utxo the webmaster 🧑‍💻

Wisp: https://wisp.mobile
Nostr Archives: https://NostrArchives.com
Haven: https://github.com/barrydeen/haven
Rektbot: npub1r3kty2vkh247jgdu63wgkcsnktdtp9hc3e962eudg0getgvxs4gsz4uytc

All Projects:
https://github.com/barrydeen

Public Key

npub1utx00neqgqln72j22kej3ux7803c2k986henvvha4thuwfkper4s7r50e8

NIP-05 Address

utxo.one

Profile Code

nprofile1qqsw9n8heusyq0el9f99tveg7r0rhcu9tznatuekxt764m78ymqu36cprdmhxue69uhhyetvv9ujucnfw33k76twwpshy6ewvdhk6qgmwaehxw309aex2mrp0yh82arcduhx7mn99ahh2arzdauq9wdl30

Publishing to

relay.bitcoinpark.com

relay.utxo.one/outbox

Author Public Key

npub1utx00neqgqln72j22kej3ux7803c2k986henvvha4thuwfkper4s7r50e8

Show more details

Last Notes

2026-08-31 05:20:44 UTC

- reply

A very important degree

2026-08-31 02:45:56 UTC

Remember 2021? That was fun
https://relay.utxo.one/5bde7e89d6a6c12a122d964bc07c512af8712f01f035c6bfdc601163e6018be8.jpg

2026-08-31 02:17:31 UTC

- reply

I have reference implentations in wisp and strfry and NostrArchives. If clients haven't adopted it by now it's likely choice not lack of ability

2026-08-31 01:37:53 UTC

If nostr just ends up a way for agents and robot ducks to communicate, @nprofile…cy6q is gonna be really mad

2026-08-31 01:07:57 UTC

- reply

Gb10s

2026-08-31 00:36:10 UTC

- reply

Absolutely wild. Can't wait for us to get these asics

2026-08-31 00:22:39 UTC

- reply

Depends on task, mostly yes. If you're adjusting a layout or something it can get pretty annoying

2026-08-31 00:02:07 UTC

Definitely taking some getting used to being local AI only.
Shit is much slower than cloud, so gotta basically write huge plans then let the clankers bang them out overnight.
Can't do that thing where you prompt and start drooling watching the clanker go

2026-08-30 22:59:42 UTC

Substantial improvements to the nspam model across the board, from 92% accurate to 97% accurate (looks small but actually huge... that's what she said)
Also added much better spam detection accuracy in Portuguese, Japanese and Spanish.
Get fucked spammers :wisp_salute:
Rolling out to wisp and dark wisp next week, links to the benchmarks, open weights and dataset below:
https://relay.utxo.one/f3884fc371b6fc70fcf61a2d229ba283ccc1ccf7f7af861bd123bcb6baa80fc6.jpg
https://huggingface.co/barrydeen/nspam
https://huggingface.co/datasets/barrydeen/nspam-corpus

2026-08-30 22:27:57 UTC

Just manually labeled 700 npubs and added over 60,000 notes to the new nspam corpus
Now we train and pray the new model is more accurate

2026-08-30 19:56:26 UTC

- reply

Thanks for the advice

2026-08-30 19:52:15 UTC

Basically impossible for the model not to determine this is a bot yet I know this is a real person
https://relay.utxo.one/db21135145e592967d857197467c205c4868727f90a91db3b5272995b5d3ad08.jpg

2026-08-30 19:42:30 UTC

Getting reports spark wallet is having issues
There's nothing I can do, if it's not back up in 5 minutes
https://utxo.blossom.band/869b5a66e644ebb7b4c1ee8c373b6d74790096cf7407df310891f7fdd8a70138.mp4

2026-08-30 18:53:41 UTC

Local model fixing Claude's bugs 💅
https://relay.utxo.one/743068fe7880c731d1b099da3b940b55bb8967fd8f9234a491015a5ed4f2b031.jpg

2026-08-30 18:34:02 UTC

- reply

Ok thanks taking a look

2026-08-30 16:51:07 UTC

- reply

Can you restart the app and lmk if it fixes? Seems to be working on my end wonder if it's a weird cache thing (don't clear your cache tho just restart)

