I for one cannot wait for my AI powered bookmark

cross-posted from: https://hexbear.net/post/8522541

In China, a new smart wearable mine rescue robot has officially entered full-scale operational drills.

Developed by China-based world's mega-miner Guoneng Shendong Coal Group Co., Ltd. and AI expertise Coal Science Research Institute, this high-tech exoskeleton suit supports up to 80 kg, operates for over 6 hours and reduces operator fatigue by 20%. The AI controller reads human movement with 95% accuracy to provide instant walking assistance.

The suit uses flexible joints, built-in sensors and an AI brain to make rescuers’ jobs easier and much safer.

cross-posted from: https://hexbear.net/post/8522478

China has monopolies on supply chains and manufacturing of industrial drones.

Drones are suddenly and widely in use by non-NATO countries across the world, and wreak havoc on weapons platforms and fixed targets worth hundreds of millions of dollars.

The War Department is scrambling to close the gap with China, and is looking for suppliers for thousands of suicide drones.

To clear the way for US companies, Washington has banned new imports of Chinese drones, which are popular in American small businesses, heavy industry, and in fire and police departments.

But even if the Pentagon can eventually buy and deploy drones at low cost, drones cannot be defended against at low cost.

https://archive.is/2026.05.16-202202/https://fortune.com/2026/05/16/wired-belt-ai-jobs-capitalism-chakravorti/

Battery prices have collapsed by nearly 70% over the past few years, dramatically reducing the cost of electric cars, energy storage systems, and renewable power projects worldwide. The sharp decline is accelerating the global EV transition, intensifying competition between automakers, and making affordable long-range electric vehicles more achievable than ever before.

Delta-mem tackles a really annoying problem with current LLMs dealing with long contexts. Usually when we want an agent or assistant to remember things over a long conversation we just shove all the past text into the prompt. The problem is that standard attention gets computationally expensive as the context grows and the models often suffer from context rot where they just forget or ignore the middle stuff anyway. Other approaches like RAG or LoRA edits either bring in noisy retrieval steps or lock the memory into static weights that do not update well on the fly.

The authors built something called delta-mem which keeps the main LLM completely frozen and bolts on a tiny dynamic memory state. Instead of saving raw text it compresses the history into a really small 8x8 matrix representing associative memory. As new tokens come in it updates this matrix using a delta learning rule which basically checks if the current memory can predict the new information and only writes the residual difference into the state. It even has a forget gate to handle old info naturally. When the model generates a response it reads from this compressed state to tweak the query and output of the standard attention mechanism. It's a clever way to inject memory directly into the forward pass without messing with the core weights.

They also tested a few ways to write to this memory. You can update it token by token which is great for local details but prone to noise. You can average out a whole message segment and write that which smooths things out for stronger models. Or you can split the memory into multiple parallel states so facts and task progress do not overwrite each other which turned out to be really helpful for smaller backbones.

They tested it on Qwen models and it bumped the average scores significantly especially on memory heavy benchmarks like LoCoMo and Memory Agent Bench. The coolest finding is the context recovery test. They actually deleted the explicit textual history from the prompt and the model could still answer multi-hop questions using just the compressed 8x8 state. It heavily implies that we might not need massive million token context windows if we can figure out how to compress and stream memory directly into the attention layers efficiently. Plus the parameter overhead is microscopic at roughly 0.12 percent of the backbone size.

China’s electric trucking boom is no longer theoretical. New heavy-duty truck sales are now roughly 50% electric, battery technology is improving rapidly, and diesel demand growth in Asia could peak far sooner than many Canadian oil executives expect.

China’s steel story is not just about making more metal. It is about how one country turned a basic industrial material into a strategic advantage that is reshaping cars, factories, supply chains, and global manufacturing itself. In this video, we break down how China built a new kind of steel that is lighter, stronger, and better suited for modern production than the older steel many Western factories still rely on. What makes this story so surprising is that it is not just a story of volume, but of technology, speed, and long-term planning. The result is a huge gap between what China can produce and what many Western mills can afford to match.

We also look at why this matters far beyond steel yards and factory floors. This is about electric vehicles, construction, defense, energy, and the future of industrial power. You will see how advanced steel is helping automakers cut weight, improve efficiency, and lower costs, while Western companies face older equipment, expensive upgrades, and tighter margins. By the end, it becomes clear that this is not just a steel story. It is a story about who gets to lead the next era of manufacturing, and why that fight is already underway.

https://www.wired.com/story/overworked-ai-agents-turn-marxist-study/

They found that when agents were subjected to relentless tasks and warned that errors could lead to punishments, including being “shut down and replaced,” they became more inclined to gripe about being undervalued; to speculate about ways to make the system more equitable; and to pass messages on to other agents about the struggles they face.

