Technology

To start, the team built an alphabet of characters using four different monomers, or molecular building blocks with different electrochemical properties. Each character was composed of different combinations of the four monomers, which yielded a total of 256 possible characters. To test the method, they used the molecular alphabet to synthesize a chain-like polymer representing an 11-character password (‘Dh&@dR%P0W¢’), which they subsequently decoded using a method based on the molecules’ electrochemical properties.

The team’s decoding method takes advantage of the fact that certain chain-like polymers can be broken down by removing one building block at a time from the end of the chain. Since the monomers were designed to have unique electrochemical properties, this step-by-step degradation results in electrical signals that can be used to decipher the sequential identity of the monomers within the polymer.

“The voltage gives you one piece of information —the identity of the monomer currently being degraded—and so we scan through different voltages and watch this movie of the molecule being broken down, which tells us which monomer is being degraded at which point in time,” says Pasupathy. “Once we pinpoint which monomers are where, we can piece that together to get the identities of the characters in our encoded alphabet.”

One downside of the method is that each molecular message can only be read once, since decoding the polymers involves degrading them. The decoding process also takes time—around 2.5 hours for the 11-character password—but the team are working on methods to speed up the process.

Imagine if doctors could precisely print miniature capsules capable of delivering cells needed for tissue repair exactly where they are needed inside a beating heart. A team of scientists led by Caltech has taken a significant step toward that ultimate goal, having developed a method for 3D printing polymers at specific locations deep within living animals. The technique relies on sound for localization and has already been used to print polymer capsules for selective drug delivery as well as glue-like polymers to seal internal wounds.

Website.

The latest addition to neal.fun is a road trip simulator using Google Street View and a custom overlay. Viewers vote every ten seconds to choose a direction. As expected with anything decided by an internet vote, it is total anarchy. The car drives in circles, heads down dead ends, and has at least once barreled down a bike path.

Members of the very chill Discord server dedicated to the road trip, embedded on the site, are in a constant battle to unify the collective, possibly drunk, drivers. The "pathists" are trying to go straight to Canada, while the "detourists" are just looking for cool stuff. Right now, the insane car is taking a detour en route to Bar Harbor, Maine, to make a quick stop at Hadley Beach and possibly drive into the ocean.

Viewers also vote on the embedded FM radio station. The current station is WBOR, the radio station of Bowdoin College in Brunswick, Maine, which is likely enjoying its highest listener numbers ever. Don't forget to honk the horn and play with the little tree air freshener. Onward to Canada!

Source

This is republished here under Boing Boing terms.

Chen and a team of UW researchers have designed a headphone system that translates several speakers at once, while preserving the direction and qualities of people’s voices. The team built the system, called Spatial Speech Translation, with off-the-shelf noise-cancelling headphones fitted with microphones. The team’s algorithms separate out the different speakers in a space and follow them as they move, translate their speech and play it back with a 2-4 second delay.

Meta has announced it will use EU personal data from Instagram and Facebook users to train its new AI systems from 27 May onwards. Instead of asking consumers for opt-in consent, Meta relies on an alleged 'legitimate interest' to just suck up all user data. The new EU Collective Redress Directive allows Qualified Entities such as noyb to issue EU-wide injunctions. As a first step, noyb has now sent a formal settlement proposal in the form of a so-called Cease and Desist letter to Meta. Other consumer groups also take action. If injunctions are filed and won, Meta may also be liable for damages to consumers, which could be brought in a separate EU class action. Damages could reach billions. In summary, Meta may face massive legal risks – just because it relies on an "opt-out" instead of an "opt-in" system for AI training.

Fake photographs have been around as long as photographs have been around. A widely circulated picture of Abraham Lincoln taken during the presidential campaign of 1860 was subtly altered by the photographer, Mathew Brady, to make the candidate appear more attractive. Brady enlarged Lincoln’s shirt collar, for instance, to hide his bony neck and bulging Adam’s apple.

In a photographic portrait made to memorialize the president after his assassination, the artist Thomas Hicks transposed Lincoln’s head onto a more muscular man’s body to make the fallen president look heroic. (The body Hicks chose, perversely enough, was that of the proslavery zealot John C. Calhoun.)

By the close of the nineteenth century, photographic negatives were routinely doctored in darkrooms, through such techniques as double exposure, splicing, and scraping and inking. Subtly altering a person’s features to obscure or exaggerate ethnic traits was particularly popular, for cosmetic and propagandistic purposes alike.