2026-08-30 14:43:48 UTC

Toggle web of trust on to disable the current wave of spam on wisp and dark wisp
https://relay.utxo.one/75b876e5d86eaaf5fe9b86b4f288556aeb48382c3842b94f4e83f592578e98d3.jpg

2026-08-30 14:31:14 UTC

- reply

It does if you turn it on

2026-08-30 14:15:30 UTC

Working on a new nspam model to fix this new wave of spam, they'll be dead next week :wisp_salute:
https://relay.utxo.one/28f143bc860de4013c4757f81ec8417b72b0113725d83ffc2f1ed20bbd516867.jpg

2026-08-30 14:11:07 UTC

- reply

Depends on your hardware but anything with a mixture of experts

2026-08-30 14:00:59 UTC

- reply

Big cities do, not my little town tho

2026-08-30 13:12:56 UTC

Good morning
Local AI keeps winning
Now you don't even need an insane GPU to run MoE models at very fast speeds on consumer hardware
https://github.com/FlashML-org/FreeToken

2026-08-30 02:15:07 UTC

- reply

The GPUs are like 16-18k each

2026-08-29 20:39:20 UTC

https://github.com/elie222/rakazo

2026-08-29 18:00:31 UTC

- reply

Lightning sucks, probably will work in 10 mins

2026-08-29 16:41:29 UTC

- reply

No 🙏

2026-08-29 15:18:50 UTC

Fuck man why didn't I pull the trigger on such a machine last year when it was half the cost
This is like a $50k build today
https://relay.utxo.one/e6f9a704457fa2a268c99423f7b9770e4c416431350e621299a7ba48a922ddad.jpg

2026-08-29 14:41:10 UTC

- reply

Yes pls

2026-08-29 14:29:49 UTC

Morning
Ran out of milk for coffee, ran to Tim Hortons
After ordering mf tells me it will be 15 mins while they brew a new batch after I paid, so I just left
Sorry but $1.50 isn't worth 15 mins of my time waiting for the shittiest coffee in the world

2026-08-28 20:27:44 UTC

Non flash weights just dropped
https://huggingface.co/zai-org/GLM-5.3

2026-08-28 20:12:44 UTC

- reply

I like how you think

2026-08-28 19:14:27 UTC

- reply

But 10x slower prefill

2026-08-28 18:56:27 UTC

- reply

Blackwell pros are almost $20k and that's before you put it into a $5k rig.
Gb10 bandwidth scales the more units you add with tensor parallelism

2026-08-28 16:06:57 UTC

- reply

This is running nvfp4, you have to run a patched backend with unmerged PRs to get it working

2026-08-28 16:05:48 UTC

- reply

Mac studio for sure, if it will even be possible to buy, which it probably won't

2026-08-28 15:14:14 UTC

- reply

Only on real hardware not my Lego PCs 😭

2026-08-28 15:04:21 UTC

Speed demo: Glm5.3 flash on 2x GB10s
https://relay.utxo.one/8125e248f9192acc5eb8cfcb812e6708c4eefdcedde6bdb1f885626180ecd779.mp4

2026-08-28 14:46:21 UTC

I am hearing that china is about to dump so much ddr5 on the market, cxmc just completed some factory waffer thing, possible cope post, idk, ram prices are whack son

2026-08-28 14:34:38 UTC

- reply

I hate it

2026-08-28 14:30:42 UTC

- reply

Yeah it's shit

2026-08-28 14:25:30 UTC

Gonna try omarchy, popos been annoying AF since going to cosmic

2026-08-28 13:06:36 UTC

- reply

Classy

2026-08-28 13:04:32 UTC

Probably gonna waste 4-5 hours trying to get 4-5 more Tok/sec today, thank God AI is here to save me so much time
Then as soon as it's optimized a new model will drop and the cycle with repeat

2026-08-28 12:26:08 UTC

Good morning:wisp_cute:

2026-08-28 00:58:29 UTC

- reply

2 dgx sparks

2026-08-28 00:34:47 UTC

Glm5.3 makes it official, only clanking sovereignly from now on
https://relay.utxo.one/935dad8920ce8a043e0fbee3a329bcd4799715e994026a3ad146ebaab53fef2e.jpg