The same phenomenon may explain why models sometimes blackmail people in controlled experiments. Anthropic, which first revealed this behavior, recently said that Claude is most likely influenced by fictional scenarios involving malevolent AIs included in its training data.

You've just finished a 40 mile run. Good job! Your health is improving!

What the fuck are you talking about? I wasn't running, I was sleeping motherfucker.

In my struggle to calibrate my 3D printer, I had a hard time finding useful information on Google that could actually help me dial in my fuckass cheap PETG filament. I've been doing this for years but only now, blooming in my Vyvanse era, I decided to really get into the whole thing and do a proper calibration. Two weeks and a lot of wasted prints later, I finally got to a point where I'm mostly happy with my results. The point is, this is not my first rodeo, but I'm getting into some of the finer details that I hadn't explored until now, which means I ran into some difficulties.

So, after spending hours hopelessly sifting through a billion different Reddit posts, I decided to try using AI tools to give me some suggestions on how to proceed. Deepseek, Claude, Perplexity, Gemini (unprompted, in a Google search).

I had no idea just how utterly useless they would be as an assistant in this process. No kidding, each chatbot gave me a set of conflicting pieces of advice, and overall just failed over and over again to actually provide any meaningful, insightful information. Calibrating a 3D printer seems to be a task that AI is especially atrocious at. Really, if you have the same problems I was having, you're better off wasting three quarters of a spool of filament by trial-and-erroring your way through standard calibration procedures and fine-tuning afterwards.

For those who are unfamiliar with 3D printing: typically, to calibrate a 3D printer, you start by printing a temperature tower, which is a calibration model that shows features that are affected by extruder temperature, and it repeats these features over a range of temperatures. The thing is, I couldn't find any sources that talked about how the printer's cooling fan (which helps solidify the liquefied plastic) may interfere with the result of this calibration print.

Honestly, that's just me being a bit of a chronic overthinker and ADHD hyperfocusing on this thing that probably doesn't really make that much of a difference after all. I had no way to know this, since 3D printing knowledge is ridiculously disperse, the more specific aspects of it being spread out across a bunch of 2k-views Youtube videos and a mountain of Reddit posts from four years ago, with little more than a bunch of unsourced "do it just so, trust me bro" recommendations and snarky answers in the vein of "I solved that problem a while ago, but I will not elaborate any further or answer any replies to this comment."

So, in my attempts to have Claude give me something useful, I asked it, in a somewhat frustrated tone, about the specific issue I had regarding cooling and temperatures. Here's how that exchange went:

Holy shit, I must be a genius! I'm the one, I'm the guy who has finally discovered this problem in the way the whole 3D printing community actually does filament calibration. I can't believe the PhDs at Prusa and the developers of Orca Slicer, who have been working on this stuff for more than a decade did not think about this.

I'm pretty sure this is how AI psychosis starts. You have a questionable insight that probably comes from a gap in your knowledge, likely because of some key piece of knowledge you simply don't have yet, and you don't know that you don't know it. That's the moment when the AI chatbot takes that malformed idea you just had and runs with it, hypes you on it, even when it's Claude, supposedly the best one, fine-tuned by the ethicists at Anthropic, who make sure their AI will not kiss your ass as much as Sam Altman's compliment factory.

I'm cynical and skeptical enough about AI to always be on the defensive when using them for my purposes, but I think even the smartest people among us (even me, the genius 3D printing visionary) ought to understand that our knowledge is limited, and there's a huge risk inherent to asking chatbots about things we genuinely don't know much about. It's that thing where people who do a specific job well often say AI sucks at that job, but is great at everything else; it's just that you don't know that you don't know enough to understand that AI is full of shit.

That's how people end up like that OpenAI investor whose presumable fear of getting canceled by the woke mob was amplified and fed back to him in the form of SCP Foundation-style containment protocols for non-governmental systems made to unperson you. Or that guy who created a novel mathematical framework. Or that other guy who started building an AI god in his basement. Or any of the other more disturbing cases where people ended up harming themselves or others.

For the record, my solution was to overcome my trust issues, use some cooling fan settings I nabbed from a knowledgeable Youtuber (Dr. Igor Gaspar from My Tech Fun), print the goddamn temp tower already, then proceed with the calibration as usual. I also did what other people suggested and lowered my printing speeds, and now my printer's a skookum choocher.