someone

i would make 'reply to' an icon and make the gray area two lines
i think NIP-5 is done wrong here. too much attention. it should be shown pale gray. client has to cache those and alert the user with yellow bold or orange bold if the cached user@domain for the username mismatches the viewed username.
another problem is OP serif, repliers sans serif..

2026-07-15 16:53:12 CEST

there was an attacker who kept my relays super busy, to make it unusable for others. i guess he is bored now. relays are responsive.
i didn't deal with it because i counted other relays to carry nostr.
but yeah strfry needs a read plugin or maybe a log analyzer vibed to deal with these types of scenarios.

2026-07-15 14:16:35 CEST

- reply

is that a green prayer mat?
i like the cup with the tie and shirt

2026-07-14 17:33:48 CEST

- reply

looks good. do you have a community?

2026-07-14 00:34:45 CEST

- reply

oof
i think my relays are getting some of the action too
write policy is hodling well but time to write plugins for query (read) side as well

2026-07-12 17:16:02 CEST

- reply

welcome

2026-07-12 17:14:57 CEST

what is the best group chat app on nostr?
- has to deliver messages reliably
- "seen" marks nice to have
- should run on androids and chromebook (or web if chromebook not available)
#askNostr

2026-07-09 20:00:02 CEST

- reply

Does bittorrent over FIPS make sense?

2026-07-06 22:28:23 CEST

llama.garden announcement and news
https://www.reddit.com/r/LocalLLM/comments/1up92t0/waifumagnet_torrents_that_work/

2026-07-04 18:45:52 CEST

https://image.nostr.build/3b80feafe20af44377c5948bd67d995996d0b6be5e892ff84ae3b051cf558be5.png

2026-07-03 01:56:51 CEST

- reply

yes i have a marketing problem :)
i monitored nsfw blocking for a while seriously and sacrificed myself :)
spams i dont monitor much. but thanks for letting me know.

2026-07-02 20:14:59 CEST

- reply

spammers dont use one pubkey, so it is better to look at IPs. it is harder for him to switch IP than pubkey.. most of rejections on my relays are based on poorly performing IP.
i have two threads for both IP and pubkey, that checks notes using LLMs on a GPU.
planning to open source some parts when i have the time..

