Hacker News: jmward01

New comment by jmward01 in "Volkswagen started blocking GrapheneOS users"

jmward01 — Wed, 17 Jun 2026 15:59:56 +0000

I want a law that requires publishing your API for apps like this as well as allowing users to crate their own frontend based on it. That would enable more privacy aware versions of these apps.

New comment by jmward01 in "SubQ 1.1 Small"

jmward01 — Wed, 17 Jun 2026 03:27:35 +0000

Yeah. That is the plan I think I have settled on. I'll release something interesting here shortly but the full architecture, including all the multimodal input/output streaming is something I am considering my options on. I may even try to get to the 1-2b moderately well trained model stage and host it to show how transformative cached states are compared to cache tokens.

New comment by jmward01 in "SubQ 1.1 Small"

jmward01 — Tue, 16 Jun 2026 18:55:11 +0000

Yeah, looks like fun stuff. You still need to preserve the entire kv cache though right? So even if compute is drastically less, memory keeps growing. The system I described keeps memory constant (well, if you keep the entire token history you technically are gaining one long of data per token generated but I think we can agree that is negligible and could be capped at something high like 1B or so with no meaningful impact). I think I will probably release trick one and see if people then believe trick two even without seeing it.

New comment by jmward01 in "SubQ 1.1 Small"

jmward01 — Tue, 16 Jun 2026 18:48:38 +0000

As a follow-up, I can see there is not a lot of belief which is why it is also hard to find a company to partner with on this. So, how -do- you make money on something like this as an independent researcher. Maybe I release trick one, show how guided window attn (and nn memory and probably a lot of robotics) can be trained? Thoughts? I can do that pretty quickly. By itself that is a pretty great tech (combined with fixed windows of full attn it is pretty amazing). The second trick, I think, is a bit more powerful although both are general purpose. If I do this, think people will believe trick two (and all the real time multi-modal streaming stuff)?

New comment by jmward01 in "SubQ 1.1 Small"

jmward01 — Tue, 16 Jun 2026 16:54:36 +0000

Yeah. I am about to the point of just releasing it all. I love the tech. It does amazing things. But I want to move to the next big things I can see doing with it and building the custom ops to get it to work efficiently is a pain. I am positive others would run with it and make it all way better which would free me up to do more.

New comment by jmward01 in "SubQ 1.1 Small"

jmward01 — Tue, 16 Jun 2026 16:42:02 +0000

Well, I know this is possible because I have built things that work just like it is promising to do. The two key technologies needed are:

- guided window attn. Predict where to attend to but in a fixed window. If you do this to just the token/vocab you can keep effectively unlimited context and perfect recall. (yes, I can do that. There is a trick to teaching it how to predict position. This also immediately opens other crazy things like NN memory)

-efficient fixed state size models. So not a recurrent mechanism because that breaks training, parallelizable like transformers, but fixed sized state instead of unbounded attn. Pick a reasonable amount of state and it is amazingly good since it doesn't need to keep separating wheat fro chaff in context (yes, it is possible to build this, I have. It works. This also opens up real streamed models. I have a true infinite context streamed model I toy with locally that I am getting to be audio/text in and audio/text out in real time.)

Put those together and you have O(1) token gen, infinite context and perfect recall. It is a whole new world of models. You can interact with a model until you have it at the state you want and then save its state and use that as if it were your system prompt. Batches pack perfectly so inference is massively more efficient. Training is massively more efficient. Transformer and unlimited attn models are a dead end. But how do you make money on this as an independent researcher? If I release the Two Weird Tricks this is all based on I get zip and the big players get even more tech for free. If I keep it all secret I get Zip and eventually the tricks will be figured out. (Yes a little frustration here) If anyone wants the model architecture of the future make me an offer :)

New comment by jmward01 in "A backdoor in a LinkedIn job offer"

jmward01 — Mon, 15 Jun 2026 23:53:19 +0000

I'm not a lawyer but it would be odd if a government agency couldn't communicate a possible threat to a tech company. It is in a company like LinkedIn's best interest to set up a phone number/channels for a centralized agency to communicate potentially malicious accounts and other emerging threats. I suspect that actually already exists for big companies. I doubt they are required to -do- anything without laws but this seems like a win that is easy for all sides. The problem is likely mostly on the US (and other govt) side of things. No clearly defined agency with a clear mandate, resources and leadership to take on this task.

New comment by jmward01 in "A backdoor in a LinkedIn job offer"

jmward01 — Mon, 15 Jun 2026 22:14:35 +0000

I see several comments like this implying nothing can be done. But that is far from the truth. First, an agency that actually answered the phone could coordinate directly with LinkedIn and other tech companies to quickly take down these fake accounts and minimize harm to others. We all know how incredibly hard it is to contact a tech company. Second, an agency that answers the phone could help less technical people find what may have been compromised and push people towards support services if needed. And finally, maybe, they could do the hard job of combining leads and working with appropriate agencies to maybe find and prevent these things over time.

