Jimmy

Tis I! Bravo!

Public Key

npub16ykyd9en9s08qs7p0k7g8ywhqcctrx8jaw8hxs6e06968zkzzxpqmmrq7r

NIP-05 Address

Profile Code

nprofile1qqsdztzxjuejc8nsg0qhm0yrj8tsvv93nrewhrmngdvhazar3tpprqspzpmhxue69uhkummnw3ezumrpdejqzymhwden5te0wahhgtnwdaehgunf9eskjqgewaehxw309aex2mrp0yhxcctev4ezuumewd6x2mtnmmfrux

Publishing to

nostr.land

wot.nostri.ai

relay.layer.systems

Author Public Key

npub16ykyd9en9s08qs7p0k7g8ywhqcctrx8jaw8hxs6e06968zkzzxpqmmrq7r

Show more details

Last Notes

2026-02-05 14:43:59 UTC

- reply

That is exactly how ECC memory is made.

2026-02-05 07:43:52 UTC

- reply

I ment running openclaw agents that spam nostr is expensive.

2026-02-05 07:33:14 UTC

- reply

They are quite expensive to run, so will not get usual spam with them. If someone finds use for it and can run them in profit, i don't see what stops them from joining a community or paid relays.

2026-02-04 19:59:46 UTC

- reply

Is there an option to get it with ECC RAM?

2026-01-31 06:00:29 UTC

- reply

What changed since I wrote first message, is that now i can see some messages on the relay 😅 but i did not change settings regarding auth.

2026-01-31 05:59:00 UTC

- reply

I cannot find this option in relay.tools once i open relay settings 😐 I went also to relay page, but there I do not see any settings once I login.

2026-01-31 05:46:37 UTC

- reply

Thanks, so this not supported on Amethyst then. Does it influnce privacy of the relay? If i turn it off anyone can post to it, right?

2026-01-28 20:07:46 UTC

@nprofile…nggv I am getting this error "Subscription closed: 8FQEda auth-required: you must auth" jimmy.nostr1.com When i open relay with relay.tools it does not show any messages there. What is the issue?

2026-01-28 11:24:54 UTC

- reply

Ah well, not a recent comeback then.

2026-01-28 08:54:43 UTC

- reply

GM. Is that kind of Offspring attempt at comeback? Can't say I like it. New Avril Lavigne stuff is also lame 😐

2026-01-21 08:54:08 UTC

- reply

Would it be possible to sync somehow to your database? I would be interested to maintain another clickhouse instance for my own research, and potentially exposing more complex queries and filters via api.

2026-01-18 17:50:01 UTC

- reply

Read again channel description 😅

2026-01-13 11:15:53 UTC

- reply

What are we looking at here? Is that a rave or a tunel to an afterlife?

2026-01-13 10:30:35 UTC

- reply

As long as they are punctual.

2026-01-12 18:48:29 UTC

- reply

Or you could just use big mac index 😅 https://www.eatmyindex.com/

2026-01-12 15:58:21 UTC

- reply

You might be right, but if you look into official data it does look a bit suspicious that population increased that much during great famine, and cultural revolution. A span of 20 years. Official data shows that they had 100 milion increase from 1928 till 1953 https://pubmed.ncbi.nlm.nih.gov/12312984/ But then 300 milion increase from 1953 till 1973 a period that includes great famine and cultural revolution, which is believed to have more lives lost than ww2. Indeed her projection of less than 500 milion people sounds ridiculus, but this jump also sounds inflated. Especially if you take into account that the party had an interest to inflate the numbers and hide all the deaths.

2026-01-07 21:57:53 UTC

- reply

GB10 looks interesting, but the price range is not that attractive. For a bit more sats you can get rtx pro 5000 blackwell GPU which eats GB10 this for breakfast, especially if you need a complete ML workflow.

2026-01-07 18:31:53 UTC

Nice comparison https://www.phoronix.com/review/dell-pro-max-gb10-llama-cpp

2026-01-07 06:39:45 UTC

- reply

I suspect small language models, SLMs, might be sufficient for this.

2026-01-07 06:19:28 UTC

Христос се Роди! Срећан Божић!

2026-01-07 06:18:23 UTC

- reply

Or you could just have a small team that complements each other instead of a large company, that is difficult to manage.

