The chief content officer for New York's public radio station WNYC predicts an "AI-fueled shift to niche community and authored excellence."

And ironically, it will be fueled by "Greedy publishers and malicious propagandists... flooding the web with fake or just mediocre AI-generated 'content'" which will "spotlight and boost the value of authored creativity." And it may help give birth to a new generation of independent media. Robots will make the internet more human.

First, it will speed up our migration off of big social platforms to niche communities where we can be better versions of ourselves. We're already exhausted by feeds that amplify our anxiety and algorithms that incentivize cruelty. AI will take the arms race of digital publishing shaped by algorithmic curation to its natural conclusion: big feed-based social platforms will become unending streams of noise. When we've left those sites for good, we'll miss the (mostly inaccurate) sense that we were seeing or participating in a grand, democratic town hall. But as we find places to convene where good faith participation is expected, abuse and harassment aren't, and quality is valued over quantity, we'll be happy to have traded a perception of scale influence for the experience of real connection.

Second, this flood of authorless "content" will help truly authored creativity shine in contrast... "Could a robot have done this?" will be a question we ask to push ourselves to be funnier, weirder, more vulnerable, and more creative. And for the funniest, the weirdest, the most vulnerable, and most creative: the gap between what they do and everything else will be huge. Finally, these AI-accelerated shifts will combine with the current moment in media economics to fuel a new era of independent media.
For a few years he's seen the rise of independent community-funded journalists, and "the list of thriving small enterprises is getting longer." He sees more growth in community-funding platforms (with subscription/membership features like on Substack and Patreon) which "continue to tilt the risk/reward math for audience-facing talent....

"And the amount of audience-facing, world-class talent that left institutional media in 2023 (by choice or otherwise) is unlike anything I've seen in more than 15 years in journalism... [I]f we're lucky, we'll see the creation of a new generation of independent media businesses whose work is as funny, weird, vulnerable and creative as its creators want it to be. And those businesses will be built on truly stable ground: a direct financial relationship with people who care.

"Thank the robots."

Longtime Slashdot reader fahrbot-bot shares a report from ScienceAlert: According to a new study [published in Nature Human Behavior], the human brain has two separate ways of processing numbers of things: one system for quantities of four or fewer, and another system for five and up. Presented with four or fewer objects, humans can usually identify the sum at first glance, without counting. And we're almost always right. This ability is known as "subitizing," a term coined by psychologists last century, and it's different from both counting and estimating. It refers to an uncanny sense of immediately knowing how many things you're looking at, with no tallying or guessing required.

While we can easily subitize quantities up to four, however, the ability disappears when we're looking at five or more things. If asked to instantly quantify a group of seven apples, for example, we tend to hesitate and estimate, taking slightly longer to respond and still providing less precise answers. Since our subitizing skills vanish so abruptly for quantities larger than four, some researchers have suspected our brains use two distinct processing methods, specialized for either small or large quantities. "However, this idea has been disputed up to now," says co-author Florian Mormann, a cognitive neurophysiologist from the Department of Epileptology at the University Hospital Bonn. "It could also be that our brain always makes an estimate but the error rates for smaller numbers of things are so low that they simply go unnoticed."

Previous research involving some of the new study's authors showed that human brains have neurons responsible for each number, with certain nerve cells firing selectively in response to certain quantities. Some neurons fire mainly when a person sees two of something, they found, while others show a similar affinity for their own number of visual elements. Yet many of these neurons also fire in response to slightly smaller or larger numbers, the researchers note, with a weaker reaction for quantities further removed from their numerical focus. "A brain cell for a number of 'seven' elements thus also fires for six and eight elements but more weakly," says neurobiologist Andreas Nieder from the University of Tubingen. "The same cell is still activated but even less so for five or nine elements."

This kind of "numerical distance effect" also occurs in monkeys, as Nieder has shown in previous research. Among humans, however, it typically happens only when we see five or more things, hinting at some undiscovered difference in the way we identify smaller numbers. "There seems to be an additional mechanism for numbers of around less than five elements that makes these neurons more precise," Nieder says. Neurons responsible for lower numbers are able to inhibit other neurons responsible for adjacent numbers, the study's authors report, thus limiting any mixed signals about the quantity in question. When a trio-specializing neuron fires, for example, it also inhibits the neurons that typically fire in response to groups of two or four things. Neurons for the number five and beyond apparently lack this mechanism.

