0.60, 0.78) | | and | 0.22 | [0.78, 1.00) | Alice's next secret bit is `1`. Look in the second half [0.50, 1.00): The words whose intervals fall within [0.50, 1.00) are: - tart: [0.50, 0.56) - sauce: [0.56, 0.60) - with: [0.60, 0.78) - and: [0.78, 1.00) Random number `0.40` → scaled: `0.50 + 0.40 × 0.50 = 0.70`. Falls in [0.60, 0.78), which is **with**. **Alice outputs: "with"** Text so far: `"I like to eat apple tart with"` #### Alice is done Alice sends Bob the text: **"I like to eat apple tart with"** (Plus whatever padding/continuation she wants to make it look like a complete sentence.) To any observer, this looks like someone generated a sentence about food. Nothing suspicious. --- ### Decoding (Bob's Side) Bob has: - The received text: `"I like to eat apple tart with"` - The same model - The same shared context: `"I like to eat"` - The same shared key (so the same random stream: `0.15, 0.62, 0.40, ...`) #### Step D1: Bob runs the model Bob feeds `"I like to eat"` into the model. He gets the **same** distribution Alice got: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.20 | [0.00, 0.20) | | date | 0.15 | [0.20, 0.35) | | banana | 0.10 | [0.35, 0.45) | | cherry | 0.05 | [0.45, 0.50) | | pie | 0.20 | [0.50, 0.70) | | juice | 0.12 | [0.70, 0.82) | | tart | 0.08 | [0.82, 0.90) | | sauce | 0.05 | [0.90, 0.95) | | for | 0.03 | [0.95, 0.98) | | with | 0.015 | [0.98, 0.995) | | and | 0.005 | [0.995, 1.00) | Bob sees that Alice chose **apple**, which is in the interval [0.00, 0.20). Now Bob asks: "Which half was apple in?" Apple is in [0.00, 0.20), which is in the **first half** [0.00, 0.50). **First half → bit `0`** **Bob recovers bit: `0`** ✓ #### Step D2: Bob runs the model again Bob appends "apple" to context: `"I like to eat apple"`. Model output: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.15 | [0.00, 0.15) | | date | 0.10 | [0.15, 0.25) | | banana | 0.08 | [0.25, 0.33) | | cherry | 0.07 | [0.33, 0.40) | | for | 0.06 | [0.40, 0.46) | | with | 0.04 | [0.46, 0.50) | | pie | 0.20 | [0.50, 0.70) | | tart | 0.16 | [0.70, 0.86) | | juice | 0.08 | [0.86, 0.94) | | sauce | 0.04 | [0.94, 0.98) | | and | 0.02 | [0.98, 1.00) | Bob sees Alice chose **tart**, interval [0.70, 0.86). "Which half was tart in?" [0.70, 0.86) is in the **second half** [0.50, 1.00). **Second half → bit `1`** **Bob recovers bit: `1`** ✓ #### Step D3: Bob runs the model again Context: `"I like to eat apple tart"`. Model output: | Word | Probability | Interval | |--------|-------------|---------------| | apple | 0.10 | [0.00, 0.10) | | date | 0.08 | [0.10, 0.18) | | banana | 0.07 | [0.18, 0.25) | | cherry | 0.06 | [0.25, 0.31) | | pie | 0.08 | [0.31, 0.39) | | juice | 0.06 | [0.39, 0.45) | | for | 0.05 | [0.45, 0.50) | | tart | 0.06 | [0.50, 0.56) | | sauce | 0.04 | [0.56, 0.60) | | with | 0.18 | [0.60, 0.78) | | and | 0.22 | [0.78, 1.00) | Bob sees Alice chose **with**, interval [0.60, 0.78). "Which half?" [0.60, 0.78) is in the **second half** [0.50, 1.00). **Second half → bit `1`** **Bob recovers bit: `1`** ✓ --- ### Result Bob has recovered: `0 1 1` — exactly the message Alice sent. --- ## Diagram ``` ┌──────────────────────────────────────────────────────────────────┐ │ ALICE (ENCODER) │ │ │ │ Secret bits: 0, 1, 1 │ │ Shared context: "I like to eat" │ │ Shared key → random stream: 0.15, 0.62, 0.40 │ │ │ │ For each bit: │ │ 1. Run model with current context → distribution │ │ 2. Use bit to pick which HALF of [0,1) to look in │ │ 3. Use random# to pick a word within that half │ │ 4. Append word to context, repeat │ │ │ │ Output: "I like to eat apple tart with" │ └──────────────────────────────────────────────────────────────────┘ │ │ (sent over channel — looks innocent) ▼ ┌──────────────────────────────────────────────────────────────────┐ │ BOB (DECODER) │ │ │ │ Received text: "I like to eat apple tart with" │ │ Shared context: "I like to eat" (SAME as Alice) │ │ Shared key → random stream: 0.15, 0.62, 0.40 (SAME as Alice) │ │ │ │ For each word: │ │ 1. Run model with current context → SAME distribution │ │ 2. Find which interval the word falls in │ │ 3. Determine which HALF that interval is in │ │ 4. That half = the recovered bit │ │ 5. Append word to context, repeat │ │ │ │ Recovered bits: 0, 1, 1 ✓ │ └──────────────────────────────────────────────────────────────────┘ ``` --- ## Key Points 1. **The LLM never saw the secret message.** The secret bits controlled *which word was picked* from the distribution. The LLM just produced distributions. 2. **The output looks natural.** "I like to eat apple tart with" is a perfectly normal sentence. A censor seeing this has no reason to be suspicious. 3. **Both sides run the same model with the same context.** This is why they get the same distributions. If Bob had a different model or different context, the distributions would differ and decoding would fail. 4. **The shared key provides the random stream.** This is needed for the encoding to be secure — without it, the censor could run the same model and try to decode. With the key, the censor can't reproduce the random choices. 5. **Each token encodes roughly 1 bit** in this toy example (because we split into 2 halves). In reality, with arithmetic coding over a 50,000-word vocabulary, you can encode **multiple bits per token** — roughly 2–4 bits per token for a real LLM, depending on how uncertain the model is at each step. 6. **The "half" splitting is the simplified version.** The real scheme (from the Meteor paper, eprint.iacr.org/2021/686) uses full arithmetic coding, which is more efficient — it doesn't waste the probability mass the way a 2-way split does. But the principle is the same: the secret bits steer the selection within the model's probability distribution.

