Drawback: You need different silkscreens for different amounts of RAM.
Btw, if you're worried about RAM-swapped units you can already verify them as described at https://geekworm.com/blogs/news/prevent-ram-swapped-raspberr...
--
(§) as can be seen here https://pip-assets.raspberrypi.com/categories/1129-pcn/docum...
There are already many Jev-like models in there.
Edit: No affiliation. Just found it and thought others might find it interesting.
For emails, I get 95% accuracy with this method, with only 50-100 examples for training
Training the model takes less than 5 minutes on a CPU
The resulting model is <1MB, and inference is sub 100ms
Some other cool things about this approach:
* the model doesn’t train on some “ideal” or general classification, instead it learns your preferences
* the model runs on pretty much any mobile device and can be retrained online on the device
* privacy, the whole training and inference is 100% local, no data goes anywhere (except whatever you feed codex/claude while building the model)
Note: to do a more general test, I made a classifier for the Banking77 dataset. The model is <10MB, trains in <30s on CPU and gets 94.5% accuracy, which puts it in the top 5?models by accuracy for that set (the best one is at 94.86%, but it’s 350MB in size and takes hours to train on a GPU).
Jev came in, and added that magic of "you dont need to train your classifier or determine the weights" if you dont want to, and just get the classified answer out. I think that's what is making people see this with a glitter in their eyes.
I’ve dismissed it every time. I don’t mind being shown that ad once. But it bothers me that they don’t care enough about my experience using their product to provide a “never show this again” option.
Or just use some metrics to know that if I dismiss it 50 times then maybe I’m not interested.
Yes I see that as an ad. Do you not? Does anyone not? And if I'm on the highest paying ad free plan, what are they promoting to me?
I’m all for consumer awareness but I’m begging everyone to stop freaking out over prosaic non-issues like this.
* Ads in YouTube feeds for other google products and services.
* Ads underneath videos for products from the channel owner.
* Sponsorships within videos from the channel owner.
* Advertising overlays (supported IN THE APP BY GOOGLE) for products and services from the channel owner.
* Email advertisements for Google products and services.
* Community post advertisements from channel owners which show up in the YouTube feed.
I contacted support to enquire and they state these are not considered advertising.
Why would anyone give them the benefit of the doubt?
[0] https://lawcouncil.au/international-law/ils-insights/tangled...
If you pay to avoid ads, you are merely letting them know that you have disposable income to spend on this sort of stuff. You're doing their job for them by segmenting yourself into the upper echelons of the market.
At some point, some shareholder value maximizing CEO is going to show up and notice how much money he's leaving on the table by not advertising to all of those people full of disposable income.
I have actually cancelled my subscription because of this.
Snowden did ask them not to get into any direct military secrets that were in the leaks, so that could explain some of the restraint.
While I heard many of the headlines the reporting is very in depth, including a bunch of smaller details worth poking into. I'd recommend people who are interested and who haven't done so yet to just read a couple of these.
Through a combination of blackmail, personality clashes, vanity and ego, incompetence, the whole thing just fizzled out completely. A real shame.
At the time, look at the situation of Chelsea Manning, Julian Assange and the many whistle blowers before Snowden? How can u blame him?
Edit: Am unable to reply to comments. HN don't let you comment 2-3 levels in or is it moderation? - Fixed.
Also, please refer to a reply to a similar comment: https://news.ycombinator.com/item?id=49781990
A lot of what sounded extraordinary in 2013 is now baked into everyday discussion like metadata and mass surveillance. He is also available less for interviews because he is in Russia.
Basically bigger fish to fry
We're in the phase where the factions are geographically sorting, identifying allies, and arming. Then we fight it out. Each faction thinks their particular Overton window will win and become dominant afterwards, but likely we'll just end up back in the Stone Age and there won't be much of a society left to have a mainstream.