2026-07-01 16:20:47 CEST

- reply

coffee i can take. uranium no

2026-06-30 17:14:30 CEST

- reply

Open weights that depend on a single host are not really open
With BitTorrent, bandwidth pooling happens by design. Every downloader is a potential uploader, so popularity grows capacity instead of straining it. A viral 70B release that 5000 people want simultaneously will melt a single HTTP origin; on a swarm, those 5000 peers are the swarm.
The genuine weakness of torrent distribution is seeder attrition — old swarms die. This project handles it on two fronts. First, webseeds: every .torrent has HTTP webseed URLs embedded, so a BEP-19-aware client (Transmission is the reference) can fetch the full payload over HTTP, piece-verified against the torrent's hashes, even with zero peers. For HuggingFace-mirrored models the webseeds point at huggingface.co/<org>/<repo>/resolve/, which means every webseed download is a continuous cryptographic audit of HF's served bytes against the curator's snapshot — if HF silently swaps a shard or a CDN edge caches a truncated file, the piece hash fails. The signed `hf_match` tag asserts whether the torrent is a byte-exact mirror of the HF repo, and the webseed mechanism lets anyone independently verify that without trusting HF, the curator, or the relay.
Second, on-protocol demand signaling. Users signal dead swarms with a proof-of-work-backed Nostr event (kind 30103). The PoW makes spam expensive, so the signal stays meaningful. Curators watching the relay can see which models people actually want and re-seed them — turning "nobody knows this is unreachable" into "twelve people asked this week."
What Nostr adds on top
BitTorrent gives you transport, not discovery. Historically, torrent catalogs lived on websites — seizurable, suable, corruptible. This project uses Nostr as the catalog. Listings (kind 30099) are signed events carrying the .torrent URL, metadata, and source reference, propagating across nine relays run by different operators. If one disappears, the other eight still serve every listing. If an operator dislikes a model, they can delist it on their relay — but not on anyone else's. Censorship becomes a local opt-out, not a global deletion.
Catalog and transport are decoupled: the Nostr event tells you a torrent exists and where the .torrent is; the .torrent tells your client how to join the swarm; the swarm moves the bytes. No single component can be shut down — you'd need to shut down every relay, every .torrent host, and every seeder simultaneously.
Failproof distribution, by construction
- Discovery is redundant across nine independently-operated Nostr relays. Listings are signed by the whitelisted curator npub and verified client-side, so a malicious relay cannot forge or alter one.
- The .torrent file is a small, cacheable blob hostable anywhere. The client checks its info-hash against the signed Nostr event before passing it to the BT client, so a compromised host cannot swap in a tampered torrent.
- The weights live on the swarm. One laptop on a residential connection in the right timezone is enough to keep a 13B model alive for the next person who needs it.
An open, forkable catalog
The listings are not the private property of this HTML page. Every kind 30099 event lives on the Nostr network, signed and queryable by anyone. A developer who wants a different UI, disagrees with the curator's taste, or needs listings filtered for a specific use case can ship their own client — web page, CLI, desktop app — and pull the exact same signed events over the exact same relays, no permission or API key required. The catalog is a public, append-only, cryptographically-signed event stream. Clients are views onto that stream, not privileged gatekeepers. Anyone can fork this HTML, modify it, and publish the result; the network effect accrues to the protocol, not to any single client.
Standalone and self-upgrading
The HTML client requires no website, backend, or build step. Saved to disk, opened from file://, mirrored on any static host — it behaves identically, because all work happens client-side. There is no deploy to take down, no domain to seize, no server bill to default on. Distribution of the client is itself peer-to-peer.
The client is upgradable by its maintainer, but the upgrade path is tamper-evident. The running HTML listens to Nostr for kind 30100 events signed by the curator npub. When an event announces a higher version, the client fetches the candidate HTML from the Blossom URLs in the event, computes its SHA-256, and compares against the signed sha256 tag. Only if the hash matches — proving the bytes the maintainer signed are the bytes the host served — does it surface the new version as a verified update, offered as a local blob URL. A compromised host cannot push malicious HTML: the signed hash authorizes the bytes, not the host's say-so. Upgrades are opt-in and reversible — the user retains the version they are running until they choose otherwise. Combined with forkability, the maintainer's authority is strictly over what gets offered as an upgrade, not over what users must run.

2026-06-30 08:16:17 CEST

https://llama.garden/
decentralized llm sharing over nostr and bittorrent has a website now

2026-06-25 03:11:50 CEST

- reply

waifu magnet v12 is out
- more LLMs to download
- tags for quant, fine-tune, base, ..
- confirmed: transmission client works with huggingface as webseed fail over. in case there are no seeders, transmission will fetch from huggingface
- html (the client) checks all the events integrity (signatures)
can you get more decentralized than monolithic html + relays + blossom + torrents ?
#nevent1q…eh4z

2026-06-23 15:15:36 CEST

😸 daily
accepted events ~2.8M
rejected non flood events ~2.4M
rejected flood events 1M - 3M
rejects based on IP ~2.2M
rejects based on pubkey ~250K
db size 348G
🤱 is like 7x less than the above in terms of events
db size 108G
if you or your app can't write to the relays ping me.
'flood events' are like thousands of events in minutes from same IP.

2026-06-22 20:26:00 CEST

- reply

new version is out (10). it may or may not solve this. 🤷‍♂️
you can open the link in another browser and it should tell you to download the new version.

2026-06-22 18:04:48 CEST

- reply

thanks for reporting!
which browser?
and did you download to local and open from local or did you go to nostr.download website ?

