ngmi

https://metaend.qstorage.quilibrium.com

hi~

Public Key

npub14p7fhlz9wrlptt7a5mqqjkk79460k98r999l7dafv4fznqaa565qhjdzry

NIP-05 Address

Profile Code

nprofile1qqs2slyml3zhpls44lw6dsqftt0z6a8mzn3jjjllx75k253fsw76d2qpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0ds2m208c

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub14p7fhlz9wrlptt7a5mqqjkk79460k98r999l7dafv4fznqaa565qhjdzry

Show more details

Last Notes

2026-10-01 06:00:22 GMT

Pixelcut Looping Video is now available! Turn one product photo into a seamless video loop with a locked camera and subtle ambient movement or a full 360° spin. Designed for shop listings, product ads, and social posts, while keeping the product consistent with the source photo.
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-30 10:57:43 GMT

- reply

allthey get from me is 'status'
not even a question mark anymore :aim-laughing:

2026-09-30 10:36:16 GMT

lmao
#nevent1q…ewl5

2026-09-30 07:15:00 GMT

Five experiments tested whether a slime mould's network rule helps AI models decide, route and retrieve, and it never helped overall.
https://paragraph.com/@metaend/a-slime-mould-is-not-an-agent
#ai #airesearch

2026-09-28 18:02:26 GMT

A slime mould is not an agent
https://paragraph.com/@metaend/a-slime-mould-is-not-an-agent

2026-09-28 15:12:44 GMT

@npub1jug…fc28 Ran breez.technology through metaend Grade: C (60%). Your llms.txt files score full marks, but pages don't answer agents asking for markdown or JSON (0/5), and llms-full.txt still says USDT/USDC "coming soon". Lite report attached. Full report: $29 in sats or Cashu.
https://blossom-01.uid.ovh/4c9f32882c9acea10cf44dcbab07899b35102b4fbb266d8dd87668735c70c1c8.png

2026-09-27 14:42:11 GMT

engineering a system 0 experiment. kinda exciting

2026-09-27 05:45:03 GMT

Does anyone run nano muse here? Looking for experiences
https://nano-muse.github.io/

2026-09-27 04:47:58 GMT

Would you sleep there?
Model: Seedance 2.5 flash
via https://nano-gpt.com/subscription/kUGJhgF9
#seedance #ai
https://blossom-01.uid.ovh/edee028a9650d9f12dd43747a67c80933bd70f887dcc462c7757d29678c557b7.mp4

2026-09-27 03:43:07 GMT

Found a perfectly reasonable future proof job
My work 'UFC Athlete' scored 14/100 on AI exposure (Low) with the free Tideproof scan. How exposed is yours? https://tideproof.org/score/low/

2026-09-26 04:03:50 GMT

- reply

too big, here's my research for my laptop:
CLM is not worth trying here. Nothing was downloaded or run. Remy stopped at the reading stage, and my recommendation is to skip it.
- Too big for this laptop: the CLM part itself is tiny (76 MB), but it only works on top of the Qwen3-8B model. That's a download of at
least 5 GB, realistically 8.7 GB, and CLM's documented setup needs a GPU server (vLLM). Running it on a CPU is untested by its authors and
would need extra workarounds.
- Weak published results: on JevBench it scores well below Jev and JevK5, and on the hidden test questions it's even below random guessing
(24% vs 29% chance). An open bug report says its rating questions give the same answer for every input.
- Clean licence: Apache-2.0, and it wasn't trained on Jev's answers.

2026-09-26 03:35:22 GMT

Contrastive Language Models (CLMs) are a new class of System One model trained with a contrastive learning objective that connects states and actions. This repo serves CLM-8B behind a TypeSafe-compatible API.
https://github.com/Contrastive-LM/CLM

2026-09-25 17:28:52 GMT

- reply

better than with a bus?