This is only because immigrants are willing to do jobs at wages and working conditions that locals do not accept. Employers in the Finnish berry industry were just jailed for human trafficking. Meanwhile the same companies have been loudly complaining that no Finn wants to work for them.
Mmmm... It seems in the world of politics, esp. populist politics, more facts don't bring better policies. They are simply ignored, or accused of being lies pushed by the ennemies.
> "We have all the facts on this we need. We don't need any more facts. In the land of truth, my friend, the man with one fact is the king."
-- Linton Barwick, In The Loop (2009)
this is insane. the "speciesist" people aside (the best way for all other species to flourish is for humans to eradicate themselves right now, which is completely mad), i have the opposite problem: the _axiom_ seems to be: we should help ME flourish. "Me" as in people who are raking in trillions for their own very special selves right now, at the cost of everyone else's future, while none of them can be trusted to hold my cell phone for a second.
> if an industry commits to the axiom of helping humanity flourish
where does he see such industries, outside of maybe nonprofits?
Unfortunately, mathematics (especially pure mathematics) is by its very nature very, very poorly understood by those who haven’t worked as a mathematician. Even worse, those who don’t understand are seemingly not at all aware of their misunderstanding and are entirely confident in their (very wrong) characterisation of the subject.
It's about many things, but perhaps the most relevant idea here is that no information matters without understanding. We could generate all possible knowledge, but unless someone--a human--can verify and understand it, it doesn't count. The cure for mortality could be written on the moon, but if no one reads it, it hasn't really been discovered.
[1] https://maskofreason.wordpress.com/wp-content/uploads/2011/0...
Yes, it was developed by Google employees, that does not imply it has the full backing of Google, or Deepmind, or GCP. Notably, the website doesn't seem to claim this either.
The same Google that pulls plugs on a whim?
I've been still just like, making VM's with proxmox, then putting my agent in the machine and letting it run free (with my dotfiles setup script making dev env pretty much free, though I could also just make a VM snapshot). What's wrong with that? Is that not the scalable solution for enterprise rn?
That said, I think Google's ADK ecosystem and this new AX platform is promising--I would expect Google to maintain this and other tooling around this for years to come.
To the Googlers out there: is Google using this at any capacity for internal projects?
> We want to make dealing with agentic infrastructure easier so you can focus on your work. AX is designed with an uncompromising focus on ergonomics, rapid iteration, and joyful workflows for both application developers and AI researchers.
On the the other hand, the readme quickstart section says
> You need a Kubernetes cluster, ko (brew install ko), a container registry your cluster can pull from, and a reachable Agent Substrate Control API (in-cluster default: api.ate-system.svc.cluster.local:443).
Call me old-fashioned but I don't find this "easier". Maybe it's easier in the same way that Kubernetes itself is easier than managing VMs and container deployments at massive scale without such a tool. But there's a vast chasm between what this tool is being sold as and what it actually is.
btw its the same google that has already killed its "gemini cli" and re-introduced it in the form of "antigravity cli"
These comments are even more annoying. Sides that are winning don't have to constantly tell their opponents that they're losing. I have no idea why you and so many others seem to have attached their ego to the use of a tool that you feel the need to attack people who have complaints about the output it's often used to generate.
It's one thing if your code is really truly "co-authored by" Claude. It's another thing if it isn't even really co-authored by you.
Evidently the majority of people see no problem with LLM prose, but me and many others think it is some of the most annoying crap possible. Please consider speaking in your own voice.
I know this gets repetitive, but it is important.
Re: why, I don’t think most people understand the very basics of the global economy in mechanical terms, and this was my attempt to explain those mechanics. I wanted the various pieces of the system to be motivated by understandable problems, hence the fable-like story.
I now see some folks were triggered by this stylized approach. Which is a pity. I think the simple setup is worth the payoff. My explanation of money creation, for example, matches the Bank of England’s whitepaper, and I am especially pleased with how the idea of a reserve currency both develops naturally and rhymes with earlier ideas lower in the hierarchy.