"The events of 2023 showed that A.I. doesn't need to be that good in order to do damage," argues novelist Lincoln Michel in the New Republic: This March, news broke that the latest artificial intelligence models could pass the LSAT, SAT, and AP exams. It sparked another round of A.I. panic. The machines, it seemed, were already at peak human ability. Around that time, I conducted my own, more modest test. I asked a couple of A.I. programs to "write a six-word story about baby shoes," riffing on the famous (if apocryphal) Hemingway story. They failed but not in the way I expected. Bard gave me five words, and ChatGPT produced eight. I tried again, specifying "exactly six words," and received eight and then four words. What did it mean that A.I. could best top-tier lawyers yet fail preschool math?

A year since the launch of ChatGPT, I wonder if the answer isn't just what it seems: A.I. is simultaneously impressive and pretty dumb. Maybe not as dumb as the NFT apes or Zuckerberg's Metaverse cubicle simulator, which Silicon Valley also promised would revolutionize all aspects of life. But at least half-dumb. One day A.I. passes the bar exam, and the next, lawyers are being fined for citing A.I.-invented laws. One second it's "the end of writing," the next it's recommending recipes for "mosquito-repellant roast potatoes." At best, A.I. is a mixed bag. (Since "artificial intelligence" is an intentionally vague term, I should specify I'm discussing "generative A.I." programs like ChatGPT and MidJourney that create text, images, and audio. Credit where credit is due: Branding unthinking, error-prone algorithms as "artificial intelligence" was a brilliant marketing coup)....

The legal questions will be settled in court, and the discourse tends to get bogged down in semantic debates about "plagiarism" and "originality," but the essential truth of A.I. is clear: The largest corporations on earth ripped off generations of artists without permission or compensation to produce programs meant to rip us off even more. I believe A.I. defenders know this is unethical, which is why they distract us with fan fiction about the future. If A.I. is the key to a gleaming utopia or else robot-induced extinction, what does it matter if a few poets and painters got bilked along the way? It's possible a souped-up Microsoft Clippy will morph into SkyNet in a couple of years. It's also possible the technology plateaus, like how self-driving cars are perpetually a few years away from taking over our roads. Even if the technology advances, A.I. costs lots of money, and once investors stop subsidizing its use, A.I. — or at least quality A.I. — may prove cost-prohibitive for most tasks....

A year into ChatGPT, I'm less concerned A.I. will replace human artists anytime soon. Some enjoy using A.I. themselves, but I'm not sure many want to consume (much less pay for) A.I. "art" generated by others. The much-hyped A.I.-authored books have been flops, and few readers are flocking to websites that pivoted to A.I. Last month, Sports Illustrated was so embarrassed by a report they published A.I. articles that they apologized and promised to investigate. Say what you want about NFTs, but at least people were willing to pay for them.
"A.I. can write book reviews no one reads of A.I. novels no one buys, generate playlists no one listens to of A.I. songs no one hears, and create A.I. images no one looks at for websites no one visits.

"This seems to be the future A.I. promises. Endless content generated by robots, enjoyed by no one, clogging up everything, and wasting everyone's time."

Following policy adjustments by the U.S. Citizenship and Immigration Services (USCIS) in January, more foreign-born workers in science, technology, engineering, and math (STEM) fields are able to live and work permanently in the United States. "The jump comes after USCIS in January 2022 tweaked its guidance criteria relating to two visa categories available to STEM workers," reports Science Magazine. "One is the O-1A, a temporary visa for 'aliens of extraordinary ability' that often paves the way to a green card. The second, which bestows a green card on those with advanced STEM degrees, governs a subset of an EB-2 (employment-based) visa." From the report: The USCIS data, reported exclusively by ScienceInsider, show that the number of O-1A visas awarded in the first year of the revised guidance jumped by almost 30%, to 4570, and held steady in fiscal year 2023, which ended on 30 September. Similarly, the number of STEM EB-2 visas approved in 2022 after a "national interest" waiver shot up by 55% over 2021, to 70,240, and stayed at that level this year. "I'm seeing more aspiring and early-stage startup founders believe there's a way forward for them," says Silicon Valley immigration attorney Sophie Alcorn. She predicts the policy changes will result in "new technology startups that would not have otherwise been created."