Hmmmm, interesting. So you would have to share prompts in common as well? Maybe an encryption prompt and a decryption prompt? So if I were to send you a secret text, I would feed it into the LLM with the encryption prompt, and then I could also check it by feeding the cypher-text back in with the decryption prompt to make sure it comes out reasonably close to what I intend.

I'm trying to understand why you need the LLM. Let's say that an LLM is around 500GB. If Alice and Bob both have access to shared 500GB, why does the fact that the 500GB is an LLM give them something special? For example, in 500GB, we could simply have a vast lookup table that converts innocuous text to secret text. GM = Meet you at the cafe at 5PM. It's sunny here = We have been discovered. .... What am I missing here?

https://blossom.laantungir.net/9dea316fc41d4868e680c81c56ea975ad97d069701a3ca055b5aefea813baa76.jpg nostr:naddr1qvzqqqrkvupzq8ky23e5mjlkle2fq88ztsx8c672tmwcj3p5zempltwry8fcmufeqq9xgatz946x2cmgdehsd4hllk

Here is the current breakdown of my relay by event type. I don't know if my relay is representative, but gift wraps are by far the most popular event type that I am hosting.  It is possible that NIP-17 as it is currently running is the most popular event type on nostr that my relay sees.  You could argue that this means it is ripe for optimization. On the other hand, you could ask why mess with what is obviously working. https://blossom.laantungir.net/9b45301b89ab7fd22074771187058a1ea5701eeaf21b52737935d68d6caaae0d.png

My vote is to always push the boundaries of what is possible with privacy and censorship resistance, and to stay inspirational in that regard.  That is our niche.   That is our thing.  Nobody else cares as much about censorship resistance as we do, but there are lots of people who care about making reasonable engineering trade-offs in software, to gain users. I want to make the best tools. Not the tools for the masses.

I added mirrors in my Gitea for Minibits, Amethyst and vnak. So these projects are accessible directly via TOR and FIPS: [tor] http://kn2jam4kyz6s5wacyozo3do3d2zsvh4to45uh7xliulwdust6zjuvnad.onion/ [fips] http://npub1crpldvy49ef8z34wlacwujnfudy4nd7k96aqdx5wgn6ckztz7z8q9t59ud.fips/

OMG, So I haven't watched the full thing, but so many bad takes from this clip and post. First, you can't prove anything in science. So yeah, you can't "prove we aren't in a simulation", but you can't prove anything else either. That's not how science works. We don't prove things. How could we? Could we go to God and ask him "Hey God, it this 100% true?" There is no oracle, God, or process we can follow that will give us absolute truth. If we had that thing, then science would be over and done with. We would just consult that oracle on everything, it would tell us "true or false" for everything we wanted to know, and we would wrap it up for a day. Second, we don't ever "experience base reality", because we are always experiencing a simulation created by our brain. Am I experiencing this keyboard in front of me via "base reality"? No. Light bounces off the keyboard, enters my eyeballs, gets converted to electrical impulses, gets sent to my brain, and from those impulses my brain creates a simulation of the keyboard. Same with the my fingers on the keyboard. My nerve endings send electrical impulses to my brain which likewise get fed into this amazingly good simulation of the universe that I get to experience every day in my brain. It is so funny how all these promoters of "The universe is a giant simulation" or "We are living in alien supercomputers" never seem to think about the obvious fact that we ARE in a simulation that is running in our brain.  It is just so obvious that it is ignored. So strange.