New comment by jmward01 in "A backdoor in a LinkedIn job offer"

jmward01 — Mon, 15 Jun 2026 20:36:21 +0000

So, this is a crime right? Why isn't there a well known '911' for cybercrime to report things like this to and get help? Society needs to catch up with the actual dangers out there and build support networks for this ASAP. This is organized crime and needs organized defense to deal with it.

New comment by jmward01 in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

jmward01 — Mon, 15 Jun 2026 19:01:42 +0000

Didn't realize they did this. I have avoided pushing data to huggingface. This is all -deeply- private info and I haven't really reviewed their privacy policies and the like. I'll give them a look.

New comment by jmward01 in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

jmward01 — Mon, 15 Jun 2026 18:41:26 +0000

Has anyone been storing their cc sessions for future training data on their own models? I'd love to build a system that fine-tunes on cc sessions and a good first step is capturing my own sessions well.

New comment by jmward01 in "Iroh 1.0"

jmward01 — Mon, 15 Jun 2026 18:02:38 +0000

I think I see the value prop here. Beyond its intended use, what about creating a full VPN out of it? This takes care of the hard part for a lot of home users, opening your vpn up in a safe way. I know this is solved by many other tools so this isn't a new thing but it may increase adoption. Is there already something like that? I imagine you have considered this and if it doesn't already exist have a good reason for not including it. If so, what is that reason?

New comment by jmward01 in "AI agent bankrupted their operator while trying to scan DN42"

jmward01 — Fri, 12 Jun 2026 15:45:28 +0000

AWS not having spending caps makes me -very- wary of using anything agentic on it.

New comment by jmward01 in "Oh good, screwworms are back (2025)"

jmward01 — Thu, 11 Jun 2026 19:09:03 +0000

Mistaking the initiation of something as the cause of something is a common mistake. A spark starting a fire in a bin of oily rags is technically the 'cause' but that thinking is clearly short sighted. In reality it was the bad practices that created the environment that was the true culprit. This is an example of the same. I suspect there are constant incursions in and near panama but a layered approach of constant study, monitoring, consistent prevention techniques, etc are what stop a surge in 'unchecked northward migration' from something that happens once or twice and is contained and corrected into something that happens constantly and isn't figured out until this has spread and established itself over thousands of miles. So, yes, the spark was 'northward migration' but the 'unchecked' was because we gave up in being good at this job. We de-funded monitoring and study. We put in place political leaders and not technical experts all because 'expert' opinions aren't important. That is what created the 'unchecked' and why we are here now.

New comment by jmward01 in "Oh good, screwworms are back (2025)"

jmward01 — Thu, 11 Jun 2026 17:55:16 +0000

A problem happening eventually is expected. The point of a good program is a layered approach that admits no layer is perfect so you have backups that kick in to minimize the impact of problems. So the problem was emerging in 2022, not great but not a tragedy. Cutting monitoring means we reacted slower and our inability to play with our neighbors well means that we can't coordinate a response quickly or as effectively. Destroying our layered, nuanced policies has real consequences and this is one of them.

New comment by jmward01 in "Oh good, screwworms are back (2025)"

jmward01 — Thu, 11 Jun 2026 17:44:13 +0000

New comment by jmward01 in "Oh good, screwworms are back (2025)"

jmward01 — Thu, 11 Jun 2026 17:32:44 +0000

Trust in science is so low in the US that this was bound to happen. The author's conclusion nails it:

But institutional failures like this aren’t just the aggregate failure of a number of irresponsible people, they are a failure of a cultural attitude that doesn’t demand excellence from everyone in order to get the job done.

I'll add though that when crisis happens the entity 'doing something about it' suddenly looks amazing, even if they were the cause. That means that it is a winning political strategy to create a lot of crisis so you can solve a few of them and look like a hero.

New comment by jmward01 in "An Ohio Valley 100k-watt FM signal is severed in broad daylight"

jmward01 — Sun, 07 Jun 2026 05:09:55 +0000

I highly doubt the people doing this look at crime and punishment stats before they do this. More punishment often just ends up costing society because courts and incarceration aren't cheap and no real rehabilitation so it often just makes the person do more bad things when they get out. I'm not saying 'no jail', but we do need evidence based criminal justice.

New comment by jmward01 in "PyTorch Custom Operation"

jmward01 — Sun, 07 Jun 2026 01:26:58 +0000

I'd love to se a side by side by side comparison of implementing as triton, cuda/c++, just using torch.compile, etc etc with a few example ops. I have broken out triton a lot for things but found that it is very hit or miss how much I will gain over just using torch.compile. Probably a lot of that is my skills and a lot is how much torch.compile can take together and optimize if raw pytorch is made available to it.

New comment by jmward01 in "New method turns ocean water into drinking water, without waste"

jmward01 — Fri, 05 Jun 2026 19:59:05 +0000

This is an interesting tech, but I have big doubts. In the picture you can see some salt coating the surface. Even just a little seems like too much for this type of system. I really hope they can make this work and scale this up.