The lack of ASML EUV machines certainly hurts, and pushing DUV so hard results in abysmal yields of good chips, but you can compensate by running more wafers or making smaller chips, and the net result is that Huawei's Ascend production volume is limited by CXMT's HBM capacity not processor dies.
The problem is that HBM manufacture requires many steps (die thinning, via drilling, plating, alignment) where the equipment used by everyone else (Samsung, SK Hynix, Micron) is also blocked by sanctions, so the Chinese are having to develop all of this themselves too, which they have, but yields are currently low, even when using shorter HBM stacks.
And the worst thing is that is the best possible strategy for them. It's essentially win-win for everyone but consumers.
- Buyer (AI) has crazy money, so will pay whatever
- Seller doesn't have to build anything new, as the buyer is willing to pay whatever
- Generate ridiculous profits from crazy money
- No oversupply risk in case of reversal
- Return from producing HBM to DDR5 in a single quarter if reversal does happen.
Hold on to your existing hardware, people, and be on lookout for your local deals.
"Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway."
https://gowers.wordpress.com/2026/09/17/why-i-didnt-sign-the...
"Nothing is worse to the demise of a society, than people who want to convince you that the cat is out of the bag and will not go back in, while the cat is being violently shook out of the bag at the same time."
> The mechanism is standard adtech. What has no precedent is running it on an AI chat product.
As someone who has been well aware of this mechanism for quite some time, I still feel icky anytime I re-read the details of it.
What a time to be alive.
why not use your own words? If you are gonna ai generate this blog, just post the prompts instead.
Firefox, Brave and Safari do. Chrome and Edge do not.
There's an empirical observation that models often have a single direction in their activation space for "hmm no I shouldn't do this". It forms naturally during pre-training, and is then surfaced during post-training to make the model refuse to engage in certain behaviour.
With a little bit of linear algebra you can zap that direction from the model's activations, and it stops refusing to do things. You can also do the opposite: magnify that direction, and the model refuses to do anything at all.
Orthogonalising activations at runtime is computationally cheap. Just distribute the refusal vectors (few thousand floats per layer), then run against the stock weights. Antirez's DS4 already supports this: https://github.com/antirez/ds4/blob/8db1d1d155cb0400a86a86b9...
Abliterated weights are just a bad habit we've gotten into. It's also deeply suboptimal from a precision point of view to take a model that's already been QATed and distributed in pre-quantised form (DeepSeek V4, Kimi K2.5 or K3...), modify its weights, and re-quantise it. Similarly, abliterated models regain some of their refusal behaviour when they're re-quantised after abliteration -- avoidable by keeping the two separate.
Positives
• It's a heck of a lot smaller than Qwen-Image 1 (20b parameters) at only 7b, making it one of the smaller open-weight models available (Z-Image Turbo is one of the few that is smaller at 6b) when compared to Ideogram, Krea2, Flux2, etc.
• It supports native transparency (Qwen's team, as far as I know, is the only one attempting to tackle this). Even though it's relatively trivial to set up background removal postprocessors, it's also neat to see it natively supported.
• It's fast using QwenImage2.1 convrot, a 1MP image took around ~5 seconds on an RTX4090.
Negatives
• The license (assuming you respect it) is far more restrictive. The original Qwen Image 1 was released under the standard Apache license; this one explicitly forbids commercial usage without obtaining a separate license. On the other hand, a lot of us didn't expect the Qwen team to ever release "weights-available" ever again.
Qwen-Image 1.0, released about a year ago, only scored 4/15 on my GenAI Showdown Benchmarks. Since that time, they've been upstaged by Krea 2 (6/15) and Ideogram4 (8/15). I'll post the new results once I have some more time to run them.
If AI labs get to ignore licenses, so do we.
https://en.wikipedia.org/wiki/Qwen#List_of_models
Unfortunately, it looks like this model is using a much more restrictive license:
Qwen-Image 2.1 is definitely a pretty big leap over the last open-weight version, Qwen-Image 1.0, released back in August of last year and managed to score 7 out of 15 as opposed to its predecessor which scored 4 out of 15.