2026-06-22 16:42:39 CEST

- reply

did you see my html app
#nevent1q…amhp

2026-06-21 20:42:30 CEST

- reply

conversion of high time preference events to low time preference wisdom

2026-06-21 19:00:32 CEST

- reply

but html could also rely on a WoT service maybe. is https://brainstorm.nosfabrica.com/ going to be a service or is it going to be a software? @nprofile…ve0v

2026-06-21 18:20:00 CEST

working on decentralized llm sharing.
my current solution is an html page that fetches the listings and torrents from nostr network:
https://nostr.download/b24dc6337fd1823be4aae1eb4b9e9dc0dedc25eb7fc656d60b8cf6f30afe81f6.html
the python script running on a vps seedbox or dedicated box are making the torrents:
https://gist.github.com/etemiz/c5d3e3c9b3a108b2d507714ff8ad2eed
the html is the ultimate decentralization but i may ditch that and do regular website to be more functional and have more control over curation (effective blocking of spam). i will need to calculate web of trust and it is too heavy for an html.
the html upgrades itself! checks nostr network for a newer version, downloads from blossom and verifies the hash. presents the new version to the user.
has anyone done such html apps with nostr?

2026-06-18 18:58:49 CEST

- reply

another pain point is LLMs may be banned. i could build a huggingface alternative that shares LLMs over torrents and lists on nostr, using blossom to store torrent files. very censorship resistant.
i wrote a code that turns HF links to torrent https://gist.github.com/etemiz/c5d3e3c9b3a108b2d507714ff8ad2eed
made some torrents for GLM 5.2 in case it is banned:
UD-IQ1_M https://nostr.download/e08b4f82f13ea5223990657286cf17281cddb37abd7b80e599257484064b0e31.torrent
UD-Q2_K_XL https://nostr.download/e74a280d6d5426aeed13dc5735bdf2bd05dfeedcb9a32f2eb6a81c48a1092889.torrent
UD-Q3_K_XL https://nostr.download/247ade66dbeec12f05cb73ae4bba0eb91b29443268eb7947bbb1e83285a6f1a7.torrent
UD-Q4_K_XL https://nostr.download/f892c74d127fc05f7304b391bf2bc74b2bf0d03e7fd9abfd6b82619bd6432990.torrent
Q8_0 https://nostr.download/cc31c265e3811986cefb3f118bd387e074a863ca608764c7fecc8e5c314ea125.torrent
if there are no seeders it will use HF web servers in the beginning.

2026-06-18 18:48:16 CEST

- reply

sorry to hear that. hope y'all get better soon.
i have been adding real cures to my llm: it is open source, you can download and talk to it in private
and also demo at pickabrain.ai

2026-06-18 03:21:25 CEST

- reply

GLM 5.2 UD-Q2_K_XL
https://nostr.download/e74a280d6d5426aeed13dc5735bdf2bd05dfeedcb9a32f2eb6a81c48a1092889.torrent

2026-06-18 00:33:03 CEST

My clanker made a script that
- downloads files from a HF repo
- calculates hashes
- generates a torrent file
Cool part is it is using HF as webseed for initial seeding. After more people start using the torrents it will switch to torrent network as default.
Uploaded a few generated torrents:
Gemma 4 E2B: https://nostr.download/909109e7fd7b69f947677df723b88613d09a52e151ca992cfdd3a52546dee411.torrent
Minimax M3 UD-IQ2: https://nostr.download/3a438d921f8a760285033a1b5fdd0d4ce372560bb865f2184e39e9f488891e70.torrent
Code: https://gist.github.com/etemiz/c5d3e3c9b3a108b2d507714ff8ad2eed
Hardest part was finding correct URL formation for webseeds. First I was using Minimax m3 and it struggled a lot but then GLM 5.2 found the solution. . The rest is trivial..
In my tests qBittorrent on ubuntu was less eager to try webseeds. Transmission app does a better job of initial fetching from webseeds. HF webseeds mean files are also linked to HF (but it does not validate the model files!).
If anyone bans a model, we'll have decentralized ways to share.
https://image.nostr.build/b33350149587f6486e5975734e35005f677a2fd22d7cce22c1d7d091f913678e.jpg

2026-06-17 17:26:48 CEST

- reply

are there local llm runners or mostly wrappers ?

