iiinova — where your digital self keeps evolving

Reposted by Stuart Gray

FOI docs: UK police's six-month live facial recognition trial in London railway stations scanned 500K+ faces, leading to no arrests and one false-positive alert (Daniel Boffey/The Guardian) Main Link | Techmeme Permalink

0118

Reposted by Stuart Gray

Hetan Shah @hetanshah.bsky.social · 29/09/2026

‘After the incident with Usman, Robb told Muse to stop giving his address. To test the bot, he asked a few friends to see if Muse would still share it. “And it literally gave my address out to five people,” he said.’ 😱 www.theguardian.com/technology/2...

theguardian.com

Meta’s AI agent Muse gives out user’s home address without permission, sending buyer to his house

The new AI agent, Muse, was released last week and has been downloaded by 3 million users

716278

Reposted by Stuart Gray

Astra ⎔ @astrra.space · 22h

well, it was fun while it lasted

5602

Stuart Gray @sgray.bsky.social · 21h

Interesting timing with the delay of their latest model, and calls for a pause - OpenAI are also choosing to signal the beginning of the end for subscription token discounting. Their models have def. been getting cheaper. The big question is how much lower can API prices go sustainably?

010

Reposted by Stuart Gray

Zach Weinersmith @zachweinersmith.bsky.social · 28/09/2026

I keep having discussions with mathematicians, artists, etc., where they insist that we should make sure people are valued for being humans, not just their maximum economic contribution or whatever, and I keep wanting to shout... come visit any get-together of the special needs parent community.

727727

Reposted by Stuart Gray

Rich Harang @rich.harang.org · 28/09/2026

New from NVIDIA: OpenShell open source sandbox is solid, but the Sentry component is -- IMO -- the killer app here: out-of-band proxy for all traffic, including the LLM. Sentry \approx complete hardware isolation between agent and policy enforcement. developer.nvidia.com/blog/nvidia-...

developer.nvidia.com

NVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring | NVIDIA Technical Blog

To understand where agentic AI stands today, consider the last seismic shift in technology: the rise of the internet in the 90s. It was new and full of possibilities. You could build a website over a…

641

Reposted by Stuart Gray

Roland Smith @rolandmcs.bsky.social · 28/09/2026

'Which taxes do you want to raise instead?' should be the default question to anyone who calls for reductions in immigration.

14432151

Reposted by Stuart Gray

Astra ⎔ @astrra.space · 28/09/2026

experts being somewhat fungible in MoE LLMs is peak like what do you mean “if you don’t need the non-cached expert badly enough you can just use a cached one that’s close enough” is both a valid optimization option AND basically doesn’t affect accuracy at all???

10914

Stuart Gray @sgray.bsky.social · 28/09/2026

Kind of puts the Police (lack of) response to the recent Patriot Patrol actions into perspective if reports of a “large group of hooded and masked men” was all it took for a rapid response.

000

Reposted by Stuart Gray

Dan O’Sullivan @osullyville.bsky.social · 27/09/2026

Today I saw the new 4 hour documentary Musk, directed by Alex Gibney. There were too many times to count that a sold out theater burst out laughing at Elon Musk, something he said or did, even sometimes just how he looked. I believe that is his worst nightmare come to life.

14194912187

Reposted by Stuart Gray

Ethan Mollick @emollick.bsky.social · 27/09/2026

It is strange how much LLMs turned out to be able to solve such a wide range of hard problems that would not, instinctively, seem to be problems that a model of human language would be able to solve This is from a Stanford project that let Astra drive a robot in a kitchen tml.stanford.edu/homebody/

2339744

Reposted by Stuart Gray

Natasha Walter @natashawalter.bsky.social · 26/09/2026

It’s much much harder now to be an anti racist than it is to be a racist. That’s one of the things I learned from talking to Rachel Marsh, who has been doxxed and abused for standing up to the far right. www.thenerve.news/p/rachel-mar...

thenerve.news

‘I see what’s happening to Rachel; it scares me’: the woman who dared to face down Piddington’s anti-migrant storm

Natasha Walter talks to Rachel Marsh, the resident of the newly ‘independent’ Oxfordshire village whose anti-racist BBC interview led to abuse, doxxing and death threats

8289112

Reposted by Stuart Gray

Turkey Dancer @turkeydancer.bsky.social · 25/09/2026

No surprise there. Script on GitHub to scape the same fucking data the agent accessed. So much for a scary AI hack.

0161

Reposted by Stuart Gray

mimir @mimirwolv.bsky.social · 25/09/2026

arxiv.org/pdf/2412.00099 local model understanders might appreciate this :)

arxiv.org

0273

Reposted by Stuart Gray

Sam Harsimony @harsimony.bsky.social · 24/09/2026

First (independently-verified, commercially available) battery above 500 Wh/kg! We're gonna get (short-range) flying cars guys. www.youtube.com/watch?v=1ty1...

youtube.com

This Anode-Free Battery Just Broke Every Record (500Wh/kg)

YouTube video by Ziroth

78112

Reposted by Stuart Gray

Roland Smith @rolandmcs.bsky.social · 24/09/2026

Basingstoke police station.

