jmward01

If 'distillation' means training on outputs then what is the legal concept of ownership of outputs? And, more broadly, is this something that could be skirted by doing it in different countries that have different legal structures? Basically, are they saying they own those outputs, not the companies that paid for the tokens, and only they can train on them? I suspect a lot of companies are saving their token histories and using them to fine tune internal models.

That is my problem. What can you trust anymore? Is there an appliance out there that can do the basics and not be a malware gateway?

Honestly I have nothing but anger for all parties involved here. We need to treat any appliance as hostile. Vote for people that will stop this stuff. Don't buy it. This is digital assault and should be treated as such.

If any of the HN crowd reading this are tech reviewers, make privacy and security a first class feature of the things you review. If it has a network connection then ask hard questions of the manufacturer about how they are managing that massive responsibility and what promises they have about how they will, and won't, use it. Ask what qualifies a 'trusted partner' and get a list of them so you can dig into them and report on how much those partners can or can't be trusted.

I really don't care how bright the screen is if this type of stuff happens. I'd rather watch nothing.

Finding a different way of thinking about a problem often leads to a breakthrough. This is what an ecosystem in nature shows us, that diversity matters in finding hard solutions. I think the great thing here is we are getting a chance to find whole new ways of thinking about problems that were hard. I suspect many old problems will fall because of it and, hopefully, some really new interesting ones will replace them.

You prepare for the 9.0 earthquake that can happen once in 500 years because it is a 9.0 earthquake and if you haven't prepared your society is dead, not because you think it will most likely happen in your lifetime.

It isn't about storing the data you send/receive. It is about having a model live through millions, or billions, of tokens doing the job. The current state becomes very important. Especially when you think about how interacting with these things works. No system prompts, just keep working with it until it is good at its job and then using that state as a starting point for other jobs. picking a bunch of different states as starting points, etc etc. You didn't use a simple system prompt to get there, you used interaction and examples and just like a person you decided they were good enough at some point and let them start doing the job for real. It is a totally different system than current models use.

So, I've been working on infinite context models (think fixed size state with a few tricks) and I think this will eventually lead to a kind of lock-in by vendor. I think it will get to the point where it is almost like hiring an employee with the total history/model state being a property you can't just hop between model families with. Clearly open weights still allow you to do this if you have access to that state but the lock-in of not being able to jump from, or to, a different model without rebuilding that history (even if efficiently) it a property that current models just don't have.

I think there is real value in going smaller/limiting resources. The trend is 'just make the weights bigger and throw more data at it'. It is a MBA's view of winning. We have a knob, keep turning it. It does work but it may not drive as much creativity as resource limits can drive. It is like urban growth boundaries in city planning. If you aren't allowed to 'just expand' you are forced to build more intelligently inside the city and those creative solutions often lead to major improvements.

It may seem extreme, but we need laws about digital breaking and entering/assault that cover things like this. It is being done specifically to harm you and without your consent or knowledge. That is assault. We just, as a society, need to start naming things what they actually are and treating them that way.

I generally think AWS is better than GCP and azure, but them not allowing spending caps is a big worry source for me and something that has made me pause and rethink using them. A bad click or a bad actor can create tens of thousands of dollars of spend nearly instantly and they can, and will, bill you for it. I can understand that stopping services is hard but some system would be good. For instance, if they had a two tier system where you could stop new services and active things like EC2 would shut down (but not delete) if spend is > x, that kind of thing. Some sort of 'stop the bleeding' concept would give me a lot of piece of mind using them.

Hmmmm... I wonder if there is a caesar cipher font or other substitution cipher fonts out there to actually obfuscate the data. So you use this to display to a user text but the unicode is re-mapped so that a is pointing to unicode g for instance. You remap the text to display correctly but contain massively swapped around unicode. Of course cut and paste would be a killer here but it is a price to pay for poisoning training data I guess.

I see a future HN post about how someone vibe coded HN to filter the AI stories. HNAI (Heck No AI)

I didn't see cost. If cost is similar to flash drives then this could be massive. Every GPU ships with a lot of ram still but for AI inference and games you have 10TB of this stuff to stuff all your textures into and all your model weights in. For fine tuning models this would also be great if using lora or similar.

AR will happen. Exactly what that future will look like I don't know. The one thing I do know though is that meta, the same company that decided you can be AI injected into an ad without informed consent, has no business paving the path to AR adoption.

Chin's emissions rose by 20% from 2017 to 2024 and then flat-lined in 2025 with emissions expected to decline in 2026. That is what decarbonization looks like, the rate of change has swung massively in the right direction. The .3 in 2025 will likely turn into -1.5 in 2026 and more after that. They have embraced alternative energy because it is now making them money and gaining them influence. That trend won't change and they will keep on decarbonizing because it is the easy, profitable, route.

You conveniently left the US out of that and the US's role in driving global emissions. China is also likely rapidly de-carbonizing right now just like they rapidly spun up. They are an issue, but the US and its policies have a massive impact on this problem. Same with Europe. Rapidly adopting renewables and not externalizing our emissions to other countries would go a long way.

I have seen the argument shift from 'That's a lie. Global warming doesn't exist. Stop pushing regulations for a non-existent problem.' To 'Maybe it does, or there is some human impact but this is all within normal variation and the climate can take some amount of pollution so stop trying to regulate it!' to, I guess the new argument of, 'It's too late so why try? We shouldn't change course because it won't matter so stop trying to regulate the problem away.' It isn't too late. Do something.

Reminds me of the boss key in a lot of 80's games.

