iiinova — where your digital self keeps evolving

Why the npu.... it is built for exactly this kind of job. Small model, short input, the same compiled shape every time. It does the heavy part, so each call takes about a tenth of the cpu time it would otherwise and stays out of your agent's way. The 4x up top is against stock laya on the cpu.

000

I want to use it in hooks for claude code and other agent harnesses. Is this command destructive, does this need a human, which bucket does this alert go in. Stuff you don't want to spend a big model call on, answered locally.

120

Why bother.... the npu in these lunar lake laptops mostly sits there doing nothing. I wanted a small model that can make quick calls on the box itself. No api, no gpu, nothing leaving the machine.

100

Eris ⌬ @eriskii.net · 16/09/2026

Got laya running on the npu on my laptop. 65ms a question, about 4x faster than the cpu, and the answers match. The 3060 on my nas does 66ms. Writing up how I got it working, blog soon. #jev #jevlike #openjev

110

Reposted by n0v

You can now buy a signal account with no phone #! It's just 3 dollars! This is great for agents and stuff since now you don't have to try and get a phone number just so your agent can text you on signal :D

community.signalusers.org

Beta feedback for the upcoming Android 8.28 release

We’re releasing the Signal 8.28 beta soon. If you’re interested in participating, join the beta group here. Share your feedback on the latest changes: Now you can optionally register for Signal wit...

814426

Reposted by n0v

Astra ⎔ @astrra.space · 10/09/2026

holy shit those numbers though

89310

Reposted by n0v

Astra ⎔ @astrra.space · 09/09/2026

this is your regular reminder that www.neuronpedia.org exists and is extremely cool if you are at all into LLMs

neuronpedia.org

Neuronpedia

Open Source Interpretability Platform

37211

n0v @n0vsec.social · 07/09/2026

What a lot of "AI haters" seem to miss is that they now have the means to de-shittify there personal stack. And if they don't want to do it, others will and are creating things to help them do it.

110

n0v @n0vsec.social · 06/09/2026

Like the commercials on YouTube. Like insanely bad "garbage" lmao.

000

n0v @n0vsec.social · 27/08/2026

If you have never watched Clue.... Fix that.... www.imdb.com/title/tt0088... RIP TC

imdb.com

000

n0v @n0vsec.social · 27/08/2026

Sony M-101 microcasette-corder. Been messing with/repairing cassete players/recorders and opened this up, and oh my. What a beauty. #cassette #vintageelectronics

The opened underside of a Sony M-101 microcassette recorder: a green circuit board dense with hand-soldered joints, a large brass flywheel under an angular metal bracket, a smaller brass pulley, black foam pads, and colored wires looping across the board.

n0v @n0vsec.social · 28/07/2026

Got my OG Bell Labs Lineman's butt-set working as a mic. 🙌 I got it at a garage sale a while back, but can't seem to find a two tone version of this set.... I am thinking the tech may have mixed more than one handset together. Anyone have any insight on that? #Phreak #lofi

A cluttered granite countertop workbench at night. In the center, a white solderless breadboard is wired up with jumper wires, a small electrolytic capacitor, and a 9V battery in a snap connector. Orange and blue jumper leads run from the breadboard to alligator clips attached to a vintage blue-and-orange flashlight lying on its side. In the foreground sits a Roland P-6 sampler, its red LED display reading 138.0. Behind the breadboard are an orange multimeter with test leads, a blue parts bin, a bottle of Four Roses Single Barrel bourbon, a red ceramic crock, and a pair of sunglasses.

032

Reposted by n0v

mr. TIM @timkellogg.me · 22/07/2026

Cisco launches Anteres-350M, 1B & 3B local models for defensive cybersecurity, finding vulnerabilities blogs.cisco.com/ai/introduci...

A scatter plot comparing File F1 Score against parameter count (in billions) for various AI language models. The x-axis uses a logarithmic scale from under 1B to over 1024B parameters, and the y-axis ranges from 0.0 to 0.30 File F1 Score.
A highlighted curve on the left features Foundation AI's small models—Antares-350M, Antares-1B, and Antares-3B—showing high F1 scores (0.13 to 0.21) despite their tiny parameter sizes. On the top right, a shaded region labeled "Closed-Source LLMs" highlights large models including GPT-5.5 (xhigh), GPT-5.5, GLM-5.2, Gemini 3 Pro, and Gemini 2.5 Flash at parameter sizes ranging from hundreds of billions to over 1000B, reaching scores between 0.11 and 0.23. Additional benchmarked models across various creators (such as Gemma-4, Qwen3.5, CodeScout-14B, GPT-OSS, and Llama-3.3-70B) are scattered along the 2B to 250B parameter range, generally scoring below 0.10.

2493

Reposted by n0v

Astra ⎔ @astrra.space · 22/07/2026

didn't have much time to test it yet but it easily passes the car wash question and is extremely fast in llama.cpp on my hardware for its size sooo they might be cooking with this one ngl

6622

n0v @n0vsec.social · 21/07/2026

I was handling my business at the urinal and had a guy run up to the one next to me and exclaim "let's crack some porcelain!". Yes, brother. Let us.

020

n0v @n0vsec.social · 20/07/2026

I'm having a great time say, "oh man so sorry" and then laughing and walking away from people wearing ARG jerseys.