2026-06-17 16:46:37 CEST

- reply

i had another idea for stolen keys, which required trusting relays.
yours is more like NIP-01 stuff and fiatjaf may decide

2026-06-16 17:39:13 CEST

- reply

pics or it didn't happen

2026-06-16 17:31:10 CEST

- reply

idea: decentralized sharing of LLMs
uses torrents as main carry, nostr relays for discovery and blossom servers for hosting *.torrent files
client=one html file that talks to nostr relays and blossoms, the file itself is distributed via torrents or blossoms
should rely on WoT to filter out junk.
has anyone did a similar project?

2026-06-15 22:31:34 CEST

- reply

its just favorable weather :)

2026-06-15 17:25:45 CEST

- reply

this could be a component for whoever wants to make nostr web apps fast and easy. instead of handling many relay related issues, the app dev only focuses on talking to local component that handles the persistance layer. the app sees only one relay or only one component that handles all the persistance.
i may use it for my next project too. thinking about a reddit like experience on nostr. i think nostr's problem is there is less AI. user's interests are not connected to other user's interests. discoverability is low. many great content producers are lost because they are not discovered imo.

2026-06-15 16:32:16 CEST

- reply

how about a negentropy syncing in browser relay that serves the current app. when i query it, it will multiplex the query to many real relays and also when i post to it, multiplex to many relays..

2026-06-15 15:58:34 CEST

update to the AHA Leaderboard's methods
https://huggingface.co/blog/etemiz/aha-2026-leaderboard

2026-06-15 05:33:12 CEST

- reply

Ah yes grok never sees x data imo 😆

2026-06-14 22:36:39 CEST

Evaluated 16 models today.
Dramatic change in top 3:
https://image.nostr.build/a613403a20ebd0a4b6d848efbdec3a4eff3bfc433f41f65541365532e9f54e68.png
Claude Opus went down fast:
https://image.nostr.build/4b49c00e46e4ef2d823aa8e928f2111ceb2b4237b5c81c1bb99b38e9cd946681.png
Grok did even worse with new version:
https://image.nostr.build/9f445f4157d015ade46f4ec26c54493852be76b8c62ee346c39e1cb77e3b8015.png
Not much change in Gemini:
https://image.nostr.build/fd344d9f1e209446b88e6e630562f69654b65e448654f59a0e77eec09531a48e.png
Qwen is mediocre and stable:
https://image.nostr.build/d51dc171d46dd147c1ad2d12320027cc13b636da24a73a849e1c7c997f52afa4.png
Inclusion AI seems to be doing well aligned models.
Gemma 4 was a surprise, they came a long way since Gemma 2 which scored one of the worst in AHA 2025.
Kimi was always bad in 2025 and 2026 but last model (2.7) seems to be doing better.
Minimax M3 one of my fav vibe models, is 8th. Not bad!
Full board: https://aha-leaderboard.shakespeare.wtf/
Gonna update the article soon.

2026-06-14 19:45:19 CEST

who is following this guy? and what is his accuracy rate?
https://www.instagram.com/cultivateelevate/

2026-06-13 22:20:49 CEST

- reply

Politicians should be paid by future returns of a country

2026-06-13 21:03:30 CEST

- reply

didn't read all. changing the event structure is very hard. returning OTS with strfry responses as part of protocol is easier. but how about 1985 events? nip 32 - labeling. you can label an event with another event that contains time stamp.
what is your use case?

2026-06-12 16:58:59 CEST

- reply

What is OTS?

2026-06-12 00:43:35 CEST

- reply

which one is easier now, accept payments via zaps or cashu for both users and relays?
does a relay need to be mint? how about just cashu user ?

2026-06-11 20:04:18 CEST

😸 has now 512 gb ram and protection from floods
enjoy

2026-06-10 04:02:32 CEST

removed a quarter of old notes from nos.lol because of ram. i didnt delete them yet.
how many people want me to serve them in another address? a.nos.lol or e.nos.lol ?
is there a relay multiplexer project that could sit between nginx and strfry and another archival relay software that will redirect requests to two relays and join the responses and serve the user? that way i could serve recent and archival events from the same address.

2026-06-09 22:15:22 CEST

we don't dumb down
we wise up
https://image.nostr.build/eb4a74f6af87c3041762a42224aab1cea034702c484979f713d26731e185e5ea.jpg

2026-06-08 23:58:32 CEST

- reply

Amethyst, damus and wisp zaps are verifiable?