Hey nostr:npub1dd9znw7585wsam4d8p84ztdmtywwjsrayld6fzk4fvqdn5hpju4st5xe7p Can you explain why you would accept bitcoin lightning for top-ups to an account, but your initial payment has to be with a credit card using Stripe? This seems a bit ... un-privacy preserving.

Hmmmm, GLM says yes it can: Amethyst CAN send nutzaps from its NIP-60 wallet — but the recipient must be set up to receive them Amethyst's peekNutzapFunding function in CashuWalletState.kt checks 4 conditions before offering the Cashu nutzap rail. If ANY fail, it falls back to asking for an external Lightning wallet: The 4 conditions (all must be true) # Condition Who is responsible 1 Sender has a Cashu wallet with mints You (the sender) — ✅ this is now fixed 2 Recipient has a kind 10019 (NutzapInfoEvent) The person you're zapping 3 Recipient's 10019 has a pubkey tag (p2pk) The person you're zapping 4 Sender and recipient share at least one mint Both parties The most likely cause Since your sender wallet is set up correctly (we just fixed the 17375), the problem is on the recipient side. The person you're trying to nutzap either: Hasn't published a kind 10019 nutzap mint list, OR Their 10019 doesn't have a pubkey tag with their p2pk pubkey, OR Their 10019 lists mints that don't overlap with yours

nostr:nprofile1qqsyvrp9u6p0mfur9dfdru3d853tx9mdjuhkphxuxgfwmryja7zsvhqpzamhxue69uhhv6t5daezumn0wd68yvfwvdhk6tcpzpmhxue69uhkummnw3ezumt0d5hszythwden5te0dehhxarj9emkjmn99uwfvhl6 I have sats in a users cashu wallet in Amethyst. Shouldn't I be able to nut-zap from there? Amethyst is asking to open an external wallet.

I have been mainly using only OpenAI and Anthropic for coding. I have tried other open source models but they all seemed to fail spectacularly. I heard about GLM 5.2 and have been giving it a go today, and so far, I am very pleased. I'm sure it isn't up to the frontier level yet, but so far this morning I am finding it completely usable for javascript coding. I have held off from looking into running local models until I have seen something that works, and this seems to work well so far. My hopes of getting sovereign as far as LLM models are concerned has increased.

Instead of tackling the important items on my todo list, I made my first font. Ask me anything. https://blossom.laantungir.net/8089f1bfd705335ba17b7da073f085321a0c212c477d9c7e0f63561b89b7835d.png https://blossom.laantungir.net/48dc3333aa73e9e47c86c7d672318bc0bc7a12e46952b1465724d9bf8dd9bba4.png

GM. If you cut off your thumb, and replaced it with a mechanical one, would you still be you? Yes, of course. https://blossom.laantungir.net/24235ed85d421e61ea9a834f5f5c01cd639f60f3afe090399328e1ad0665f249.jpg Replace your whole hand, it's still you. Replace your body, it's still you. Replace your skull, it's still you. Now here is the good one: Replace your brain... still you? Yes, it's still you. Replace every atom, one by one, and it is still you. Why? Because you are the software, not the hardware. Turing-complete computers, which include our brains are commodities. They can all run any software in our universe. They can run your software. They can run me. They can run you. On #NOSTR we aren't just creating a social network, we taking baby steps toward creating our future bodies. Our future homes. Suppose the computer you are running on (your brain) is malfunctioning or dying, and you could be uploaded to "the cloud". What would you want that "cloud" to look like? Would you want an off switch controlled by others? By the government? Would you want to be uploaded to one computer or decentralized across many? Would you want to be in a robot? Be in control of many robots? Stay here on earth or be shot off into space away from everyone else and potential danger? Would you want to be a slave or be free ... and what exactly does that even mean? How could you ensure communication with others?

