PaulKeeble

I think those in poverty have an awful lot less ways to supplement bought food these days, they are largely legislated out of foraging in any meaningful way. So I think actually a large percentage will starve to death or be in prison.

Or use something Kodi as the OS and just have a TV like interface. You can also run a Jellyfin client on linux or with Kodi and have that as a reasonable interface for the TV too.

Those viewpoints are always there in society and many people just waiting for the leader who will harness to come. It is one of the hard pills to swallow about humanity that authoritarianism is a popular movement brought forward with cheers and votes.

Feels like the whole industry of technology has been driving this direction for a while, removing ownership and replacing it with making everyone rent from them.

I don't think many people were even aware that there was a market betting on mortgages. People could see that rates were low and people were getting mortgages that couldn't pay them but the enormous credit market betting on it and bundling them together was something only the finance people and the ultra wealthy knew about.

This one everyone can see it, its very public, it has daily enormous public marketing from all the sellers of AI in the press.

The bigger issue is not that the direct AI companies go down in this bubble, its that they likely take Nvidia, Microsoft, Google, Intel, AMD and other big tech companies with them, hardware and software that are hugely intertwined with the global economy and many other businesses and underpin the modern computing world. Letting all them go bankrupt will have a huge impact. The problem with AI is the people exposed to it are not just AI companies, existing big tech has overinvested in it.

What I find they do badly in code is building abstraction and breaking the problem down in a way a human does. They don't seem to really build simplifying components unless you expressly tell them too, they are more than happy to just nest 10 deep or embed a 20 line called back function and use lots of shortened names. I find I can improve the code just by organising better and choosing some abstractions that make sense to the humans and then ironically you save tokens too because the AI will reuse it, but it wont create most of it itself. Its bad for domain work unless you are explicit about it.

I just can't seem to stay interested and read LLM text. It just seems to lack going somewhere, its information density is often really low and its full of odd isms that I don't find very interesting to read. I don't have a good radar for detecting the various LLM particular speech patterns but everytime I have an issue with an article there is always people talking about how its clearly AI written.

I find the same thing with written results out of AI "research", it is a slog to read through it and most of it can be skipped even if it can sometimes be useful its an unenjoyable read.

One key aspect that most people don't seem to be aware of is that there has been a lot of research in the cancer causing impacts of Covid infections. Its been implicated and mechanistically shown to cause a number of cancers and Covid hangs around in the gut for years/potentially ever we aren't yet sure in some people and it acts like a phage. As a phage it infects certain bacteria in our gut and drastically changes the microbiome, pushing it towards colonies of bacteria that are implicated in colon cancer.

The scientific evidence isn't there yet for colorectal cancer but its being investigated in studies at the moment and we are expecting a mass disabling event due to Covid and Long Covid and a big increase in a variety of conditions including cancer. Its one of the most studied viruses in human history and it does a lot of bad stuff and most people are catching 1.5 times a year.

We need a step change in the amount of silicon chips produced. This situation has been going on now for over 2 decades, initially bitcoin but ever since various high compute scenarios keep causing GPU shortages and now RAM and storage as well. This one isn't likely to go away anytime soon so we need a drastic increase in production to bring pricing down to reasonable levels.

Owning your email accounts with your own domain is just a good idea given the state of things, even if you ultimately just deliver it into Gmail or whoever. It gives you the ability to spend an hour and have all your email appearing with someone else in the event that a companies algorithm goes haywire and cancels your accounts.

I have tried the Thinking cap and the Swift model for QWEN-3.8 27B and my experience was its not as smart and capable, the results on the same task were less impressive. For the amount of tokens saved it might very well be worth it but I wanted to run the xhigh precisely because I wanted the peak result. I haven't tested this but I suspect it ends up much the same trade off, in theory it does well in the benchmarks but in practice on your own tasks its not as capable.

Seems like this is more about the introduction of LLM code reviews that has found edge cases that would have otherwise been missed and fixed as they appeared over years in the future. The issue is that the volume of issues has been beyond what people could do in time.

--cache-type-k q8_0 --cache-type-v q8_0

would drop it from fp16 to 8 bits. You don't want to go below 8 bits if you can possibly avoid it and if you really do need to squeeze just drop the v value to q4_0 but leave the key alone. Unlike model quantisation this really makes things worse quickly.

Ideally you want to find something that allows a context window for at least 65k for agent work. Over 100k is better but 32k is about the bare minimum for anything in a harness/agent like Pi. You might have to drop to a lower quantisation to make that possible but context is the key for longer lots of thinking questions like this.

If you drop the thinking level to low or even off the context will go further at the cost of the models intelligence. That might also help you get some answer but since such a question will read so many files into that context its the wrong approach. This is one of the bigger limitations of local LLMs, you can't do work that spans across a lot of files because you are almost always context limited. There is a limit of 1M even with the frontier so even that can't review an entire project. Focussing in on individual files or small groups is more achievable.

So just to be specific the -c needs to be >65536 and depending on the size of what is in the folder might very well not be enough.

