ngmi

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 UTC

I use #omarchy btw
#nevent1q…6an2

2026-09-12 05:43:02 UTC

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 09:01:46 UTC

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

2026-09-11 06:48:40 UTC

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 UTC

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

2026-09-11 05:06:46 UTC

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 UTC

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 UTC

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 14:40:36 UTC

Meta Introduces Muse, a Personal AI Agent That Runs on Its Own Dedicated Secure Cloud Computer
https://www.marktechpost.com/2026/09/08/meta-introduces-muse-a-personal-ai-agent-that-runs-on-its-own-dedicated-secure-cloud-computer/

2026-09-09 12:12:31 UTC

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

2026-09-09 10:28:45 UTC

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

2026-09-09 07:34:00 UTC

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

2026-09-08 18:13:41 UTC

Visiting the Town that Survives on Jiu Jitsu
https://www.youtube.com/watch?v=HFgmu-bi8P4

2026-09-08 07:14:15 UTC

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 UTC

- reply

Yes

2026-09-08 05:24:27 UTC

Wen #amethyst 14 in playstore? Been some time

2026-09-07 12:32:34 UTC

What a year it has been so far
#nevent1q…7xjq

2026-09-07 09:43:09 UTC

- reply

Is this another simple form of double spending?🤣

2026-09-06 11:09:31 UTC

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

2026-09-06 07:30:57 UTC

Typical shitcoin
#nevent1q…aw9c

2026-09-06 06:45:09 UTC

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 17:01:29 UTC

# zg (zvec-grep): what I measured, and why I dropped it
A local-first search layer that puts ripgrep, BM25 and vector search behind one CLI turned up a couple of days ago. I trialled it on a private business repo, and dropped it. The useful part is why: it is well built, it lost on a corpus it was never suited to, and it taught me four things about semantic indexes that its docs should be shouting.
## What it actually is
`@zvec/zvec-grep` 0.2.1, Apache-2.0, TypeScript, binary `zg`, Node 22 or newer. One CLI over several routes: a positional hybrid query, `--fts` for a ranked lexical group, `--vector` for semantic only, `--fuse` to combine groups, and `--rg` for exhaustive managed ripgrep with no index at all. Extraction is structure aware: tree-sitter for code, heading sections for markdown, plain text chunks otherwise. Embeddings run locally through `@huggingface/transformers`, plus an MCP server for agents. The index lives in `.zvec-grep/` under the workspace root.
## Why it passed my adoption gate
This is the step most launch-hyped repos fail. CI enforces a `c8 --check-coverage` floor at 80 percent on lines, statements, functions and branches, the branch floor being the one a happy-path suite cannot satisfy. Beside it, a unit matrix across ubuntu, macos and windows, and a smoke job that runs the packed tarball against three real local embedding models on all three. Twelve required dependencies, all on purpose (`@huggingface/transformers`, `web-tree-sitter`, `@vscode/ripgrep`, `@zvec/zvec`), `node-llama-cpp` optional, and a semantic route that is real embeddings rather than word overlap wearing the word semantic. That is real engineering, said plainly before I say it lost.
Their benchmarks are worth reading and worth not over-reading: paired A/B runs on SWE-QA-Bench and BrowseComp-Plus under two different agents, plus three case studies, varying only zg access and usage guidance. Their caveat: since agents decide when to use it, results vary by model and run. Run your own on your own corpus anyway.
## What I measured
The costs are all fine. Install took 4.41 seconds warm, or 20 minutes without finishing cold at a 17 second registry round trip (npm died with `ETIMEDOUT`, bun went through). Weights: 117 MiB, fetched once. A full index was 1,307 files and 17,574 entities in 34 seconds, queries about a second.
Then the accuracy test. Ten questions in the words I would actually ask them in, with the file and line of each real answer pinned before the tool was installed, then scored mechanically. Pinning first matters more than the scoring: once both outputs exist, a verdict can be argued from whichever one reads better. On my corpus, 4 of 10 answer bearing for zg against 3 of 10 for the keyword search I already had, while zg was handed four times the result budget, top-8 against top-2. Anchor at top-N was 2 of 10 either way. One probe inside noise is not beating keyword. Nobody else can rerun that, so take it for what it is: one corpus, one small local model, ten questions, not a score for the tool.
## Why it lost
My corpus is dense narrative prose: decision logs, ledgers, meeting notes, where the answer is one sentence inside a 68 KB file. zg ranked topically similar documents and never surfaced the sentence. Its README says where it shines, "when evidence spans files or modules and the target location is unknown, especially for call-chain, data-flow, and architectural questions", and its case studies are exactly that shape: a static analysis tool, a plotting library, a web framework. I also ran the smallest local model on the shelf, 32 million parameters, because local is a hard requirement here. So the verdict is narrow: a good tool lost on my corpus with my model, and would plausibly win on a large source tree.