Here is something fun to do today on NOSTR.  Start an anonymous account.  https://blossom.laantungir.net/930853bed8afe98a9b6228079458f44389d2c72a0e736026d1a30acc7670aaf1.jpg Think about privacy. Use TOR or a VPN, but don't worry about it too much. Don't overthink it. It is more important to start than to get too hung up on all the details. Just start one.  Maybe post your meals. Maybe post art. The more specific the better. Enjoy the fact you don't have any followers. There is no pressure. You are free to post whatever you want.  If you really want to have fun, try designing the person you would like to become. You can just become someone.

Hashes changed which probably messes up whatever it is you're doing, but: https://blossom.laantungir.net/e4f216aeb1eec5e0af56085eabe0c71b5966358c8d8b14699a7832ce61a58a5b.jpg https://blossom.laantungir.net/957af05d110fd12f0b1fb37af8cfabb422e316f6fb6af6fa73c1611cc2448f24.jpg

The Mrs likes buying and trying out esoteric fruit, and then subjecting me to them. In my opinion, there is a reason why the esoteric fruits such as jackfruit and dragon-fruit, remain esoteric. They just don't taste as good.  If there was a really delicious unpopular fruit, it would become popular. nostr:naddr1qvzqqqrkvupzq8ky23e5mjlkle2fq88ztsx8c672tmwcj3p5zempltwry8fcmufeqqr8yet8vask2a68qwj

GM. Live Reggae this morning. https://blossom.laantungir.net/e1eac8df0a208760b8aeb61961e1c4dcd4c88538651a64d2c716e916fd5570d1.webp nostr:naddr1qvzqqqrkvupzq8ky23e5mjlkle2fq88ztsx8c672tmwcj3p5zempltwry8fcmufeqqr8yet8vask2a68qwj

To be clear, I am not making a metaphor about the brain being a computer. I'm stating that the brain IS a computer. It has memory where it can store information over time, and it can run computations on that stored information. The brain is a computer. Not an analogy. And no, understanding that a brain is a computer does not imply that all thoughts and emotions (computer code) can therefore be understood at the atomic or chemical level, or be derived from physics. The rules of code running inside computers is emergent. It doesn't depend on what the computer running it is made up of. Just like the next best move in a game of chess don't depend on whether the chess pieces are made from glass or brass.

Slight critique that may or may-not apply depending on what exactly you mean by "Morality can be derived from physics." Ethics occurs in computers. It's obvious that rocks for example can't think about ethics, or anything else. Only computers [like our brain for example] can think about or have ethics. When a computer becomes Turing-complete, it means that it can now calculate ANY calculation that is allowed in our universe, given memory and speed constraints. That minimum limit of capabilities - becoming Turing-complete - is also the ceiling of capabilities. A Turing-complete computer can calculate EVERY CALCULATION allowed by the laws of physics, given memory and processor speed constraints. Why do I bring up Turing-completeness as a critique? Because what it means is that all the software in the universe, will run the same on every Turing-complete computer, whether you make a Turing-complete computer from neurons, transistors, or marbles rolling down tracks. That means that the physical properties of the computer fall away, and all you are left with are the emergent properties of what can happen inside computers. What can happen with code. If this is hard to grasp, imagine the analogy of chess. Fool's mate is the fastest known checkmate in chess. Does fool's mate change if you are playing with chess pieces made from plastic instead of metal? Obviously not. Once you have the rules of chess, it doesn't matter the physics of the board, the atoms bouncing around in the pieces, or anything else on the physics level. Thermodynamics has nothing to tell us about fools-mate. Likewise morality can not be derived from physics, anymore than the best moves in chess can be derived from physics. This point I am making does not mean that physics and morality do not interact in any way. It's not so simple. Is it moral to throw a pillow at you? Fire a bullet? These are moral questions that depend on the physics of pillows and guns. But we are talking about code here that emerges within computers, and the rules of code within computers does not necessarily have anything to do with physics. Code has it's own emergent properties: like prime numbers, beauty, morality, the number pi, etc. I was going to go on with 2 or 3, but this is quite a bit and I'll leave it there for now. Looking forward to what you come up with.

Why I think AGI will deserve ethical consideration. nostr:nevent1qvzqqqqqqypzq8ky23e5mjlkle2fq88ztsx8c672tmwcj3p5zempltwry8fcmufeqyd8wumn8ghj7um9dejxjapwdehhxenvv9ex2tnrdakj7qg4waehxw309aex2mrp0yhxgctdw4eju6t09uqzpa2yhjvnwgv0fwdw7ac8eptkllvmdsx3wt3nj09ulpjffwmznt0l84uslk

So what is this constraint that you believe in that keeps a silicon based computer dumb? Is it the dry silicon instead of our wet carbon? Is it a god given soul? What is it?