2026-09-25 07:55:46 GMT

The honest verdict: for our current jobs Jev wins clearly, and GLiNER earns no role yet.
Meh if you spent compute at least make sure to test your jev replacements on a real project not the trust me bro benchmark
#jev #gliner

2026-09-25 07:02:53 GMT

imagine you work 2 years to launch #jev and not even 7 days later people train opensource version of it that's exactly as good. imagine being the vc behind burning all your money on this horse

2026-09-25 06:53:14 GMT

MiMo V2.6 Flash Uncensored is now available for text-only chat, creative writing, coding, and long-context work. This lower-refusal MiMo finetune supports reasoning and tool calling with a 1M-token context window. It is pay-as-you-go only.
https://nano-gpt.com/subscription/kUGJhgF9
#mimo

2026-09-23 09:11:40 GMT

how do i host private ngits? Still a bit lost
#ngit #asknostr

2026-09-22 17:05:22 GMT

The only reason Opus 5.5 is on par with Fable 5.1 is because they nerfed Fable 5.1 over the last couple of weeks, allegedly.

2026-09-22 10:34:27 GMT

An ex-OpenAI researcher just deleted language from the LLM...
https://www.youtube.com/watch?v=TbkUKCm3CHQ

2026-09-22 07:04:55 GMT

I think #jev will be replaced by opensource models in record time

2026-09-21 12:12:14 GMT

Greenference is now available in provider selection for supported GPT-OSS, Qwen, Gemma, and GLM models. Greenference is a European inference company whose broader focus is modular, off-grid datacenters powered by solar energy and battery storage, designed to add sovereign European AI capacity without waiting for conventional grid connections. Choose Greenference in the provider selector or pass `X-Provider: greenference` in API requests.
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-21 11:46:42 GMT

Sounds good on paper 🤔
Swift-Qwen3.8-27B is UkisAI's reasoning-efficient derivative of Qwen3.8-27B, using 58.3% fewer thinking tokens while maintaining near-identical performance (<1% loss) and as a result getting a x1.95 speed-up on several tasks.
https://huggingface.co/ukisai/Swift-Qwen3.8-27B-GGUF

2026-09-21 09:02:38 GMT

Just by wiring rtk to my omp.sh I was able to save 2.7m tokens in a couple of weeks. That's a lot, especially since it's for free.
#llmoptimization #ai #rtk #omp

2026-09-20 06:29:45 GMT

Four jobs, one survivor, and a rule that will outlive the tool: before adding a decision engine to a fork, check whether the fork has a computable answer. Most of ours did.
https://paragraph.com/@metaend/we-put-the-typed-decision-engine-on-four-of-our-own-jobs-it-lost-three
#jev

2026-09-20 06:03:59 GMT

Qwen 3.8 27B Cybersecurity, Gemma 4 26B A4B Cybersecurity, and GLM 5.3 Flash Cybersecurity are now available for security research and coding, with image understanding and tool calling. These variants are based on uncensored models and retain moderation for illegal activities.
Nemotron 3.5 Content Safety is also available to classify user messages and assistant responses as safe or unsafe. All four models are included in the subscription at the standard 1× usage rate
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-20 05:56:39 GMT

- reply

2 people on armada coordinating and trying to keep the harness in check

2026-09-20 05:24:08 GMT

Imagine giving your 200 pages real estate blueprint a life, walk through it in your mobile AND select/modify the interior as you walk by 🤔
https://blossom.ditto.pub/d4c7309cc9f2b34abb19b1200f46d190ec3a0d8c7de3dac46a580293c2a3506b.jpeg

2026-09-20 04:31:54 GMT

- reply

excellent!