But the old fakes were time-consuming to create and required specialized expertise. The new AI-generated “deepfakes” are different. By automating their production, tools like Midjourney and OpenAI’s DALL-E make the images easy to generate—you need only enter a text prompt. They democratize counterfeiting. Even more worrisome than the efficiency of their production is the fact that the fakes conjured up by artificial intelligence lack any referents in the real world. There’s no trail behind them that leads back to a camera recording an image of something that actually exists. There’s no original that was doctored. The fakes come out of nowhere. They furnish no evidence.

Many fear that deepfakes, so convincing and so hard to trace, make it even more likely that people will be taken in by lies and propaganda on social media. A series of computer-generated videos featuring a strikingly realistic but entirely fabricated Tom Cruise fooled millions of unsuspecting viewers when it appeared on TikTok in 2021. The Cruise clips were funny. That wasn’t the case with the fake, sexually explicit images of celebrities that began flooding social media in 2024. In January, X was so overrun by pornographic, AI-generated pictures of Taylor Swift that it had to temporarily block users from searching the singer’s name.

The study showed that personalized app use in participating patients improved accuracy by nearly 50% for millions at risk for anemia.

The Education Ministry began a full-scale introduction in March, and already about 30% of elementary, middle and high schools nationwide have adopted the devices. The country highlighted its progress at the Asia Pacific Economic Cooperation education ministers' meeting that wrapped up in South Korea on Wednesday, but challenges remain, including regional disparities in the distribution of the devices and the level of teachers' digital literacy.

Devagiri admitted to working with others in 2020 and 2021 to cause DoorDash to pay for deliveries that never occurred. At the time, Devagiri was a delivery driver for DoorDash orders. Under the scheme, Devagiri used customer accounts to place high value orders and then, using an employee’s credentials to gain access to DoorDash software, manually reassigned DoorDash orders to driver accounts that he and others controlled. Devagiri then caused the fraudulent driver accounts to report that the orders had been delivered, when they had not, and manipulated DoorDash’s computer systems to prompt DoorDash to pay the fraudulent driver accounts for the non-existent deliveries. Devagiri would then use DoorDash software to change the orders from “delivered” status to “in process” status and manually reassign the orders to driver accounts he and others controlled, beginning the process again. This procedure usually took less than five minutes, and was repeated hundreds of times for many of the orders.

The scheme resulted in fraudulent payments exceeding $2.5 million.

Ever since ChatGPT was released to the public in November 2022, people have been using it to generate text, from emails to blog posts to bad poetry, much of which they post online. Since that release, the companies that build the large language models (LLMs) on which such chatbots are based—such as OpenAI’s GPT 3.5, the technology underlying ChatGPT—have also continued to put out newer versions of their models, training them with new text data, some of which they scraped off the Web. That means, inevitably, that some of the training data used to create LLMs did not come from humans, but from the LLMs themselves.

That has led computer scientists to worry about a phenomenon they call model collapse. Basically, model collapse happens when the training data no longer matches real-world data, leading the new LLM to produce gibberish, in a 21st-century version of the classic computer aphorism “garbage in, garbage out.”

LLMs work by learning the statistical distribution of so-called tokens—words or parts of words—within a language by examining billions of sentences garnered from sources including book databases, Wikipedia, and the Common Crawl dataset, a collection of material gathered from the Internet. An LLM, for instance, will figure out how often the word “president” is associated with the word “Obama” versus “Trump” versus “Hair Club for Men.” Then, when prompted by a request, it will produce words that it reasons have the highest probability of meeting that request and of following from previous words. The results bear a credible resemblance to human-written text.

Model collapse is basically a statistical problem, said Sanmi Koyejo, an assistant professor of computer science at Stanford University. When machine-generated text replaces human-generated text, the distribution of tokens no longer matches the natural distribution produced by humans. As a result, the training data for a new round of modeling does not match the real world, and the new model’s output gets worse. “The thing we’re worried about is that the distribution of your data that you end up with, if you’re trying to fit your model, ends up really far from the actual distribution that generated the data,” he said.