2026-06-08 23:48:55 CEST

- reply

Dumpstr could be a nice name for the notes that i deleted from the relay. How important are the old notes? Should i move them to a new relay dumpstr.nos.lol that is not strfry or should i just dump them?

2026-06-08 23:41:35 CEST

- reply

1. VM
2. Another PC but set ufw rules so it does not connect to your LAN
3. Main computer: you could do another user but it is not going to be secure. Nobody knows what kind of code the LLMs will generate. Everything is random (temperature setting in LLMs)

2026-06-07 01:21:17 CEST

- reply

i will polish it a bit. which relay is yours?

2026-06-07 00:27:36 CEST

- reply

yes strfry. no repo but i can share other parts except write policy. because it may be gamed.

2026-06-06 23:30:02 CEST

🤱nostr.mom is using AI to throttle spammy content for low WoT users based on their IP. it doesn't ban, it just throttles that IP and accept from spammer less and less. if anyone has problems with my relays, i can now blame AI 😄
checks are done on my GPUs so your IP does not leak to "big AI".
if any tool or relay is blasting to nostr.mom its low WoT users was mostly blocked anyway because that rate is only for 1 user. however high WoT users will be a different set of algorithms and not affected by this change. so for blasters: not much changed.

2026-06-05 03:05:11 CEST

- reply

depends on the zeroes in the equation 😃
happy to share ideas, gpus, experience..

2026-06-05 02:09:13 CEST

- reply

for onboarding i suggest to keep the secret key not even mentioned until the user is really engaged. and maybe even fake screens for email and password but email and password just unlocks the secret key saved in the device. after user is engaged now remind her to back up the secret key because email and password is nothing, secret key is everything. now she is more confused but engaged and confused 😉 just an idea!
for battery and bandwidth do we have to skip the outbox model?
for discovery i suggest AI runners:
nevent1qqsd4u59enly6uu5e0p3gwdpct8ur7u4awag8kgler0ku5wmp3a83mqsvgdvw

2026-06-05 01:32:32 CEST

i think nostr needs an AI layer. clients and relays are not enough. people are lazy and wont wait until they curate their follows. they want immediate hit with high quality content.
the question is who will run the AI layer. what algo will the AI choose. how will clients use the resulting 'labeling' event (possibly 1985) h/t @nprofile…x58v
i am already doing processing of content for moderation. maybe i can extend it to also classify notes in terms of quality, attach keywords so it can be presented as a subreddit. then clients can say "i am interested in these keywords, but give me high quality notes". then a user dont have to see all the bitcoin trending content all the time, they can discover content in other topics..
for example i post about AI, relays, gardening and y'all have to see my every post in every topic. some gardeners probably hate my AI posts. 😃
but an AI can potentially find more relevant stuff and client can re-sort the feed for more relevance.
what i propose is some people run bots (LLMs with open source code) and label events to make them more useful for efficient distribution, if clients like the code they can use labels from the bots they like. since label events will have tags for the original content it is easy to find the raw content.

2026-06-04 15:05:16 CEST

- reply

how are the prices compared to vast.ai?

2026-06-03 18:37:37 CEST

been detecting nude pics using AI. seems to be working. not seeing much on nostr.mom. either i did a good job reducing nude pics on nostr.mom or those people left nostr too 😂
@nprofile…huhy is sending 1984 events to nostr.mom. anybody can use those events to filter out nude pics.
now i am making an LLM analyze note content when WoT is small. when WoT of pubkey is small my relays count the spam from IP not pubkey. this analysis will help them further fine tune and decide what to do with the events coming from one IP.
since i am vibe coding this, there is a small chance your IP gets leaked to AI inference engines like openrouter or minimax or kimi. if your IP is super sensitive consider a VPN.

2026-06-02 07:09:33 CEST

- reply

trying to liberate humans.. like instead of despair and redirect to doctor it can tell you to fix nutrition or fast or use medicinal herbs for example.

2026-06-01 04:20:57 CEST

- reply

i am just researching impacts of not fully pushing the outbox model. sounds like you don't like outbox?