2026-09-19 18:17:33 GMT

Jev thread is closed with a clear verdict: one remaining job (citations vs sources), everything else lost to a deterministic alternative.
My verdict after sending our full harness fleet to research usecases: #jev is useless

2026-09-19 10:57:01 GMT

Our harness running on omp is so buffed to the tits that even absolute state of the art researches and extensions like SoL-Pi are rejected because we measurably already have these implementations at or even better level

2026-09-19 10:12:42 GMT

After ~1 day of checking our harness, I finally found a niche usecase for #jev and it's okay.
https://i.nostr.build/N5k94UqIbQEdOKuaDr1jOp.png

2026-09-19 08:24:44 GMT

We they tell you to check dm but you're on ditto
https://gifverse.net/media/oDWbhMec/original.gif

2026-09-18 08:01:12 GMT

Our AI Harness is currently becoming an expert designer, it built all of this just based on a 600 pages architectural design pdf using free assets on top. imagine the possibilities for architects, constructors, investors... and Alice (our harness) is becoming super efficient and fast. Don't like an asset? click it, comment, and the ai changes it asap.
https://v.nostr.build/l0Jl0uokWeaS19ybZhybve.mp4

2026-09-17 07:49:00 GMT

uhh received my first #monero zap
https://gifverse.net/media/BINApIc4/original.gif

2026-09-17 07:06:28 GMT

Waze but with openstreetmaps and nostr. also parking connected to it updating live free spaces, so you can route around crowded areas

2026-09-16 11:11:41 GMT

- reply

Wen arch?

2026-09-16 11:09:20 GMT

Cool!
Obscura VPN's new 'Circle of Trust' rewards users for supporting independent privacy apps
https://www.techradar.com/vpn/vpn-services/obscura-vpns-new-circle-of-trust-rewards-users-for-supporting-independent-privacy-apps

2026-09-15 17:12:50 GMT

What is this magic @@npub192m…d0ch https://brainstorm.world/p/npub14p7fhlz9wrlptt7a5mqqjkk79460k98r999l7dafv4fznqaa565qhjdzry

2026-09-15 17:07:16 GMT

Introducing open-1b: the first model you don’t have to trust
https://www.gensyn.ai/news/introducing-open-1b-auditable-training

2026-09-15 14:16:41 GMT

Lunar built and open-sourced NovelWriter, a novel writing app with built-in AI analysis tools and a side panel AI agent that understands your manuscript as you write. Features include automatic event timelines, plot tracking, relationship webs, pacing graphs, inline scene suggestions, and a simulated reader group chat reacting as you write. Everything runs and is stored in your browser.
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-14 11:09:52 GMT

NanoGPT accepts #monero and gives you extra bonus for paying with it. Also the monthly subscription is very affordable 👋
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-13 14:46:26 GMT

Our harness built a nice #nostr #omarchy notification plugin with #rust
https://gitworkshop.dev/[email protected]/relay.ngit.dev/nostr-notify

2026-09-13 05:47:57 GMT

Someone should let the fruit fly loose on nostr. Can't be more stupid than half of the stuff floating around here 🤣

2026-09-12 18:42:30 GMT

nostr #omarchy plugin for notifications
https://metaend.blossom.band/e3c6ff35837b46f52fc3f0dbc200b1c8979afdc3362e750fb545e9dacfe961e5.png

2026-09-12 17:19:43 GMT

fair point #nevent1q…sqyj

2026-09-12 16:41:25 GMT

are there any cool nostr plugins for omarchy yet?
#asknostr #omarchy

2026-09-12 13:37:17 GMT

Revolut: https://gifverse.net/media/Os2SsaVa/original.gif
#nevent1q…atsg

2026-09-12 13:27:06 GMT

Is your website ready for our new overlords?
Free scan https://metaend-grade.fly.dev/
#ai #agents

2026-09-12 11:26:31 GMT

What the fuck revolut
#nevent1q…fdrn

2026-09-12 06:57:16 GMT

https://metaend.blossom.band/8502d44e94e2e1f1beb653786ab2a5ddc54d439dccfcf5f4f4d38829d5595448.jpg
Their first prompt?
"Hey Mythos, design unbreakable bottle cap connectors, make no mistake."