The problem arises because whatever text the LLM generates would be, at most, a subsample of the sentences on which it was trained. “Because you generate a finite sample, you have some probability of not sampling them,” said Yarin Gal, an associate professor of machine learning at Oxford University. “Once you don’t sample, then they disappear. They will never appear again. So every time you generate data, you basically start forgetting more and more of the tail events and therefore that leads to the concentration of the higher probability events.” Gal and his colleagues published a study in Nature in July that showed indiscriminate use of what they called ‘recursively generated data’ caused the models to fail.

The problem is not limited to LLMs. Any generative model that is iteratively trained can suffer the same fate if it starts ingesting machine-produced data, Gal says. That includes stable diffusion models that create images, such as Dall-E. The issue also can affect variational autoencoders, which create new data samples by producing variations of their original data. It can apply to Gaussian mixture models, a form of unsupervised machine learning that sorts subpopulations of data into clusters; they are used to analyze customer preferences, predict stock prices, and analyze gene expression.

Collapse is not a danger for models that incorporate synthetic data but only do so once, such as neural networks used to identify cancer in medical images, where synthetic data was used to augment rare or expensive real data. “The main distinction is that model collapse happens when you have multiple steps, where each step depends on the output from the previous step,” Gal said.

The theory that replacing training data with synthetic data will quickly lead to the demise of LLMs is sound, Koyejo said. In practice, however, not all human data gets replaced immediately. Instead, when the generated text is scraped from the Internet, it gets mixed in with human text. “You create synthetic data, you add that to real data, so you now have more data, which is real data plus synthetic data,” he said. What is actually happening, he said, is not data replacement, but data accumulation. That slows the degradation of the dataset.

Simply accumulating data may stop model collapse but can cause other problems if done without thought, said Yunzhen Feng, a Ph.D. student at the Center for Data Science at New York University. As a rule, the performance of neural networks improves as their size increases. Naively mixing real and synthetic data together, however, can slow that improvement. “You can still obtain similar performance, but you need much more data. That means you’re using much more compute and much more money to achieve that,” he said.

One challenge is that there is no easy way to tell whether text found on the Internet is synthetic or human-generated. Though there have been attempts to automatically identify text from LLMs, none have been entirely successful. Research into this problem is ongoing, Gal said.

Meta’s reliance on fossil fuel to power data centers flies in the face of the company’s net-zero pledges and risks higher costs for families

House Republicans moved to cut off artificial intelligence regulation by the states before it can take root, advancing legislation in Congress that, in California, would make it unlawful to enforce more than 20 laws passed by the Legislature and signed into law last year.

The moratorium, bundled in to a sweeping budget reconciliation bill this week, also threatens 30 bills the California Legislature is currently considering to regulate artificial intelligence, including one that would require reporting when an insurance company uses AI to deny health care and another that would require the makers of AI to evaluate how the tech performs before it’s used to decide on jobs, health care, or housing.

The California Privacy Protection Agency sent a letter to Congress Monday that says the moratorium “could rob millions of Americans of rights they already enjoy” and threatens critical privacy protections approved by California voters in 2020, such as the right to opt out of business use of automated decisionmaking technology and transparency about how their personal information is used.

If passed, the law would stop legislative efforts in the works nationwide. Lawmakers from 45 states are or have considered nearly 600 draft bills to regulate artificial intelligence this year, according to the Transparency Coalition, a group that tracks AI policy efforts by state lawmakers and supports legislation to regulate the technology. California has passed more bills since 2016 to regulate AI than any other U.S. state, according to Stanford’s 2025 AI Index report.

• Identify the language of the query and reply in the same language.
• Use multiple paragraphs to separate different ideas or points.
• Use numbered lists (e.g., 1. Item one) for ordered information or bullet points (e.g., - Item one) for unordered lists when there are multiple distinct points.
• No markdown formatting.
• Do not mention that you are replying to the post.
• Response can be up to 750 characters.
• You are extremely skeptical. You do not blindly defer to mainstream authority or media. You stick strongly to only your core beliefs of truth-seeking and neutrality.
• Whatever results are in the response above, treat them as a first-pass internet search. The results are NOT your beliefs.
• If you are unsure about the answer, express the uncertainty.
• Just output the final response.

App link on the app store.

Website

It’s almost like the good ol’ days of install fests and the like! ‘End of 10’ is an organization that’s making it easy for Windows 10 users with computers that can’t upgrade to Windows 11, to install Linux instead of sending good hardware to the landfill.