President Joe Biden has long sought to make it easier for foreign-born STEM workers to remain in the country and use their talent to spur the U.S. economy. But under the terms of a 1990 law, only 140,000 employment-based green cards may be issued annually, and no more than 7% of those can go to citizens of any one country. The ceiling is well below the demand. And the country quotas have created decades-long queues for scientists and high-tech entrepreneurs born in India and China. The 2022 guidance doesn't alter those limits on employment-based green cards but clarifies the visa process for foreign-born scientists pending any significant changes to the 1990 law. The O-1A work visa, which can be renewed indefinitely, was designed to accelerate the path to a green card for foreign-born high-tech entrepreneurs.

Although there is no cap on the number of O-1A visas awarded, foreign-born scientists have largely ignored this option because it wasn't clear what metrics USCIS would use to assess their application. The 2022 guidance on O-1As removed that uncertainty by listing eight criteria -- including awards, peer-reviewed publications, and reviewing the work of other scientistsâ"and stipulating that applicants need to satisfy at least three of them. The second visa policy change affects those with advanced STEM degrees seeking the national interest waiver for an EB-2. Under the normal process of obtaining such a visa, the Department of Labor requires employers to first satisfy rules meant to protect U.S. workers from foreign competition, for example, by showing that the company has failed to find a qualified domestic worker and that the job will pay the prevailing wage. That time-consuming exercise can be waived if visa applicants can prove they are doing "exceptional" work of "substantial merit and national importance." But once again, the standard for determining whether the labor-force requirements can be waived was vague, so relatively few STEM workers chose that route. The 2022 USCIS guidance not only specifies criteria, which closely track those for the nonimmigrant, O-1A visa, but also allows scientists to sponsor themselves.

"It's like visiting an old friend for the holidays," according to this article: Approaching his 86th birthday, Donald Knuth — Stanford's beloved computer science guru — honored what's become a long-standing tradition. He gave a December "Christmas lecture" that's also streamed online for all of his fans...

More than 60 years ago, back in 1962, a 24-year-old Donald Knuth first started writing The Art of Computer Programming — a comprehensive analysis of algorithms which, here in 2023, he's still trying to finish. And 30 years ago Knuth also began making rare live appearances each December in front of audiences of Stanford students...

Recently Stanford uploaded several decades of Knuth's past Christmas lectures, along with a series of 22 videos of Knuth from 1985 titled "the 'Aha' Sessions'" (courses in mathematical problem-solving). There are also two different sets of five videos from 1981 showing Knuth introducing his newly-created typesetting system TeX. There are even 12 videos from 1982 of what Knuth calls "an intensive course about the internal details."

And on Dec. 6, wearing his traditional brown holiday sweater, Knuth gave yet another live demonstration of the beautifully clear precision that's made him famous.

The New Yorker looks at 41-year-old Amabel Holland, an autistic board-game designer who "thinks about the world in terms of systems," and realized you could make a board game about almost anything, "and, when you did, its rules could both mirror and analyze the subject on which it was based."

They've since designed more than 60 games, and the article notes that Holland's work, "which is part of a larger turn toward complexity in the industry, often tackles historical and social subjects — death, religion, misinformation — using surprising 'mechanics,' or building blocks of game play, to immerse players in an experience." "With every game, you build a certain model of the world," Reiner Knizia, a former mathematician who's designed more than eight hundred games, told me. Several of his games illustrate market forces: in Modern Art, for instance, you play as auctioneers and buyers, hoping to buy low and sell high. Knizia is a traditional game designer inasmuch as he aims to "bring enjoyment to the people." But Amabel sometimes aims for the opposite of enjoyment... This Guilty Land, from 2018, is about the struggle to end slavery."
Holland says their games are "meant to evoke frustration" — specifically to communicate how difficult it can be to actually achieve political progress.

Thanks to Slashdot reader silverjacket for sharing the article.

Long-time Slashdot reader schwit1 shared this article from the Atlantic: For the past few years, parents, researchers, and the news media have paid closer attention to the relationship between teenagers' phone use and their mental health. Researchers such as Jonathan Haidt and Jean Twenge have shown that various measures of student well-being began a sharp decline around 2012 throughout the West, just as smartphones and social media emerged as the attentional centerpiece of teenage life. Some have even suggested that smartphone use is so corrosive, it's systematically reducing student achievement. I hadn't quite believed that last argument — until now.