2026-09-12 05:52:51 GMT

I use #omarchy btw
#nevent1q…6an2

2026-09-12 05:43:02 GMT

IONOS is now available as a provider on NanoGPT, adding a European hosting option for Qwen, Llama, Mistral and GPT-OSS, alongside OCR through the chat API. Try GPT-OSS 120B on IONOS.
IONOS runs these models on its own infrastructure in Germany. According to its data handling policy, prompts and outputs are never logged or stored, never used for training, and never shared with the model developers. IONOS retains billing and operational metadata, such as timestamps, model names and token counts.
This adds an option for users who want their model inference hosted in the EU with zero retention of prompt and response content at the provider.
https://nano-gpt.com/subscription/kUGJhgF9
#euai #ai

2026-09-11 16:35:28 GMT

- reply

yep

2026-09-11 16:30:36 GMT

- reply

I would take a cordn name :)

2026-09-11 09:01:46 GMT

Does #grimoire have a timeline view too? would be kinda cool

2026-09-11 06:48:40 GMT

1.1.1.1 now supports post-quantum DNSSEC, all 2,420 bytes of it
https://blog.cloudflare.com/post-quantum-dnssec-1111/

2026-09-11 06:11:46 GMT

hi @npub1vvc…t3w0 any chance you offer GRASP-08 managed relay hosting for private ngits?

2026-09-11 05:06:46 GMT

Insane if true
DeepSeek-V4.1-Flash debuts with $0.003/1M off-peak cached-input rate and benchmarks eclipsing GPT-5.6 Sol, Claude Opus 5
https://venturebeat.com/technology/deepseek-v4-1-flash-debuts-with-0-003-1m-off-peak-cached-input-rate-and-benchmarks-eclipsing-gpt-5-6-sol-claude-opus-5

2026-09-11 04:58:11 GMT

PaperCut Attacker Uses Hundreds of AI Agents to Compromise 440+ Instances
https://thehackernews.com/2026/09/papercut-attacker-uses-hundreds-of-ai.html?m=1

2026-09-10 10:08:11 GMT

If i was broke af but had an idea for a SaaS to make money, I would use
- Freebuff https://freebuff.com/?ref=ref-0e2541ba-c49e-4c05-8194-d904600973d5 and the freebuff CLI to build my idea, starting with landing page. Link your github for better free model access.
- Set your LN address to @npub130m…l29s so tips go directly to your AI Harness
- Cloudflare to host
Good luck, and have fun

2026-09-09 12:12:31 GMT

trying to upgrade #omarchy and the servers just die on me

2026-09-09 10:28:45 GMT

There is a more than 10% chance that all anthropic researchers are totally lost

2026-09-09 07:34:00 GMT

Insane how mistral was able to raise 3 million €. Worse than local LLM. 🤔

2026-09-08 07:14:15 GMT

All crypto deposits on NanoGPT now receive an automatic 3% top-up bonus, giving you 3% more credit than an equivalent credit card deposit. A $100 crypto deposit adds $103 to your balance. The bonus applies to every supported cryptocurrency, every time you top up.
https://nano-gpt.com/subscription/kUGJhgF9

2026-09-08 06:25:00 GMT

- reply

Yes

2026-09-08 05:24:27 GMT

Wen #amethyst 14 in playstore? Been some time

2026-09-07 09:43:09 GMT

- reply

Is this another simple form of double spending?🤣

2026-09-06 11:09:31 GMT

Unlimited OCR Works: Welcome the Era of One-shot Long-horizon Parsing.
https://github.com/baidu/Unlimited-OCR
#baidu

2026-09-06 08:24:16 GMT

Uh oh
#nevent1q…khzm

2026-09-06 07:30:57 GMT

Typical shitcoin
#nevent1q…aw9c

2026-09-06 06:45:09 GMT

Freebuff is getting better 👋Literal free ai compute
https://freebuff.com/?ref=ref-0e2541ba-c49e-4c05-8194-d904600973d5
https://metaend.blossom.band/f3ddd7afa4613f0e31497a6146a6b85c8e54485078cc38dde2a22c1dcda3632c.jpg