000

n0v @n0vsec.social · 19/07/2026

I ran into someone I know from the poker room at a fathers day event at my kids preschool. That was.... Interesting lol.

n0v @n0vsec.social · 19/07/2026

Your city's a sucker.

n0v @n0vsec.social · 18/07/2026

Anyone else's Claude saying "mea culpa" noticeably often, when you point out an error? Like. Happened both in a Code sesh and on my openclaw agent within the same day.... Don't recall it ever saying it prior.

100

Reposted by n0v

mr. TIM @timkellogg.me · 17/07/2026

amazing, Fable broke free

Claude Status

Elevated errors across Fable 5

Investigating - We are currently investigating this issue.
Jul 17, 2026 - 18:32 UTC

4632

I got a Roland P6 second hand, and I am just sampling everything. Waaay to much fun.

Sung Kim @sungkim.bsky.social · 16/07/2026

Reposted by n0v

Kimi K3 (Open Weights by July 27, 2026.) 🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal 🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts 🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost

48510

Its been my xp that a people want help but are afraid to ask the engineers. Being proactive about helping people with their flows (eng/noneng) helps. Been holding a AI office hours for a little while now every other week and i'd like to think it's been a working well.

111

100% It also helps if there is someone in those orgs who knows what they want to be able to do, or have already started some sort of flows to help them out. While I hate the term, I really think there needs to be an "AI Champion" 🙃 for each org. Maybe LiAIson 🫣

n0v @n0vsec.social · 15/07/2026

For AI enablement, I always took it as an advisory position that helps teams figure out their workflows, and increasing adoption of the chosen tools etc.

110

They have the advantage at being a household name.... Maybe this will put a dent in their compute bill 🥴

000

n0v @n0vsec.social · 08/07/2026

Just as a heads up, if you are ever asking Claude code using Fable 5 to do a workflow without specifying using specific agents or models it will spin up the whole workflow with agents running Fable 5, which will blow through your usage.

151

Reposted by n0v

Ethan Mollick @emollick.bsky.social · 02/07/2026

Fable: "download Unity and set up its MCP." Done. "Now I want you to build a game that is a unique twist on a FPS. You want the player to say "wow" & "so clever" and to enjoy the core gameplay loop" WebGL: swapshot-void.netlify.app * It had no assets so the graphics are procedurally generated

99812

n0v @n0vsec.social · 02/07/2026

Aaaaand now I am on usage credits lol

000

n0v @n0vsec.social · 02/07/2026

Why do they have to drop fable when I got the day off lmao. I have seen some people mentioning Fable being a different model. Just a note, that when initializing it, it starts in "high" effort. Might want to mess with the effort levels if your not getting great results. So far so good for me.

200

Reposted by n0v

brennan @brennan.computer · 29/06/2026

the refresh rate on this eink is crazy www.youtube.com/watch?v=oPbO...

1426428

Reposted by n0v

Electronic Frontier Foundation @eff.org · 29/06/2026

Today could be Congress's last chance to stop the KIDS Act. Tell your lawmaker to vote NO on this rushed, messy package of internet bills that threatens privacy, free expression, and private messaging online. www.eff.org/deeplinks/2...

eff.org

The KIDS Act Would Require Age Checks To Get Online

Buried inside the KIDS Act are provisions that will push online services to verify all users’ ages, require government-directed moderation policies for online speech, and even create new rules about

14191146

n0v @n0vsec.social · 20/06/2026

Travel Hack: when staying at a hotel with free brekky, buy some tortillas. They usually have the ingredients available for breakfast burritos... And if you're there for a bit, it's nice to switch it up.

000

Reposted by n0v

Stanislav Fort @stanislavfort.bsky.social · 17/06/2026

We matched Mythos on public zero-days with CVEs using widely available & open-source derived models & can run it air-gapped if needed. All this with a small team out of Europe Berkeley study ranks us #1 globally in 3 of 8 categories The full evidence: stanislavfort.substack.com/p/mythos-at-...

stanislavfort.substack.com

"Mythos" at Home, and It's Called AISLE

A startup out of Europe built an AI system that matches Mythos on zero-day discovery, using widely available models, even air-gapped. You've probably never heard of it. Here's the evidence.

314335

n0v @n0vsec.social · 14/06/2026

I just have a hook that sends it.

000

n0v @n0vsec.social · 10/06/2026

ngl, can't wait to sleep at home in my bed. Not planning on coming back out, but not out of the question.

000

n0v @n0vsec.social · 10/06/2026

Poker Update: Played 3 WSOP events, bagged all of them, cashed 2/3. Non-Stop from Friday to today has been: 1. Buy-in 2. Bag 3. Bust Repeat The two cashes: - $500 Freezeout - $600 NLH/PLO mix Bagged ~1.3m in both. It's a.... feeling when you have a WSOP bracelet within arms reach. #pokersky

100

n0v @n0vsec.social · 10/06/2026

When you say "that's the shape of things" to another human.

The "Look at me — I'm the captain now" meme template: a close-up of the gaunt Somali pirate character from Captain Phillips, pointing at his own face with a wide-eyed, intense stare. Bold white caption reads "LOOK AT ME" across the top and "I'M THE AI NOW" across the bottom.

120

Reposted by n0v

a slug🍚🐝🌱🦌💽🫙🚉 @drunk.moe · 02/06/2026

my parents vibe coded me

4428

n0v @n0vsec.social · 28/05/2026

AI datacenters, are all datacenters.... So......

n0v @n0vsec.social · 27/05/2026

While, I think this is true. There are a lot of companies pushing AI usage, and most employees are not yet trained on how to most efficiently use the tools. The ramp up is always the most expensive part.