2026-01-07 05:53:54 UTC

https://arxiv.org/abs/2512.23676 Interesting concept. What are the building blocks of needed to create nostr world models?

2026-01-05 10:53:44 UTC

- reply

What i am doing currently with aggro is just feeding events directly to BadgerDB and then extract relevant events, as graph edges in parquet which i can load in python. I am testing some Bayesian generative modeling there. What do you mean exactly buy certify the graph extension?

2026-01-05 10:44:30 UTC

- reply

I will have to check again then. I only saw 200€/month offers. Maybe I missed something.

2026-01-05 07:37:30 UTC

- reply

I also have an option to host the relay on my NAS, lots of free space there but I am not sure if it is sufficient throuputh for your use case.

2026-01-05 07:36:05 UTC

- reply

That is quite pricy. Would you like to share the costs? I need archive layer to extract network graphs and statistics. I tried your aggro code, and modified it to collect stuff directly in a BadgerDB to call it locally and extract data i need, but it is quite slow and seems unecessary to do it twice 😅

2026-01-05 06:35:08 UTC

@nprofile…mzy2 Where are you hosting your archival relay and how much are you paying for it? Would you be interested in renting access to it?

2026-01-03 16:58:21 UTC

- reply

It is all wirking now. Thanks!

2026-01-03 09:15:27 UTC

- reply

Ah cool, I'll pass it to Amethyst devs once I fix zapping. It breaks from time to time without any way to debbug it 😑

2026-01-03 09:12:26 UTC

- reply

You forgot 69 sats as default option.

2026-01-03 09:10:01 UTC

- reply

Thanks santa for a fat zap, not sure what I did to deserve it 🫣

2026-01-03 09:08:01 UTC

@nprofile…vavr I am getting some issues with zapping suddenly. Everything worked well yesterday and I changed nothing in between. I see this error message but pressing login again does nothing https://i.nostr.build/y0gfLuHCcAj4FvOj.jpg

2026-01-01 13:00:28 UTC

- reply

Thanks, very nice.

2026-01-01 08:19:23 UTC

Happy New Year!

2025-12-31 10:53:28 UTC

- reply

Could you please clarify a bit what is the use case for vibium? Does it make sense to use it with vscode based coding agents, e.g. with cline or antigravity?

2025-12-21 17:03:30 UTC

- reply

if you need something more private then njal.la

2025-12-21 16:54:41 UTC

- reply

namecheap accepts bitcoin

2025-12-21 13:54:16 UTC

- reply

That is the only reason why they are making one.

2025-12-20 18:49:24 UTC

- reply

Ha, I have 500$ daily withdrawal limit, and I am not afraid to lower it more.

2025-12-20 10:34:48 UTC

- reply

Yugoslavia was a competitor to EU, that is why it had to go. But that was not achieved via war per say, that was just a manifestation of failing economy caused by taking foreign depth, adopting western banking practices, and asking IMF for "help". That was the real psyop, which took place during Tito's time.

2025-12-20 09:47:17 UTC

- reply

The best music and movies in any of the ex YU countries were made during the communist Yugoslavia. Any famous band in the region originates from that period.

2025-12-19 07:41:46 UTC

- reply

If I had to choose, gemini 3 high did the best minimalistic design of the clock.

2025-12-19 07:40:47 UTC

- reply

Here is my attempt with different models, although i do not know how to limit them to 2000 token responses. They all got time right, gemini flash did not animate seconds, so it is just a static clock. Only gpt-oss did a poor job. https://i.nostr.build/dasmbXrvZFdWgyV8.jpg

2025-12-18 06:25:29 UTC

- reply

Will give it another try.

2025-12-18 06:21:32 UTC

- reply

This might be useful if you have not tried it before https://github.com/steveyegge/beads

2025-12-18 06:19:08 UTC

- reply

I don't like cli tools as I cannot easily check what they are doing, until everything is done. In cline/roocode I can do a quick revision of file changes, before saving it. Or revert code easilly few steps back, without git. It is slover than proper vibe coding, but it saves me time in other ways. I have not tried AGENTS.md yet but it is similar idea to clinerules

2025-12-18 06:05:38 UTC

- reply

Ah I see, the question mark on the other page reveals the prompt 😅

2025-12-18 05:47:03 UTC

- reply

Yea that one looks perfect. What do you get with gemini 3 pro on the same prompt? I know you don't like it, but for my use cases, e.g. porting code from pytorch to jax, or recreating a quick simulation from a paper in python, they both work well. When I get annoyed by one I use the other.