Even though it's significantly smaller, 7b vs 20b, it's multimodal (so you don't need a separate image-to-image model like you did with Qwen-Edit), more coherent, and significantly faster even when outputting at higher 2K resolutions. However, in my testing, I found that I had to play with dialing up the CFG depending on the complexity of the prompt.
I've also added a progress dropdown under Model Performance so you can see how cloud vs. local models have been trending since 2024. Spoiler: June of this year released some of the biggest bangers (Krea 2, Ideogram 4, and the kind of slept-on Boogu-Image 0.1).
Downsides:
- It was clearly trained on at least some level of synthetic training data, and it shows in some of the subpar outputs in terms of fidelity. Some of this you might be able to iron out with a refiner model downstream or a custom LoRA but time will tell.
- They've moved away from the permissive Apache license. Commercial usage is only allowed by request.
Comparisons:
https://genai-showdown.specr.net
If you just want to compare local models only:
This has been a big part of the job for anyone on a team for at least 20 years. I do agree that it’s the hardest and worst part of the job, and has now become the majority of the job for anyone who isn’t vibe coding. So, that sucks.
However, I just don't think that's realistic. It's asking an author to suddenly become an editor. It's asking somebody who writes code to now read and debug others code.
It can actually be harder to find the the bug in a tricky piece of code than it can be to write your own correct code from scratch. I see AI introduce all sorts of bugs all the time in my personal projects that I would never introduce, and would never think to test for, especially around anything graphical.
It was supposed to launch in 2018, then was pushed to the early 2020s on a Russian rocket. For obvious reasons, it got pushed again, now launching in 2028.
The state of the world isn't great for space exploration, but I'm hopeful this mission will be revived at some point in the future.
I wonder what their official explanation for this behavior is.
For an industry that’s stagnant in progress yet relies on new frequent releases to survive (non-progress being an existential risk), this could make sense.
I have no idea if that’s what’s happened, I completely pulled it out of my butt. And I have no idea is the actual frontier is stagnating.
AI companies should be subject to the OWM like any other company that sells a product that varies in weight. Perhaps when a sane administration is re-elected; one that can read history books and comprehend why our regulations exist in the first place. Or have even a semblance of respect for its citizenry.
Exactly what I am saying for months now. And it's exactly the reason why I am shifting to open weight models now. Just bought myself a 2x DGX Spark Cluster. Will run Qwen3.8 Flash Next on it, maybe Qwen4 when it comes out.
Not only do I have full control over quantization and inference, but also will I experience a constant level of quality. It won't be frontier. But it will be stable, and that's enough reason for me to switch. Also I will likely save some money on subscriptions.
and so I leave this comment as off topic, but, is it pretty easy to get your new app to appear in that ad slot, above the main contender?
Between something opensource and trusted, and a proprietary potential backdoor that can send the data back to the Zuck himself you'd probably want to choose the former (cause you never know with closed sources).
I'm glad that he has shared the link as for me it's definitely useful. It's not a competition and I don't care what media talk, that's not about money.
Why would I want a bluetooth scanner to phone home? Seems untrustworthy. Nearby Glasses wins from a data safety perspective.
It was an amazing book.
There was an article on HN that one city's police force, when there were protests going on, took to playing Disney songs from loudspeakers; so that any recording of their interactions would be impossible to publish on any significant internet platform (or at least, the audio would be muted, by IP protection filters).
Decided by whom?
This feels like astroturfing. The amount of new users recommending and hyping this proprietary product seems odd.
There's no human to appeal to in social media platforms. It's automated, algorithmic enforcement; and when it's wrong (or someone exploits it to be wrong, like this), you're powerless to do anything. It means nothing if you're technically in the legal right: the platforms are risk-averse and don't care about your false positive.