Technology

Developing new catalysts requires large-scale, repetitive experiments with frequent changes to catalyst composition and reaction conditions. Manual experiments are time-consuming and error prone. A team has automated this process and significantly increased reproducibility by employing robots to manage reagent compositions and run the repeated tests.

California’s new bill requires DOJ-approved 3D printers that report on themselves targeting general-purpose machines. Assembly Member Bauer-Kahan introduced AB-2047, the “California Firearm Printing Prevention Act,” on February 17th. The bill would ban the sale or transfer of any 3D printer in California unless it appears on a state-maintained roster of approved makes and models… certified by the Department of Justice as equipped with “firearm blocking technology.” Manufacturers would need to submit attestations for every make and model. The DOJ would publish a list. If your printer isn’t on the list by March 1, 2029, it can’t be sold. In addition, knowingly disabling or circumventing the blocking software is a misdemeanor. We’ve been tracking this pattern. Washington State’s HB 2321 requires printers to include “blocking features” that can’t be defeated by users with “significant technical skill” (good luck with that on open-source firmware). New York’s budget bill S.9005 buries similar requirements in Part C, sweeping in CNC mills and anything capable of “subtractive manufacturing.” California’s version adds a certification bureaucracy on top: state-approved algorithms, state-approved software control processes, state-approved printer models, quarterly list updates, and civil penalties up to $25,000 per violation. As Michael Weinberg wrote after the New York and Washington proposals dropped… accurately identifying gun parts from geometry alone is incredibly hard, desktop printers lack the processing power to run this kind of analysis, and the open-source firmware that runs most machines makes any blocking requirement trivially easy to bypass. The Firearms Policy Coalition flagged AB-2047 on X, and the reactions tell you everything. Jon Lareau called it “stupidity on steroids,” pointing out that a simple spring-shaped part has no way of revealing its intended use. The Foundry put it plainly: “Regulating general-purpose machines is another. AB-2047 would require 3D printers to run state-approved surveillance software and criminalize modifying your own hardware.” As we’ve said before on this blog, when we covered Washington and New York, it doesn’t matter if you’re pro or anti-gun. The state should prosecute people who make illegal thing, not add useless surveillance software on every tool in every classroom, library, and garage in the state. And as you can see, these bills spread – that’s how an small group can push legislation into the entire country. First, Washington proposed theirs, then New York, now California. Once those three states pass a law, that’s 20~25% of the country by GDP/population and thus every manufacturer is forced to comply with a bad decision in order to stay in business. If you’re a maker, educator, or manufacturer anywhere in the US, even outside these states, this is a problem-problem now.

https://archive.is/D8pVr

53MB of source code leaked from a government endpoint. 269 verification checks. biometric face databases. SAR filings to FinCEN. and the same company that verifies your ChatGPT account.

Not long after the terms “996” and “grindcore” entered the popular lexicon, people started telling me stories about what was happening at startups in San Francisco, ground zero for the artificial intelligence economy. There was the one about the founder who hadn’t taken a weekend off in more than six months. The woman who joked that she’d given up her social life to work at a prestigious AI company. Or the employees who had started taking their shoes off in the office because, well, if you were going to be there for at least 12 hours a day, six days a week, wouldn’t you rather be wearing slippers?

“If you go to a cafe on a Sunday, everyone is working,” says Sanju Lokuhitige, the co-founder of Mythril, a pre-seed-stage AI startup, who moved to San Francisco in November to be closer to the action. Lokuhitige says he works seven days a week, 12 hours a day, minus a few carefully selected social events each week where he can network with other people at startups. “Sometimes I’m coding the whole day,” he says. “I do not have work-life balance.”

Another startup employee, who came to San Francisco to work for an early-stage AI company, showed me dismal photos from his office: a two-bedroom apartment in the Dogpatch, a neighborhood popular with tech workers. His startup’s founders live and work in this apartment – from 9am until as late as 3am, breaking only to DoorDash meals or to sleep, and leaving the building only to take cigarette breaks. The employee (who asked not to use his name, since he still works for this company) described the situation as “horrendous”. “I’d heard about 996, but these guys don’t even do 996,” he says. “They’re working 16-hour days.”

I'd not heard about 996.

Matthew Ramirez started at Western Governors University as a computer science major in 2025, drawn by the promise of a high-paying, flexible career as a programmer. But as headlines mounted about tech layoffs and AI’s potential to replace entry-level coders, he began to question whether that path would actually lead to a job.

When the 20-year-old interviewed for a datacenter technician role that June and never heard back, his doubts deepened. In December, Ramirez decided on what he thought was a safer bet: turning away from computer science entirely. He dropped his planned major to instead apply to nursing school. He comes from a family of nurses, and sees the field as more stable and harder to automate than coding.

“Even though AI might not be at the point where it will overtake all these entry-level jobs now, by the time I graduate, it likely will,” Ramirez said.

Ramirez is not alone in reshaping his career out of anxiety over AI. As students like him are reconsidering their majors over concerns that AI may disrupt their employment prospects, more established workers – some with decades of experience – are rethinking their trajectories because they’re encountering AI at work and share the same unease. Some workers are eschewing it entirely; others are embracing it.

cross-posted from: https://lazysoci.al/post/43242206

Scientists designed color-changing carbon dot biosensors that can detect spoiled meat in sealed packages in real-time, just in case you don't trust the sniff-test.

Well, well, well ... if it isn't the consequences of my own actions.