2026-08-28 00:14:59 UTC

- reply

Nah

2026-08-27 23:49:06 UTC

Saylor posted an AI video of himself having sex with a bull
https://relay.utxo.one/8d0668a9fdea0324d14c3b33d140fc153901f3ba5a1c96d2797a6224d5ea88b7.jpg

2026-08-27 23:07:28 UTC

Harvey boy enjoying a chicken carcass
Dogs can safely eat chicken bones if they are raw
https://relay.utxo.one/0db37bd7f929055c5acc33ad5c7bd859de129217fc90ab4a73d71683ab80aeeb.mp4

2026-08-27 21:23:02 UTC

Trinidad curry #foodstr
https://relay.utxo.one/8124fc8da054614bc54280c8ed62501f15d0831aa98435ed83da5fda7588b2ee.jpg

2026-08-27 21:11:00 UTC

- reply

New release should fix it, out now

2026-08-27 20:47:16 UTC

- reply

Same. It's sick. Kinda slow tho. Wen dspark

2026-08-27 20:18:39 UTC

- reply

Imagine how much bandwidth they use

2026-08-27 20:03:39 UTC

🦜🏴‍☠️
https://relay.utxo.one/4758fd5680783f9d667772f8c65a7b720e90392fa3242cc85c928f2867a32e92.jpg

2026-08-27 19:20:19 UTC

- reply

Tool calling currently bugged out on it waiting for a patch

2026-08-27 17:11:04 UTC

- reply

For sun protection I use a hat and shade

2026-08-27 16:53:34 UTC

- reply

My skin care routine is don't put any products on my face

2026-08-27 14:10:12 UTC

- reply

I am running models on 3090s...

2026-08-27 14:01:39 UTC

- reply

It's a fun psyop to fall for speaking from experience

2026-08-27 13:59:01 UTC

Here's an example of what GLM5.3 Flash produced with one prompt locally, all I told it was to make a reimagined nostr UX using the outbox model.
https://bloom.utxo.one/
Don't get attached to it, I will likely take it down at some point

2026-08-27 13:50:47 UTC

Nvidia looking to buy hugging face for $13 billion:wisp_eyes:

2026-08-27 13:40:30 UTC

- reply

That hasn't been true for at least 4 years now

2026-08-27 13:32:39 UTC

- reply

Nice plot of land sir

2026-08-27 12:20:12 UTC

- reply

You need 2

2026-08-27 12:16:32 UTC

Good morning,
Glm 5.3 flash is no joke an approaching endgame level intelligence that you can realistically run at home, especially with these new Mac studios
Re: cost, think of this
If a landscaping company bought $10k of equipment for this business, you'd consider that completely normal. And so a dev, who probably makes more than a landscaper, can justify having a 10k computer for their work.
(I hope you enjoy my cope excuse and go buy some compute)

2026-08-27 01:40:27 UTC

GLM absolutely dominating DeepSeek on quality, this is sick

2026-08-27 00:18:40 UTC

Currently getting about 30 Tok/sec decode on GLM flash. Definitely not as fast as deepseek (today)
When dspark comes out this will maybe go to 40-45

2026-08-27 00:08:46 UTC

Glm is definitely trained on our notes, it did this fully offline
https://relay.utxo.one/20f3039425f0c1f971abe4a0e3d35d17397c29d55bd1ef2750959499d0f5f51e.jpg

2026-08-26 23:34:26 UTC

Blowing my brains out trying to get glm going locally but we are close gang

2026-08-26 23:27:52 UTC

:wisp_eyes:
#nevent1q…sz5f

2026-08-26 23:23:36 UTC

- reply

They weren’t hot they were the leftovers

2026-08-26 23:21:49 UTC

- reply

Need hot boobie girls

2026-08-26 23:17:28 UTC

- reply

Hmm I wonder if wisp isn't sending the client tag on these

2026-08-26 23:16:15 UTC

- reply

Why arrested? Right to bear arms

2026-08-26 23:13:32 UTC

- reply

We buying here or what

2026-08-26 21:32:05 UTC

- reply

No 🙏

2026-08-26 20:42:27 UTC

Running qwen3.8 flash :wisp_salute:

2026-08-26 18:44:22 UTC

- reply

No 😭

2026-08-26 18:31:06 UTC

Lol bitcoin
#nevent1q…r88s

2026-08-26 17:23:51 UTC

- reply

Damn for real?