Is this even a fight that is possible to win here? Run enough functions and between timing comparisons (x takes 2.5 y) and rounding (things like this) and I suspect you can nail os/exact machine and possibly even other tasks running on that machine. I'm not sure there is a viable way to stop this. At best just make it a little harder? Society and legislation need to catch up here. It is like a lock on my door. Locks don't stop people. They just deter a few people and delay a couple more but a determined person can (likely very easily) break into my house. That is why we need society to call it out that it isn't right (people avoiding buying things they think are stolen, shunning those that do it, etc etc) and laws that step in (it is a crime to break into my house. real resource dedicated to tracking down and stopping people that do it, etc). A similar approach needs to happen here. It should be illegal to track a person like this and people that get hired at a place using the gains of it should shun those companies and society should shun those companies.

it becomes very hard to have a common culture which is the absolute basis for a high trust society.

I think a large set of problems have come from this statement. We have, over time, become more intolerant of other cultures which is directly related to trust. Excluding the other isn't the path we need to take. We need to get back on the path where we can include other cultures.

These are often not disclosed until you start. Day 1 'paperwork' that drastically changes the agreed upon terms of employment well after you have left a previous position. In some cases they are even forced on you after having worked somewhere for a while and future employment is predicated on them. They are evil. I almost quit my job over this, but of course how can you afford to not work while searching? And searching while out of work is far harder than while still working. We need penalties on companies that do this, not just a ban. 'Sign this or you are fired' documents after you are hired should have repercussions.

their main mem is LPDDR5x with DDR5 SODIMMs for expansion. LPDDR5x is 256GB/s? on the machines you see it implemented on which is a lot faster than the DDR5 expansion. For comparison, gddr7 on a 5070ti pumps ~900GB/s. 4x the tokens/sec if you are memory bound.

It seems like there is a very health space for an MOE targeted GPU where it has essentially an 5070ti with 16gb ish GDDR7 but then also has 128 GB LPDDR5x (or even just DDR5 as expansion dimms on it?). Putting this into the same card would likely reduce the transfer hit when a cache miss happened and the gpu had to load from the slower LPDDR5x. No need to have PCIe 5x16 limiting memory transfer if it is on the card. MOE models could then get near native performance and even models where the active parameters + context didn't quite fit the thrashing would be less of a problem. Not UMA but gets the UMA 'lots of system memory to play with' benefit.

All new laws should be given a trial period where the lawmakers are forced to live with them for 90 days before the public is subjected to them. At any time during that period lawmakers can change their vote.

This court has had little respect for precedent so maybe an argument here is more about the fact that rulings like this one may become more likely.

The weapons use saddens me. Ignoring that, this is interesting for electric aviation as a whole. One of my arguments about electric aviation is that burning fuel for propulsion has a lot of challenges, especially at altitude. Pushing the speed envelope for electric will lead to more and more novel ways of pushing things and once they get efficient, fast and, very importantly, high altitude, the amount of power you need drops considerably. The argument against electric aviation is always that (pick your favorite fuel x) has 100x the energy per gram. This is an apples to bananas comparison for many reasons but at a high level if you need far less energy to cover the same distance then that argument goes away in a hurry. Electric propulsion with aircraft redesigns taking advantage of it has a chance to do that. I'm really looking forward to hitting FL1000 and getting to my destination in an hour using electric propulsion.... eventually.

I realize the article is pro buttons. I think a huge thing missing from the button discussion (well, maybe lightly touched on in the article) is that physical buttons and controls help guide without looking. Other buttons give feedback that your hand is in the right place. Sure, at first contact that (very bad) reference radio is worse than the touch screen but within a few days of using that I would not need to look to make sure I was hitting the button I wanted because I could feel the face of it with my hand and know I was hitting the right button. So basically, even though the paper picked essentially the worst radio on the planet, it would likely be better than a well designed touch screen after just a few days of use. First day though? That thing is a nightmare.

A first version is always going to be low in scale and problems will of course be found. The question isn't 'are there problems', of course there are, the question is 'are the benefits here worth more than the problems'. If you already have the land, a maintenance operation, a user of energy, etc etc then integrating solar into that system has a lot of strong benefits that can make the problems manageable. Start looking for the upside here and you may actually find the cost benefit equation balances out, especially if this scales out.

This is a great example of the solar ecosystem finding new niches. Tracks are likely good because there is already constant maintenance and inspections on them as well as ready hookup to the grid or, as the article mentions, directly providing power for trains. The point here isn't that this is 'the most efficient way to use solar' it is that it works and provides benefit. It is like agrovoltaics. The solar production is lower than a dedicated solar installation, but the dual use of the land and potential secondary benefits make it worth while. It looks like this could even get to the point where solar could actually power the trains completely assuming this expanded a little beyond just the track surface which could be interesting for the design of trains.

Cooperation leads to higher value for everyone. We are now only getting the maximum minimum value because we are actively destroying cooperation. In other words, borders are stupid. When they go away you see prosperity and reduced tensions. When they go up you see inefficiency and distrust. Borders aren't the result of distrust and economic issues, they are the cause.

There is a valid debate but the example I gave clearly coerced the consumer. They had paid for something with the expectation of use and then were hit with a requirement to give consent after the transaction. We shouldn't let some grey area prevent us from stopping the ongoing harm. One side has clearly been abusing the other. If a law goes just a little to far in favor of the consumer I think we can all agree that is better than letting the consumer be completely abused without protections. You don't let an attacker keep punching their victim because we gotta get the laws perfect to act to stop them. Act and reduce the harm and then adjust to get the balance right.