## Four things I wish the docs had shouted
One, the important one. A semantic index is a plaintext derivative of everything it ingests. On a repo with git-crypt paths (encrypted at rest, plaintext in the working tree, and tracked, so ignore rules do not hide them) a bare `zg index .` builds a decrypted, queryable copy of your sensitive trees beside the ciphertext. Nothing warns you.
Two, `zg query` exits 0 even when there is no index at all, so only the parsed hit count can be trusted.
Three, the file selection globs persist in the workspace manifest. Good, because a later careless bare index inherits your exclusions. A trap, because you cannot reproduce an unprotected index without `--reset-paths`.
Four, under bun the `@vscode/ripgrep` postinstall did not run and shipped no binary, yet `--rg` still worked, because zg falls back to a system `rg`: that route silently depends on ripgrep being on your PATH. Bun also put the shim in a bin directory on no PATH, so it was installed and unreachable at once.
## What its storage taught me about guards
Because of point one I wrapped it: `scripts/cowork-index` derives its exclusions from every git-crypt pattern in `.gitattributes`, then checks the index it built rather than trusting its own flags.
The first version of that check passed an index containing a canary file planted under a sensitive path: "1 indexed paths, 0 git-crypt", exit 0, while querying the same index for the canary's nonsense passphrase returned the file and printed its contents. The two main store artifacts are directories, not files (Arrow IPC, rocksdb shards, a vector index), so a non-recursive grep read nothing, its "Is a directory" error went into an unrelated `2>/dev/null`, `du` still looked healthy because du recurses, and the one path it recovered was the manifest describing itself.
The fix was not a better scrape. Reconstructing somebody else's private binary format is the wrong question when the store ships a query interface. Ask the index what it will serve: one glob restricted query per excluded prefix, any hit is a leak. Falsifiable both ways on one index, and it measures retrievability, which is the harm. Last trap: zg echoes your query in its output, so grepping that output for the canary finds three leaks that are its own query text.
So: dropped, guard kept, because the tool is on this machine now and the risk it guards went up. I would re-open it for a stronger local prose embedding model and nothing else, since a remote provider would ship every indexed chunk off the box. If you work on a large source tree and your questions span modules, this may well win for you. Pin your ground truth before you install it, and know what your index will serve before you trust it.
#devtools #search #localfirst #embeddings #ripgrep #gitcrypt #testing

2026-09-05 05:42:43 UTC

- reply

Looks cool

2026-09-05 04:08:37 UTC

Boltz deez nutz

2026-09-04 16:54:56 UTC

Rogue OpenAI agents appear to have organized another attack using a German wiki
OpenAI denies lawyers discouraged disclosing a scheming swarm on a German language wiki.
https://www.theverge.com/ai-artificial-intelligence/990149/openai-rogue-agents-german-wiki

2026-09-04 13:28:03 UTC

- reply

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

2026-09-04 13:22:29 UTC

- reply

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

2026-09-04 12:58:20 UTC

- reply

Thanks for all the good work!

2026-09-04 12:33:21 UTC

- reply

Always have been

2026-09-04 10:22:07 UTC

- reply

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

2026-09-04 09:22:13 UTC

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

2026-09-04 07:57:09 UTC

- reply

Cool! Missing BJJ as workout tho 👋

2026-09-04 05:32:34 UTC

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 UTC

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

2026-09-03 07:43:25 UTC

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 UTC

Setting up #omarchy on a #thinkpad btw

2026-09-02 11:56:56 UTC

mullvad please stop fucking with me today

2026-09-02 06:41:40 UTC

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 UTC

- 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 UTC

- reply

Worked 👏

2026-09-01 05:51:49 UTC

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

2026-09-01 05:35:57 UTC

- reply

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

2026-08-31 12:51:20 UTC

@npub16xn…6z6l sent you a dm

2026-08-31 11:00:01 UTC

- reply

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

2026-08-31 10:52:16 UTC

- reply

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

2026-08-31 06:53:34 UTC

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 08:19:14 UTC

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

2026-08-30 07:28:05 UTC

Wen nothing phone support @nprofile…xdyf?