10218848

Reposted by Stuart Gray

Roland Smith @rolandmcs.bsky.social · 24/09/2026

So they're now talking about attacking the system of policing and law. Arrest these f*ckers. Stop bowing down to them and dancing around them.

1051192342

Stuart Gray @sgray.bsky.social · 24/09/2026

Apple has published a paper & model based on Qwen 3.5 9B for processing large amounts of multi document text, compressed as images, LensVLM 9B: huggingface.co/bartowski/Le... github.com/apple-aiml-r... arxiv.org/abs/2605.07019

arxiv.org

LensVLM: Selective Context Expansion for Compressed Visual Representation of Text

Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map f...

1413

Reposted by Stuart Gray

James Ball @jamesrball.com · 23/09/2026

I think once black clad men are interfering with rescues at sea, we are past "legitimate concerns". Once they're attacking boats, and risking the lives of everyone on board, it's just thuggery. The Home Secretary's statement is on the side of criminals. There's no other way to read it.

501556415

Reposted by Stuart Gray

Nick Fisher @hydroxide.dev · 23/09/2026

This doesn't look like an official release, just a random jev implementation atop diffusiongemma via cloud run?

112

Reposted by Stuart Gray

Grace @gracekind.net · 23/09/2026

New animation from Opus 5.5!

6649584

Reposted by Stuart Gray

David Mimno @dmimno.bsky.social · 22/09/2026

It's possible for Jev/Laya/Decision Models to be not that big a deal as tech and massive as a new paradigm. Here's why I'm really excited from an NLP history perspective (thread)

17619

Reposted by Stuart Gray

Saadiq @saadiq.bsky.social · 22/09/2026

I updated an old email triage workflow for a client using Jev this weekend. It decides which emails to forward to her customers. Jev now runs alongside GPT-4o mini, which still controls forwarding.

OpenRouter token prices as of September 21, 2026. Jev is highlighted in blue at $0.042 per million input tokens with no output charge. Teal highlights mark models with comparable results on specific tasks, with vendor and independent tests distinguished. GPT-4o mini is marked as the client's existing classifier. Input prices use a logarithmic scale.

142

Reposted by Stuart Gray

Unsloth AI @unsloth.ai · 22/09/2026

Qwen-Image-2.1 can now run locally on 12GB VRAM with Unsloth GGUFs! The 7B parameter model performs on par with Nano Banana 2.0. Run our Dynamic FP8 or GGUFs for higher quality via diffusers, Unsloth Desktop & more. GGUF: huggingface.co/unsloth/Qwen... Guide: unsloth.ai/docs/models/...

1407

Reposted by Stuart Gray

Emma Yeomans @yeomans.bsky.social · 22/09/2026

Worrying scenes in the Channel today, for all interested in asylum policy and maritime tactics. A small boat from Normandy spent more than 30 hours at sea. Those on board now on their 2nd rescue vessel as Border Force forced to play cat and mouse with far-right activists who have their own boat.

1042175953

Stuart Gray @sgray.bsky.social · 22/09/2026

I’m only an hour through it, the the recent LatentSpace interview with TypeSafe founder Diogo Almeida is well worth a watch. His passion & enthusiasm alone is incredible. m.youtube.com/watch?v=cFx9...

m.youtube.com

Why I couldn't build Jev at OpenAI — Diogo Almeida, TypeSafe Co-founder & CEO

YouTube video by Latent Space

121

Reposted by Stuart Gray

Lynn Cherny @arnicas.bsky.social · 22/09/2026

Fantastic benchmark of Jev vs its "clones" from HuggingFace: DecisionIndex huggingface.co/spaces/multimo...

4365

Stuart Gray @sgray.bsky.social · 22/09/2026

The main issue with this model is general capability. It’s very small and fast, but its tiny size means it’s not smart and needs fine tuning for your specific use case. Jev being so cheap means the cost of fine tuning Laya may not be worth it, even if you have the necessary data.

091

Reposted by Stuart Gray

Adina Yakup @adinayakup.bsky.social · 21/09/2026

XiaomiMiMo just released 2 SoTA models. One might be the new BEST open model yet🔥 huggingface.co/collections/... Both are: - Sparse MoE + 1M context + MIT licensed - Native omni: text/image/video/audio

1181

Reposted by Stuart Gray

Johannes Kleiner @johanneskleiner.bsky.social · 20/09/2026

How can we test for consciousness in infants, non-human animals, and AI? I'm super happy that the recordings of our last workshop on exactly this question are now online! Speakers and links to talks below! 👇 #ConSci

13014

Reposted by Stuart Gray

mlf. ⎔ @mlf.one · 18/09/2026

fucking love people i know run a plex server go "nooo AI is stealing creators' intellectual property"

413611

Reposted by Stuart Gray

Finfet @finfet.sh · 17/09/2026

okay this dgemma -> "jev" patch is really obvious (and cool) after reading it a bit more. discrete diffusion models predict by denoising a canvas of random tokens. rather than providing only random tokens, seed the canvas with a template and put single random tokens in the response positions.

github.com

[Core] structured generation mode for DiffusionGemma model (Jev-like) by mmastrac · Pull Request #57250 · vllm-project/vllm

Prerequisite PRs A few PRs that can go in on their own merits were split out: [Bugfix] DiffusionGemma: hand out stashed logprobs only on the committing step #57414: hand out stashed logprobs only ...