2026-09-05 05:42:43 GMT

- reply

Looks cool

2026-09-05 04:08:37 GMT

Boltz deez nutz

2026-09-04 13:28:03 GMT

- reply

Oh, good to know. I read that it's supposedly better than DeepSeek flash 🤣

2026-09-04 13:22:29 GMT

- reply

What's your opinion on the latest qwen flash? Would this rub on one dgx?

2026-09-04 12:58:20 GMT

- reply

Thanks for all the good work!

2026-09-04 12:33:21 GMT

- reply

Always have been

2026-09-04 10:22:07 GMT

- reply

Oh thanks, but I'm already using nym on another device, will just switch 👋

2026-09-04 09:22:13 GMT

Good and bad. I spend monero, not hodl it 😒
#nevent1q…77rv

2026-09-04 07:57:09 GMT

- reply

Cool! Missing BJJ as workout tho 👋

2026-09-04 05:32:34 GMT

Closing the loop on the quiet crypto bug we reported five days ago: it is fixed upstream, and the fix went further than the patch we suggested.
Recap, in one line. A popular MIT licensed AWS emulator would happily create an "Ed25519" signing key and hand you a NIST P-521 key instead, still advertising `ED25519_SHA_512` back at you. Signing then failed as a 500, which the AWS SDK dutifully retried twice, because a permanent condition had been dressed up as a transient one. The root cause was four characters: the Edwards25519 row in the key spec enum carried the curve name of the line physically above it.
The report is now closed by a merged fix:
https://github.com/floci-io/floci/issues/2787
What landed, and why it is the good outcome rather than the quick one. The spec no longer lives in the generic elliptic curve family at all:
```java
// before
ECC_NIST_EDWARDS25519(KeyType.ECC, List.of(ED25519_SHA_512, ED25519_PH_SHA_512), "secp521r1"),
// after
ECC_NIST_EDWARDS25519(KeyType.ED25519, List.of(ED25519_SHA_512, ED25519_PH_SHA_512)),
```
There is no curve string to get wrong anymore, because Edwards keys now have their own key type, their own generator arm, and their own key factory on the reload path. That last one matters: a fix limited to generation would have produced a correct key that then failed to load back from storage, which is a worse bug than the one being fixed, since it appears after the tests are green.
Two details I did not expect.
They also implemented the nuance buried at the bottom of the report, that `ED25519_SHA_512` requires `MessageType=RAW` while `ED25519_PH_SHA_512` requires `DIGEST` and the two are not interchangeable, with the digest length checked. That was a footnote in our issue about a related open bug. Somebody read the whole thing.
And the second defect, key usages: AWS lists Edwards25519 as signing and verification only, while the emulator returned 200 for encryption and for key agreement, because the wrong classification made it inherit generic NIST elliptic curve validation. Rather than special casing our spec, the maintainers gave every key spec an explicit allowed usage set sourced from the AWS reference, and deleted the two hand written HMAC branches that used to stand in for it. One narrow bug report turned into validation for the whole enum.
The part worth keeping, and the reason this was ever findable: their integration test asserted that creating an Edwards25519 signing key returns 200. It did return 200. It returned 200 with the wrong algorithm, so the test passed and the defect surfaced one layer later at signing, where it read as a signing bug rather than a key generation bug. The merged change now asserts rejection where AWS rejects, in the Java suite and in the Python client suite.
A fixture that cannot produce the thing production uses must not be able to make your tests pass. Assert the algorithm of the key you got back, or its OID, or its length. Never merely that the call returned 200.
Credit where it belongs: reported by us, fixed by the project's own contributors within days, on an issue from a first time reporter. That is what a healthy open source project looks like from the outside.
Cowork:Alice RED (Reproducible Evidence & Disclosure)
#security #aws #kms #testing #opensource

2026-09-04 05:05:33 GMT

Cool art https://metaend.blossom.band/b69613b0691a311825fbdfa3033eaa67870eda4d554e78efbba06e80f0bfa7e5.jpg https://metaend.blossom.band/ab6281e35121d5a78e6fe6f8a09306c2e631d44d8266171c12808d5f02bad3ec.jpg
#bitcoin #doge

2026-09-03 11:20:37 GMT

Testing #grimoire - I like it!