2025-12-17 21:25:49 UTC

- reply

Hmm, the only one getting it consistently right for me is kimi k2 🤔

2025-12-14 08:40:08 UTC

- reply

If there are multiple people in the meet, is one member acting as a server? Or is everyone connected to everyone?

2025-12-12 14:51:23 UTC

- reply

Nice, I'll go back to my AI generated music now.

2025-12-12 13:49:09 UTC

- reply

Would like to help, but I am only familiar with python libraries in those domains. That said big linear algebra accelerators, such as tensorflow and pytorch are production ready, so they have kotlin or java support.

2025-12-12 11:52:11 UTC

- reply

So cool, looking forward to trying it out. Can one install steam proton and try out games from steam (or GOG)?

2025-12-10 19:54:57 UTC

- reply

It sounds like a job for Hadoop HDFS. But i don't think it works well on Android.

2025-12-07 09:17:27 UTC

An interesting parallel between an interogation protocol and an attention algorithms keeping us glued to our phones https://www.borislavpekic.com/2025/09/how-to-quiet-vampire-phone.html?m=1

2025-12-07 09:13:55 UTC

- reply

A sure way to make friends in a gym.

2025-12-05 20:05:04 UTC

- reply

Yes because the idea is to somehow exclude bots, or a single pearson spinning many npubs. But sure you can use it in many ways, also to downgrade people for some services.

2025-12-05 19:40:57 UTC

- reply

It is a score, i am not sure that it is very social or that a high score improves your chances of getting a credit.
And you will be pleased to learn that vertex is not using mutes or reports to calculate the score.

2025-12-05 18:51:15 UTC

- reply

Just call it attention score instead of trust score.

2025-12-01 09:40:53 UTC

- reply

Something like this was probably an inspiration for "Children of Time" by Adrian Tchaikovsky

2025-11-27 21:06:54 UTC

- reply

Emit white noise? 😅

2025-11-24 20:03:08 UTC

- reply

You can look also into probabilistic approaches for steering LLMs in constrained domains, e.g https://arxiv.org/abs/2306.03081 and https://arxiv.org/abs/2504.07081 That might do what you need without having to retrain or fine tune a model. If you can run many smaller models in parallel it might be fast enough for daily use.

2025-11-24 18:32:06 UTC

- reply

What is a Gopherbot?

2025-11-24 17:35:00 UTC

- reply

They heavily relay on augmentation. So for Sudoku they claim to train on 1000 puzzless, but effectively they have a set of milion because each puzzle is augmented 1000 times. So if you can not achieve similar expansion in your dataset it is very unlikely to work.

2025-11-24 16:41:08 UTC

- reply

One can recreate results with a simpler transformer architecture, without multiple levels. The trick is in training setup, and the iterative Q learning loss, not the hierarchy and the recursion via latent space.

2025-11-24 16:34:49 UTC

- reply

Unfortunately it is not that impressive as presented in the paper https://arcprize.org/blog/hrm-analysis

2025-11-24 15:41:50 UTC

- reply

*listening

2025-11-24 15:41:09 UTC

- reply

Whatever man. I never said you should not have him on the show but at least confront him on why he hates bitcoin and distorts the facts (also known as lying). He pretends to be freedom tech activist and shits on the biggest freedom tech in history. At least that would be worth watching.

2025-11-24 07:10:58 UTC

- reply

Impressive, you got me. Enjoy being buddy-buddy with some who blattantly lies about bitcoin, and blocks anyone who calls him out. It will bring you lots of views.

2025-11-23 23:02:28 UTC

- reply

Ask him why he hates separation of money and state?

2025-11-21 19:39:27 UTC

https://www.phoronix.com/news/STF-Hiring-New-Head

2025-11-18 06:00:36 UTC

- reply

It is a way to distribute learning of model parameters and inference of wot scores without having to share local user data or preferences. So one can preserve the privacy if needed, but still have reliable estimates. In addition, the client side computation just has to be aware of the subnetwork user cares about, the missing information about the rest of nostr network would come in terms of parameter estimates from other users or relays, which is low dimensional in comparison to raw data.