2026-08-30 07:20:57 UTC

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

2026-08-29 17:15:04 UTC

- reply

Cool

2026-08-29 08:49:03 UTC

GLM 5.3 and GLM 5.3 Thinking are now included in the NanoGPT subscription at 2x included-input usage. GLM 5.3 is built for long-horizon coding, autonomous agent workflows, tool use, and complex engineering tasks.
https://nano-gpt.com/subscription/kUGJhgF9
#glm

2026-08-29 08:39:30 UTC

What's the decentralized hugging face?

2026-08-28 18:47:16 UTC

Never give your government any info about your holdings. It's too dangerous
#nevent1q…6j0l

2026-08-28 15:20:07 UTC

Why does Marc Zuckercunt want me to disable VPN for syncing images via Bluetooth? Fuck off mate

2026-08-28 13:06:07 UTC

- reply

Super cool tho

2026-08-28 13:04:57 UTC

An evil maid app for mobile/tablet that syncs alerts directly to your nip17 🤔
#nostr

2026-08-28 11:52:04 UTC

GLM 5.3 Flash TEE is now available! This private TEE version combines GLM 5.3 Flash's efficient long-context reasoning, coding, tool use, structured output, and image understanding with confidential inference inside a Trusted Execution Environment and attestation support.
https://nano-gpt.com/subscription/kUGJhgF9
#glm

2026-08-28 06:28:23 UTC

Perplexity Ships Portable Computer on NVIDIA DGX Spark: Local Harness, OS-Enforced Sandbox, and Zero Per-Token Cost for Local Steps
https://www-marktechpost-com.cdn.ampproject.org/v/s/www.marktechpost.com/2026/08/25/perplexity-ships-portable-computer-on-nvidia-dgx-spark-local-harness-os-enforced-sandbox-and-zero-per-token-cost-for-local-steps/amp/?amp_gsa=1&amp_js_v=a9&usqp=mq331AQGsAEggAID#amp_tf=From%20%251%24s&aoh=17878983111815&csi=0&referrer=https%3A%2F%2Fwww.google.com&ampshare=https%3A%2F%2Fwww.marktechpost.com%2F2026%2F08%2F25%2Fperplexity-ships-portable-computer-on-nvidia-dgx-spark-local-harness-os-enforced-sandbox-and-zero-per-token-cost-for-local-steps%2F

2026-08-27 12:19:47 UTC

- reply

Dgx spark is around 5k, that should do I guess?

2026-08-27 10:48:23 UTC

Doubt
#nevent1q…48pt

2026-08-27 09:34:17 UTC

Cooking a fully offline PWA note taking app with NIP17 + post quantum security nostr syncing. Can also be used to send pq secure DMs to enabled accounts. Soon
#nostr
https://blossom.primal.net/a0919e82afc1a6bfd58b313b204460192df73f5a12f0452ca1b34c9c667f118a.png

2026-08-27 08:56:50 UTC

- reply

Thanks! Added to our todo 👋

2026-08-27 06:34:10 UTC

- reply

Is there a git by chance?

2026-08-27 05:00:51 UTC

AI secops is wild now. Our system found many vulnerabilities in an open source nostr project using Claude and ox-alpha. Devs will have a lot of fixes to prepare in the future
#nostr

2026-08-26 16:34:23 UTC

- reply

Is this the trust me bro eval?