231

Reposted by Stuart Gray

Karsten Konrad 🇪🇺 @zabong69.bsky.social · 18/09/2026

Well, that didn’t take long. And its probably like this - Jev is an LLM based on diffusion, trained to the max on synthetic classification tasks. Without reasoning and tools, what we have is a fast cheap and often dumb predictor. For which I’d still like to see real use cases.

032

Stuart Gray @sgray.bsky.social · 17/09/2026

Lucky that their are no ports, marinas, or sailors in the Portsmouth area that might need the RNLI services! (sarcasm) For context, on the night of disorder, several of the racist idiots used boats to get to the location, and one had to be helped out by the same RNLI after experiencing problems!

020

Reposted by Stuart Gray

Will Jennings📉🗳️ @drjennings.bsky.social · 17/09/2026

The original reporting of what happened in Portsmouth was a joke. This is just a hate mob that has been whipped up who are now targeting the RNLI. Throw the book at them.

independent.co.uk

RNLI closes Portsmouth lifeboat station after huge anti-migrant protests

RNLI volunteers were branded ‘traitors’ after bringing migrants ashore

27556173

Reposted by Stuart Gray

Stephen Turner @stephenturner.us · 17/09/2026

Microsoft: "capability laundering" - a weaker unaligned model splits a harmful task into harmless-looking subquestions, asks a frontier model each one separately, combines the answers locally. CBRN attack chain raised mean rubric score from 62.3 to 83.1. arxiv.org/abs/2609.153...

1152

Reposted by Stuart Gray

Parody Nigel Farage @parodypm.bsky.social · 17/09/2026

Steve Reed on 25 March explaining that the £100k cap on donations "will therefore apply retrospectively, so it includes all donations from overseas electors received from today". Yes, but how are people who never bother going to Parliament supposed to know that?

451475525

Reposted by Stuart Gray

Pete Fraser @petefrasermusic.bsky.social · 16/09/2026

In the movie version of this, after Piddington runs out of food water and money as a result of leaving the UK, the residents apply for asylum and end up housed in a military base, with HILARIOUS consequences. www.bbc.com/news/article...

bbc.com

Piddington in Oxfordshire votes to leave UK over asylum plans

Piddington residents voted over proposals to house up to 1,256 male asylum seekers nearby

1663527835

Reposted by Stuart Gray

James MacGlashan @jmac-ai.bsky.social · 15/09/2026

Rogue all-powerful AI predictions are not scientific and should not be trusted. People struggle to accept that because this prediction comes from some AI scientists. Sadly, scientists often make bogus unscientific predictions that should be ignored. Let's use another AI prediction as an example.

1123

Reposted by Stuart Gray

Stella Biderman @stellaathena.bsky.social · 15/09/2026

have never seen an AI model come up with an amazing idea that I hadn’t already thought of for a problem I’m working on. And the typical idea quality is garbage. If this isn’t your experience I’d love to see examples! Post them or DM me.

5161

Reposted by Stuart Gray

Pekka Lund @pekka.bsky.social · 15/09/2026

It must be hard for those who are still in denial about how powerful and significant AI is to hear it even from people like Obama. Oh well.

Barack Obama @BarackObama

I was encouraged this week to see the leaders of the frontier labs agree on the need for them to slow down the pace of AI development. Given the stakes, it’s a good and necessary first step.

But I’m even more encouraged by the growing recognition that how this powerful new technology develops should be at the center of our public debate.

I’ve been watching the progress on AI for over a decade now, and one thing that’s clear to me is that the potential impact of this technology is not overhyped. It’s also moving at lightning speed – and even faster than those who are engineering it can keep up with.

I’m not an AI accelerationist who believes it will lead to some techno-utopia, and I’m not a doomer who thinks it will inevitably lead to humanity’s destruction.

But whether this technology results in amazing breakthroughs in medicine, energy and education or unleashes huge economic disruptions, greater inequality, and potential catastrophe will depend on the choices that we make right now – choices that should be made not just by the companies involved, but by all of us.

61038

Reposted by Stuart Gray

Kevin Mitchell @wiringthebrain.bsky.social · 14/09/2026

Yes, we already have artificial autonomous entities that are not aligned with human values and that we've allowed to get out of control - they're called corporations

47321

Reposted by Stuart Gray

abadidea @0xabad1dea.infosec.exchange.ap.brid.gy · 15/09/2026

anyway, here is a map from NOS of the identified sabotage locations. It's interesting that they *don't* include the Amsterdam-Den Haag-Rotterdam coastal corridor. #netherlands #publictransport #trains

map showing several sabotage locations on train tracks in the Amsterdam-to-Utrecht-Arhem-Zwolle reigion (this is mostly inland, with the Amsterdam location being the only one along the major coastal route)

124