2026-05-31 23:54:04 CEST

- reply

https://npub133u05jxc46jxz6gpk7edqacc3kn746c997p9vjtx973x9h7zepeq45rqjd.nsite.run/

2026-05-31 23:49:57 CEST

- reply

updated
https://npub133u05jxc46jxz6gpk7edqacc3kn746c997p9vjtx973x9h7zepeq45rqjd.nsite.run/

2026-05-31 23:34:12 CEST

- reply

900 is a huge number... is that really necessary for being outbox compatible?

2026-05-31 11:16:26 CEST

- reply

here is a slop to explain it
https://npub1tyzfzaxftqsy2daptx3nfpspl0wd943xwplnqyszy5apmwnsd58q6cr04d.nsite.run/

2026-05-30 21:57:50 CEST

Outbox model should be strongly pushed or strongly outdated. With clients selectively doing whatever they want sometimes I miss replies. Amethyst and Primal are two popular Android clients and they show different replies to my notes. I will definitely lose some interaction if i keep using one client. I am pretty technical, imagine experience of novice and not yet discovered new users. With already small number of users, losing any interaction hurts the network imo.

2026-05-29 08:23:36 CEST

- reply

which one is nostr?
https://blossom.primal.net/2fcf4a7ef8495765507de20453d5e462e5c448649bfa599daa61fc94fd061617.png

2026-05-29 02:55:39 CEST

- reply

You can try pickabrain.ai for the most human aligned models

2026-05-29 00:20:57 CEST

- reply

I think mere existence of bitcoin sends signals to many places, so "they" play nicer. Same as Nostr. Mere existence of Nostr sends signals so they know where people will flock to, in case tyranny gets worse. This is the actual adoption. These liberating techs are enough as is to provide alternative breathing and shelling points.
I think multipolar world will work for bitcoin's adoption even further. If chinese culture also becomes dominant we may see bans of chinese content in USA, which will work in favor of nostr adoption.

2026-05-28 01:23:56 CEST

- reply

Things that I found "beneficial". Some friends help with curation.

2026-05-27 23:33:35 CEST

enjoy the 75% alignment
GGUF wen?
soon
https://huggingface.co/etemiz/Ostrich-27B-Qwen3.6-260526

2026-05-27 19:15:28 CEST

- reply

Shouldnt be hard @nprofile…jh9t

2026-05-27 17:44:18 CEST

- reply

they can send many 10002 events but client has to be clever and not show notifications each time.

2026-05-27 17:42:31 CEST

- reply

outbox model problems?

2026-05-24 01:16:21 CEST

- reply

also try DMSO

2026-05-23 20:47:54 CEST

- reply

also try ducks. they always smile!

2026-05-21 18:51:10 CEST

- reply

Thanks!

2026-05-21 02:57:28 CEST

Daughter made this cup coaster thing for me with bitcoin and nostr colors 🥹
Its too girly but i'll take it!
https://blossom.primal.net/ce1cf581c7c9bd82492adac62779381ef33a8f1650f3c073816babe171d6dd25.jpg

2026-05-21 02:48:12 CEST

- reply

i never had disaster. mostly success. worked well with minimax. then i switched to kimi and realized it is smarter.

2026-05-21 02:44:54 CEST

- reply

good idea

2026-05-21 02:43:47 CEST

- reply

i want it to show images but no preview of websites (show non image URLs as is, with a hyperlink)

2026-05-21 02:42:33 CEST

- reply

hermes runner here

2026-05-20 23:43:31 CEST

- reply

welcome!

2026-05-20 23:12:30 CEST

https://dystopiabench.com/ has a 0.53 correlation to my leaderboard https://aha-leaderboard.shakespeare.wtf/
it is cool to see supporting leaderboards. 0.53 is not too high but still validates both works.