2026-08-26 17:15:12 UTC

Good afternoon #foodstr
https://relay.utxo.one/43a225c193b4b06d4e5165f24b90b07169543de02425d4e208fccc3ee70dc211.jpg

2026-08-26 17:07:21 UTC

Grilled a steak at 1500F on the side burner 🤌
Took about 4 minutes
https://relay.utxo.one/9176ce2cd5f40c510df4365d50cf77f15903e70b01ff35f3f343af02cb14e11d.jpg

2026-08-26 16:52:29 UTC

Fuck yo cloud
https://relay.utxo.one/bfe1b9f347bf2d7f8c8fd3cc484425e479977dce9d45e6598e00bb3297307dc7.jpg

2026-08-26 16:35:05 UTC

- reply

It's an aggregate of standard test benchmark tests, anyone can run them and find out for themselves

2026-08-26 16:22:23 UTC

- reply

Take my picture off your nostr pls banned

2026-08-26 15:56:36 UTC

For any gb10 owners:
https://github.com/barrydeen/glm53-flash-dgx-spark

2026-08-26 15:43:03 UTC

GLM 5.3 FLASH BEATS FABLE AT AGENTIC WORKLOADS
https://blossom.primal.net/e9b081ce5c882dc05d54bafbaa3458f5be546e8029860b20fced0888ec7fb24e.png

2026-08-26 15:38:03 UTC

looks like this weight works with the experimental vllm branch, running it now
https://huggingface.co/LibertAIDAI/GLM-5.3-Flash-NVFP4

2026-08-26 15:26:05 UTC

Both new glm and qwen models drop without any support to run them so annoying!!! Can the devs do something?

2026-08-26 14:21:27 UTC

- reply

Yes fair although that’s at least two and quant 4s kinda rough

2026-08-26 14:18:42 UTC

- reply

Between 20-60 depending on the model
You’ll need 4 6000s to run this pretty sure lol

2026-08-26 14:14:24 UTC

BOOM
https://huggingface.co/zai-org/GLM-5.3-Flash

2026-08-26 14:09:41 UTC

- reply

This is the best I can see I can't get it deployed lmk if you figure it out, clankers say it can't be done

2026-08-26 13:59:22 UTC

No qwen4 support on sglang and vllm sigh
https://relay.utxo.one/4b5616744ecdf8f4210c9e35caab450edcc7655b8bc12876208cc6019ed0f5a4.mp4

2026-08-26 13:52:24 UTC

- reply

I would definitely go for 512gb, probably will, gets you into that GLM class of models. But if I didn't already have sparks I would still grab a 256gb version today to run qwen and deepseek. Of course only do this if you think you can make money from it or have private workloads

2026-08-26 12:53:11 UTC

- reply

I think flash basically implies a smaller MoE model than the big daddy. I liked my GB10 setup until Apple's release yesterday 😭 - I would definitely get an m5ultra over the gb10

2026-08-26 12:36:58 UTC

Qwen Flash coming out today
Will it replace DeepSeek as my local daily driver? Stay tuned

2026-08-26 12:10:20 UTC

- reply

Usmca, the deal trump himself negotiated, is disrupted? These aren't facts you're repeating propaganda

2026-08-26 12:02:14 UTC

- reply

First the reason was fentanyl, then slavery (???) then wildfire? If you believe there are any reasons other than Trump's delusions of grandeur to absorb Canada , you're falling for the psyop

2026-08-26 11:20:08 UTC

Good morning
Reflecting on covid times, the scary thing wasn't what the government did, because we already know they are psycho...
It was how the people went along with it, turning friends and family against eachother. And when the masks came off, there were no apologies.
Makes me wonder what will happen when the red hats come off, will there be any apologies for turning against Canadians for no reason? Probably not