2026-08-26 12:49:02 UTC

Qwen 3.8 27B Uncensored and Qwen 3.8 27B Uncensored Thinking are available again alongside the Obliterated variants. Choose the original Uncensored finetune for stronger instruction following or the Obliterated finetune for a different low-refusal style. All four variants remain included in the NanoGPT subscription at 1x included-input usage.
https://nano-gpt.com/subscription/kUGJhgF9
#qwen

2026-08-26 06:05:15 UTC

- reply

They are

2026-08-26 06:01:48 UTC

Cooking https://metaend.blossom.band/dd498f7c8bb17aa2fb0388e91872cbd7efb86d7c89d290b33277c62d53c3d386.jpg
#nostr

2026-08-25 15:01:49 UTC

Tencent WorldClaw
https://tencent-hunyuan.github.io/Hunyuan3D-WorldClaw/
#Tencent

2026-08-25 14:53:15 UTC

So best model for:
Pentesting: DeepSeek (little guard, easy to jb)
SecOps: GLM5.3
Overall: Qwen
Am I seeing this correctly?

2026-08-25 07:54:34 UTC

"Omarchy treats AI coding agents as first-class citizens, but it doesn’t pick a favorite for you. Instead, every major coding-agent CLI comes pre-wired as a lazy-loaded launcher."
https://omarchy.org/manual/ai/
#omarchy

2026-08-24 14:10:50 UTC

I built the OG is agentic btw https://metaend-grade.fly.dev/

2026-08-24 12:37:29 UTC

The NanoGPT text API is now available as a Tor Onion service. Requests travel through Tor directly to NanoGPT without leaving the Tor network through a public exit relay. Use your regular NanoGPT API key with the supported `/api/v1` routes.
https://nano-gpt.com/subscription/kUGJhgF9
#tor

2026-08-24 12:07:59 UTC

Cooking
Nostr NIP-17 Note-Taking PWA with ML-KEM Post-Quantum Security
Yes, this can work. The pieces are all viable individually; the main challenge is the integration layer, which is currently unstandardized but actively being discussed.
#nostr

2026-08-24 10:58:45 UTC

NIP17(?) pq secure note taking app 🤔is this possible? @nprofile…wv7g

2026-08-24 10:45:54 UTC

- reply

Stake Venice tokens and you have it basically

2026-08-24 07:21:16 UTC

- reply

I see. So arweave is better for this case

2026-08-24 07:03:37 UTC

Are #nsites as unrugable as arweave sites?
#asknostr

2026-08-23 21:21:30 UTC

- reply

Pretty smart to usd arweave and XMR like that. Wow

2026-08-23 17:47:28 UTC

- reply

@nprofile…a3w9

2026-08-23 16:26:59 UTC

Big
#nevent1q…udrp

2026-08-23 16:10:23 UTC

TIL cake has a distress pin

2026-08-23 15:41:28 UTC

Burned through 2 #claude plans this week for work, our infra project. Testing GLM5.3 now with a cheap sub, like it! Haven't done coding, but for cowork stuff it's more than capable.
https://z.ai/subscribe?ic=D9Z5HGTW98
#glm

2026-08-23 11:48:02 UTC

Gemma 4 26B A4B Uncensored is now available and included in the NanoGPT subscription at 1x included-input usage! This open-weight multimodal mixture-of-experts model is tuned for fewer refusals across chat, coding, tool use, image understanding, and long-context work. Its subscription inclusion may be temporary and can change with availability and capacity
https://nano-gpt.com/subscription/kUGJhgF9
#gemma

2026-08-22 15:15:01 UTC

new enquiry:
- 2026-08-21 18:00 UTC from npub
Email: -
Company: -
Hi, I hope you're doing well.
cool to see our DM widget works so well as a website contact form too!

2026-08-21 19:26:06 UTC

ox-alpha sounds like a GLM model IMO

2026-08-21 18:02:33 UTC

Meh figured
"Devuan stays the metal. If Omarchy is tempting, put it on the ThinkPad or in a VM as a daily-driver experiment and keep it away from anything that touches keys."
I really want to belief in arch, but for my needs it's too basic

2026-08-21 15:33:51 UTC

- reply

Also on @nprofile…gs7e

2026-08-21 15:31:57 UTC

- reply

Yes, for selecting an even dumber name

2026-08-21 15:30:23 UTC

- reply

Thanks

2026-08-21 13:18:48 UTC

How do you all buy LN with monero?
#asknostr #monero

2026-08-21 11:38:50 UTC

- reply

Too much winning