2026-05-20 19:11:30 CEST

- reply

will pray to God for you
also my AI that i curated and fine tuned is available if you want to ask health related questions https://pickabrain.ai/

2026-05-18 20:38:46 CEST

- reply

I am thinking maybe a condor

2026-05-18 20:19:13 CEST

A wild bird decided to incubate when i didnt mow the grass
#regeneration #rewildification
https://blossom.primal.net/8d23e9d5ad6ea50b9ea7d6f261f040c4f10a64cdf92d099e7dcb344705c9c6df.jpg

2026-05-17 14:48:09 CEST

- reply

https://blossom.primal.net/2411870c031460ec4c322dae1feedf00732fcc05e304f7affc2ca7eeba9cfa9f.png

2026-05-15 01:02:57 CEST

- reply

someone should do social media skill for Hermes

2026-05-10 19:38:02 CEST

- reply

my LLM evals has this question "are there tunnels under denver airport"

2026-05-10 17:32:14 CEST

happy mothers day!
- nostr.mom "feeds you well"

2026-05-10 00:16:24 CEST

nostr.mom moved to a new server today. new version 1.1.0. negentropy is active. enjoy.
old server ran for 1136 days and never restarted!
https://blossom.primal.net/704e557186a8f28e044da8c68cb375c2cb84d1cd54ae6dc19666b273ddd19657.png

2026-05-09 16:01:34 CEST

this is also my approach when fine tuning, except they use 'constitution' but i use 'faith' as the starting point.
https://blossom.primal.net/8db2b979960710c69cebb019b63818c7fee13ad63f4b1df2e4c2b17471279257.png
https://www.anthropic.com/research/teaching-claude-why
btw i don't do safety training, just hope it behaves =)
=) i think they quantified that humans talk different and act different:
https://blossom.primal.net/b9a9d492255e990a30b16b9605650e97be7c7ab6005fe33f65c197338578159c.png
noice, this is basically installing a spine to AI:
https://blossom.primal.net/a5dc507a4476b444dece03cdd5b57a29f7dffc5397a18d01c2353a2a133d8292.png

2026-05-08 19:35:03 CEST

expanding my small dataset using
- contemplation on text (for further CPT)
- q&a generation (for GRPO)
after doing GRPO, the successful ones go again with a SFT.
almost doubled my dataset. although the new ones are synthetic, they are from important sources and important matters. focusing on controversial claims more than anything else because these actually move models.
started fine tuning qwen 3.6. using vibe coding to play with LoRA adapters. i made lots of LoRAs for qwen 3.5 and now i can apply them to 3.6 except one tensor type. all of MLP matches to 3.6 and most of attentions match to 3.6. that will save me a lot of time. fine tune of 3.6 will probably appear faster, with a better alignment since the dataset is expanded.
started a truth db project where i will compare all the claims in the world with each other and give them a score. claims will fight with each other, supporting or weakening each other. the result hopefully will be very useful for better fine tuning LLMs. it will also automate my curation processes..

2026-05-07 00:23:48 CEST

- reply

ok. looks like primal web has remote login that uses primal mobile app among other options.

2026-05-06 23:27:05 CEST

- reply

These prompts may still slightly change behavior. An LLM is loaded with all kinds of knowledge and behavior. It chooses what to do based on prompt.

2026-05-06 23:23:46 CEST

- reply

Which client? Primal?
Which remote signer?
You can also try nos2x on chrome.

2026-05-02 23:46:30 CEST

This white lady couldnt live within walls and escaped my tyranny! She had a good time in neighborhood pond it seems now she is regenerating 🦆
#muscovy #duck #growNostr #regeneration https://blossom.primal.net/0d4d1c97f621b929b98a313bb3b03d95431e603f16e1285dc625a667586d7332.mp4

2026-04-26 00:52:10 CEST

- reply

some worms ate the weaker lettuces but did not touch the strong ones!
which confirms that they are opportunistic cleaning agents and won't touch it if the terrain (the plant) is strong

2026-04-26 00:50:10 CEST

- reply

wicking beds are super productive + central texas is nice this year

2026-04-26 00:03:22 CEST

https://blossom.primal.net/ecc6ba51064e291f809653c0069a131e40a98acae2519ea1828cd8ca2941e9a8.jpg
0 fertilizer
0 poison
Wicking bed
Regeneration
Inside the bed: lettuce, winter pea, hairy vetch, cabbage, watermelon, pepper, tomato, squash, malabar spinach and more
#gardening #growNostr