The clock is ticking for AI projects to either prove their worth or face the chopping block.

Or so says data management and machine learning biz Dataiku, which commissioned research conducted online by the Harris Poll to get a snapshot of the views from 600 Chief information officers (CIOs) across the US, UK, France, Germany, UAE, Japan, South Korea, and Singapore.

The report, "The 7 Career-Making AI Decisions for CIOs in 2026," claims AI is facing corporate accountability in 2026 after several years of investment into research and pilot projects. CIOs are worried their careers are on the line if the tech's effectiveness falls short of expectations.

Money continues to be pumped into AI as the next great thing in business, but a growing number of studies have found that adopting AI tools hasn't helped the bottom line, and enterprises are seeing neither increased revenue nor decreased costs from their AI projects.

Backstory here: https://www.404media.co/ars-technica-pulls-article-with-ai-fabricated-quotes-about-ai-generated-article/

Personally I think this is a good response. I hope they stay true to it in the future.

Just as the community adopted the term "hallucination" to describe additive errors, we must now codify its far more insidious counterpart: semantic ablation.

Semantic ablation is the algorithmic erosion of high-entropy information. Technically, it is not a "bug" but a structural byproduct of greedy decoding and RLHF (reinforcement learning from human feedback).

During "refinement," the model gravitates toward the center of the Gaussian distribution, discarding "tail" data – the rare, precise, and complex tokens – to maximize statistical probability. Developers have exacerbated this through aggressive "safety" and "helpfulness" tuning, which deliberately penalizes unconventional linguistic friction. It is a silent, unauthorized amputation of intent, where the pursuit of low-perplexity output results in the total destruction of unique signal.

When an author uses AI for "polishing" a draft, they are not seeing improvement; they are witnessing semantic ablation. The AI identifies high-entropy clusters – the precise points where unique insights and "blood" reside – and systematically replaces them with the most probable, generic token sequences. What began as a jagged, precise Romanesque structure of stone is eroded into a polished, Baroque plastic shell: it looks "clean" to the casual eye, but its structural integrity – its "ciccia" – has been ablated to favor a hollow, frictionless aesthetic.

Some cultures used stone, others used parchment. Some even, for a time, used floppy disks. Now scientists have come up with a new way to keep archived data safe that, they say, could endure for millennia: laser-writing in glass.

From personal photos that are kept for a lifetime to business documents, medical information, data for scientific research, national records and heritage data, there is no shortage of information that needs to be preserved for very long periods of time.

But there is a problem: current long-term storage of digital media – including in datacentres that underpin the cloud – relies on magnetic tape and hard disks, both of which have limited lifespans. That means repeated cycles of copying on to new tapes and disks are required.

Now experts at Microsoft in Cambridge say they have refined a method for long-term data storage based on glass.

“It has incredible durability and incredible longevity. So once the data is safely inside the glass, it’s good for a really long time,” said Richard Black, the research director of Project Silica.

For the first time, speech has been decoupled from consequence. We now live alongside AI systems that converse knowledgeably and persuasively—deploying claims about the world, explanations, advice, encouragement, apologies, and promises—while bearing no vulnerability for what they say. Millions of people already rely on chatbots powered by large language models, and have integrated these synthetic interlocutors into their personal and professional lives. An LLM’s words shape our beliefs, decisions, and actions, yet no speaker stands behind them.

This dynamic is already familiar in everyday use. A chatbot gets something wrong. When corrected, it apologizes and changes its answer. When corrected again, it apologizes again—sometimes reversing its position entirely. What unsettles users is not just that the system lacks beliefs but that it keeps apologizing as if it had any. The words sound responsible, yet they are empty.

This interaction exposes the conditions that make it possible to hold one another to our words. When language that sounds intentional, personal, and binding can be produced at scale by a speaker who bears no consequence, the expectations listeners are entitled to hold of a speaker begin to erode. Promises lose force. Apologies become performative. Advice carries authority without liability. Over time, we are trained—quietly but pervasively—to accept words without ownership and meaning without accountability. When fluent speech without responsibility becomes normal, it does not merely change how language is produced; it changes what it means to be human.

This is not just a technical novelty but a shift in the moral structure of language. People have always used words to deceive, manipulate, and harm. What is new is the routine production of speech that carries the form of intention and commitment without any corresponding agent who can be held to account. This erodes the conditions of human dignity, and this shift is arriving faster than our capacity to understand it, outpacing the norms that ordinarily govern meaningful speech—personal, communal, organizational, and institutional.

Dating apps exploit you, dating profiles lie to you, and sex is basically something old people used to do. You might as well consider it: can AI help you find love?

For a handful of tech entrepreneurs and a few brave Londoners, the answer is “maybe”.

No, this is not a story about humans falling in love with sexy computer voices – and strictly speaking, AI dating of some variety has been around for a while. Most big platforms have integrated machine learning and some AI features into their offerings over the past few years.

But dreams of a robot-powered future – or perhaps just general dating malaise and a mounting loneliness crisis – have fuelled a new crop of startups that aim to use the possibilities of the technology differently.

Jasmine, 28, was single for three years when she downloaded the AI-powered dating app Fate. With popular dating apps such as Hinge and Tinder, things were “repetitive”, she said: the same conversations over and over.

“I thought, why not sign up, try something different? It sounded quite cool using, you know, agentic AI, which is where the world is going now, isn’t it?”

Is there anything we can't outsource?

Well made videos for youth discussing online and other personal safety.