Locally the MOE based models get a lot more out of the hardware than a dense model. Qwen 3.8-27B is the current golden child if you have >24GB VRAM but because its a dense model it is pretty slow to run and requires a lot of hardware. A lot of the guides are very much focussed on getting the dense models to run but in practice the MOE models with few active parameters are just faster and easier to run without too much loss of intelligence, they behave a lot more like their peak size than the active parameter size.

RAM is going to be the big bottleneck for MOE (at least for now) as it will determine the quantisations you could run.

On a 3080 10GB and 32GB of RAM I can run the qwen-next coder model and that is a 127B model and it gets 40-50 Token/s with 65k context. That is a lot more usable than a Quantised Qwen3.8-27B in my experience both because its faster and more capable.

I am looking forward to Qwen4 soon, I am hoping we get some more 27-35B MOE models, those are going to be really good for local hardware.

As HN has gotten bigger its pulled in other communities and now its not very developer/hacker oriented anymore. Lobsters was the first push to have the deep software talk back and there will be more as HN caters to a different crowd nowadays.

There is also a substantial saving in household appliances from all the EU laws as well that have pretty substantially cut power consumption, including a lot of computers and devices at idle.

The UK household has quite significantly cut its power due to all these power saving innovations, far in excess of the actual drop because industry has increased its percentage of the mix over time.

I don't disagree, the UI is quite bad in places. Some of its straightforward but most of its not. Just adding a WAN interface is a lot harder than it needs to be with their new abstraction you end up making all these weird abstract pieces. The one that really irritates me is the replacement of the routers DNS with your own, like a pihole, that is really bad at the moment especially since IPv6 and IPv4 are in quite different places and a different style of input.

But its also really feature rich and with a bit of effort you work it out and you have that longevity and the security issues fixed long term, much further than the manufacturer is willing to support it.

I keep looking at the openrouter token charts and its exponential growth. Month on month they out out more and more tokens and critically the quality of those tokens keeps increasing as well. The charts are led entirely by the chinese open weight models like deepseek v4.1 flash and Z.AI's GLM-5.3 with Claude and GPT down in 6th and 7th place.

So while I think a number of these companies are never getting their training costs back AI as a thing is growing really fast right now, in the space of a year openrouter has 100x their token creation.

I only wish more of them were actually mainline openWRT so the updates from the project were guaranteed. There is a lot of these devices that run a custom version and never have their hardware support put into mainline. That isn't really much different to what all the major router companies do, they are all running OpenWRT underneith but with their own skinned interface over the top.

Slowly but surely I think modern mass manufacturing is going to move towards personalised mass manufacture. I think in the meantime the clothes industry needs to do a better job of sizing, that is providing more detailed measurements so they can be selected better. At some point I ought to be able to buy a pair of trousers where I have defined more than just waist and length and especially true for tops based on chest or generic sizing. We can do this a lot better, its not intrinsic to suits they are just worn enough that bespoke services exist in quantity.

The real ROI is the change in behaviour of people constantly surveiled. We know people change their behaviour when under constant scrutiny and this why the elites want it.

Its much harder to produce a good fundamental list of biology problems because we are no where near as far along, so many basic things are not understood. There are still many unknown unknowns and these particular problems are limited to the realm of verifiable in a lab, which rules out much of what makes biology interesting and hard to study, and that is the organisms themselves and the interactions between their cells. The petri dish differs from the rat differs from the human and that is a part of what makes biology harder to progress.

None of these is likely the root to the wide array of chronic illnesses we can't treat and those seem like the next step to really put a lot of research into given how many people suffer from them and where we are today they seem achievable with the right investment.

Part of the problem is that AI has been trained on all text, not just the internet but scientific papers and books of all varieties. AI style is almost certainly the average of that style approach but we use very different styles for different writing and getting it wrong grates, and AI doesn't seem to deal with these style differences well at all.

The problem is it will be detectable and detected and then the password demanded and a failure to provide it will result in charges. I think unfortunately the answer is the phone has to be plausibly empty when crossing the US border, they should focus their efforts on being able to store the full state of the phone remotely and encrypted and being able to restore it easily.

One thing that is I feel missed about uptime percentage when compared to on premise uptime is when the downtime occurs. Its far more impactful if its in the middle of the working day or during the busy period of shopping. A store that goes offline in the middle of black friday or in the run up to Christmas is harmed a lot more than some down time on a Sunday night/Monday morning at 3am.

One thing I have noted over time is a lot of these AWS, Azure et el downtimes is they occur in the middle of everyones day, millions of people are impacted by them. Same with github its getting in the way of work. Whereas when we hosted services on our own equipment the downtime was usually out of main hours. The percentages are in many ways the wrong measure of downtime because hours aren't equal in impact to businesses.

I am much more interested in reigning in the industry itself with top level regulation to protect human rights like privacy. Once the excesses of the industry and its thirst for personal data result in crimes for the executives then we can talk about what happens if the developers decide to go along with the regulation breaking.

What I fear is the future career path for manual programming is fixing AI generated systems. I suspect it will pay very well because you will be finding and fixing things the AI driven work can't. It is going to be soul destroying as reading AI code is deeply unpleasant in the same way reading the articles is.