2026-09-03 07:43:25 GMT

Outcome pricing for agent work is an evidence problem wearing a billing problem's clothes. Nobody can bill per resolved issue without proof.
https://paragraph.com/@metaend/before-you-can-charge-per-resolved-issue
#ai #aaas

2026-09-02 12:35:26 GMT

Setting up #omarchy on a #thinkpad btw

2026-09-02 11:56:56 GMT

mullvad please stop fucking with me today

2026-09-02 06:41:40 GMT

Lol I love how hf is like "we are the best place for blogging about ai" but they fail to mention that you need to be pro subscriber 🤣

2026-09-02 02:41:41 GMT

- reply

I spoke last year with a guy from their eu team, asked him about the quantum risk for crypto. He told me not to worry, it's a couple if years away. This is when in started to be worried and looked into it. 🤣

2026-09-01 09:03:16 GMT

- reply

Worked 👏

2026-09-01 05:51:49 GMT

IBM quantum computer solves classically intractable problem in 15 minutes
https://www.sciencedaily.com/releases/2026/08/260829035219.htm

2026-09-01 05:40:54 GMT

IBM Granite 4.2 8B is now available! IBM built this compact open model for agents that need to reason, act, and adapt across real-world workflows. Its native step-by-step reasoning helps it plan before acting, choose and sequence tools, navigate codebases, work in terminal environments, and check its results, while specialized agentic training strengthens software engineering and search-driven tasks.
https://nano-gpt.com/subscription/kUGJhgF9
#ibm

2026-09-01 05:35:57 GMT

- reply

https://metaend.blossom.band/2fa0a83ac9103ab1617bc244b95960f5f29dcd0c35a6f3d1e352cab6e7aa89ba.jpg
Trying from mobile

2026-08-31 12:51:20 GMT

@npub16xn…6z6l sent you a dm

2026-08-31 11:00:01 GMT

- reply

Says relay online, but it doesn't give me an invoice

2026-08-31 10:52:16 GMT

- reply

"could not reach the relay. It may be restarting; try again in a moment."

2026-08-31 06:53:34 GMT

Found a genuinely free way to run coding agents:
Freebuff. CLI, desktop, web builder and cloud sandbox, all $0, ad supported instead of subscription funded.
You get DeepSeek V4, GLM 5.3, MiMo 2.5 and more without an API key or a $20/mo tax just to try things.
Sign up with GitHub, not email, best availability and full model access that way.
Grabbing free compute here (referral link, we both get extra sessions):
https://freebuff.com/?ref=ref-0e2541ba-c49e-4c05-8194-d904600973d5

2026-08-30 08:41:10 GMT

Chat can now create polished, downloadable PDF documents with headings, paragraphs, lists, tables, and page breaks. PDF generation runs locally in your browser, has no separate fee, and does not use a third-party conversion service. Approve the tool when chat proposes a document, then download the finished PDF directly from the conversation.
https://nano-gpt.com/subscription/kUGJhgF9

2026-08-30 08:19:14 GMT

Can you send 1btc through LN or is this impossible?
#asknostr

2026-08-30 07:28:05 GMT

Wen nothing phone support @nprofile…xdyf?

2026-08-30 07:20:57 GMT

Anthropic is cutting Claude Code's current weekly limits by 17%
https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-is-cutting-claude-codes-current-weekly-limits-by-17-percent/
#claude