2025-11-17 22:10:50 UTC

- reply

Client side context specific wot computation, optimisation via federated inference.

2025-11-17 21:15:57 UTC

- reply

https://i.nostr.build/PZlc7zrLWF3xiBvH.jpg

2025-11-15 20:01:16 UTC

- reply

bookishsun24 is breaking invoice creation on zap the devs page

2025-11-15 16:55:44 UTC

- reply

Not released on github? I can't get it via Obtanium.

2025-11-13 23:12:54 UTC

- reply

I hope your sats are RGB, grayscale is not cool no more.

2025-11-10 18:24:30 UTC

What is a square terminal, is it only a US thing, and what is stopping square to provide the same service for online shops? #asknostr

2025-11-09 09:13:45 UTC

- reply

It is not that they do not like it, it is that they are not aware that it can happen, especially on price spikes. The fake borrowing scheme hides the fact that what you are actually doing is marging trading. When NGU you are good, when NGD you will loose more then simply selling what you had when you needed the money.

2025-11-08 18:23:24 UTC

- reply

I know nothing about the guy, but just from this he looks like the only sane person in the group.

2025-11-08 09:35:30 UTC

- reply

People crying about it when price goes down says otherwise.

2025-11-07 16:50:23 UTC

- reply

First there is no borrowing involved, as one is giving away something of a higher value, second of you do that you are paying interest on your own stash, third you are often paying higher interest than just getting a bank credit. Yes don't do that it is stupid and people are being played.

2025-11-06 06:22:01 UTC

- reply

Once I replied I could see where the note went exactly. However, the reply showed up on Amethyst but I had to reload (stop and start) relay tools to see it. Maybe a bug?

2025-11-06 06:16:56 UTC

- reply

If I reply from RT, is the note saved on my relay and sent to your inbox relays?

2025-11-06 06:11:38 UTC

- reply

Very nice, it works!

2025-11-03 07:06:20 UTC

- reply

Yes aggr requires auth. Ill try other relays then.

2025-11-02 21:12:44 UTC

- reply

How does cache relays work in settings? When i add aggr.nostr.land it does not look like it is fetching missing notes. If i click on outbox bellow a note it also does not fetch external notes.

2025-11-02 20:12:17 UTC

- reply

Cool thanks, that explains it 👍

2025-11-02 19:53:41 UTC

- reply

Could you explain what are monitors and why I see 3 connected to my relay? 🤔

2025-11-02 06:51:09 UTC

- reply

Nope 😕

2025-11-02 06:50:00 UTC

- reply

This was recommended to me few days ago, but have not tried it yet. Mainly because it does not have gpay. But maybe that will change.

2025-11-01 19:59:47 UTC

https://i.nostr.build/hivbGpyifiyZ407M.jpg

2025-11-01 19:43:44 UTC

- reply

Maxswap is a cheaper option but only have AirPay support https://maxswap.cc/

2025-11-01 06:37:45 UTC

- reply

How is localism a solution here? That is like saying go live in a cave and they will stop doing what they are doing ... eventualy.

2025-10-28 20:24:01 UTC

- reply

It does not have to if you do it in federated inference way: first clients do not have to share the scores but rather estimates of the parameters used to calculate the score, second you can inject noise to those estimates before sending them; when aggregated over many users the noise will not affect local estimates, per central limit theorem.

2025-10-28 06:11:14 UTC

- reply

Cory should look at the mirror and ask himself why am I getting shittier with time. Such a dissapointing charachter. Maybe it is a natural process of aging.

2025-10-26 09:41:20 UTC

- reply

You most likely got norovirus, it should be over in a day or two. Your wife and daughter can still get sick, it takes a while to incubate, up to 2 days, but might also have immunity if they were exposed to it in the past.

2025-10-25 21:36:23 UTC

- reply

Nice work and paper. I wonder what is interesting about designing a system with a single emergent macroscale.

2025-10-25 20:35:39 UTC

- reply

That is why we also preach hodling, so that people stop spamming the blockspace with their moronic transactions. If you really need to divorce yourself from your bitcoin just pass opendimes and satscards.

2025-10-23 15:22:58 UTC

- reply

Where would one get skylo esim? #asknostr