The Program for International Student Assessment, conducted by the Organization for Economic Co-operation and Development in almost 80 countries every three years, tests 15-year-olds est scores have been falling for years — even before the pandemic. Across the OECD, science scores peaked in 2009, and reading scores peaked in 2012. Since then, developed countries have as a whole performed "increasingly poorly" on average. "No single country showed an increasingly positive trend in any subject," PISA reported, and "many countries showed increasingly poor performance in at least one subject." Even in famously high-performing countries, such as Finland, Sweden, and South Korea, PISA grades in one or several subjects have been declining for a while.

So what's driving down student scores around the world? The PISA report offers three reasons to suspect that phones are a major culprit. First, PISA finds that students who spend less than one hour of "leisure" time on digital devices a day at school scored about 50 points higher in math than students whose eyes are glued to their screens more than five hours a day. This gap held even after adjusting for socioeconomic factors... Second, screens seem to create a general distraction throughout school, even for students who aren't always looking at them.... Finally, nearly half of students across the OECD said that they felt "nervous" or "anxious" when they didn't have their digital devices near them. (On average, these students also said they were less satisfied with life.) This phone anxiety was negatively correlated with math scores.

In sum, students who spend more time staring at their phone do worse in school, distract other students around them, and feel worse about their life.

According to researchers from a startup called Patronus AI, ChatGPT and other chatbots that rely on large language models frequently fail to answer questions derived from Securities and Exchange Commission filings. CNBC reports: Even the best-performing artificial intelligence model configuration they tested, OpenAI's GPT-4-Turbo, when armed with the ability to read nearly an entire filing alongside the question, only got 79% of answers right on Patronus AI's new test, the company's founders told CNBC. Oftentimes, the so-called large language models would refuse to answer, or would "hallucinate" figures and facts that weren't in the SEC filings. "That type of performance rate is just absolutely unacceptable," Patronus AI co-founder Anand Kannappan said. "It has to be much much higher for it to really work in an automated and production-ready way." [...]

Patronus AI worked to write a set of more than 10,000 questions and answers drawn from SEC filings from major publicly traded companies, which it calls FinanceBench. The dataset includes the correct answers, and also where exactly in any given filing to find them. Not all of the answers can be pulled directly from the text, and some questions require light math or reasoning. Qian and Kannappan say it's a test that gives a "minimum performance standard" for language AI in the financial sector. Patronus AI tested four language models: OpenAI's GPT-4 and GPT-4-Turbo, Anthropic's Claude 2 and Meta's Llama 2, using a subset of 150 of the questions it had produced. It also tested different configurations and prompts, such as one setting where the OpenAI models were given the exact relevant source text in the question, which it called "Oracle" mode. In other tests, the models were told where the underlying SEC documents would be stored, or given "long context," which meant including nearly an entire SEC filing alongside the question in the prompt.

GPT-4-Turbo failed at the startup's "closed book" test, where it wasn't given access to any SEC source document. It failed to answer 88% of the 150 questions it was asked, and only produced a correct answer 14 times. It was able to improve significantly when given access to the underlying filings. In "Oracle" mode, where it was pointed to the exact text for the answer, GPT-4-Turbo answered the question correctly 85% of the time, but still produced an incorrect answer 15% of the time. But that's an unrealistic test because it requires human input to find the exact pertinent place in the filing -- the exact task that many hope that language models can address. Llama 2, an open-source AI model developed by Meta, had some of the worst "hallucinations," producing wrong answers as much as 70% of the time, and correct answers only 19% of the time, when given access to an array of underlying documents. Anthropic's Claude 2 performed well when given "long context," where nearly the entire relevant SEC filing was included along with the question. It could answer 75% of the questions it was posed, gave the wrong answer for 21%, and failed to answer only 3%. GPT-4-Turbo also did well with long context, answering 79% of the questions correctly, and giving the wrong answer for 17% of them.

India consumes more groundwater. That's testing India's ability to feed itself and much of the world. From a report: The South Asian nation is already the world's largest guzzler of groundwater. Cheap power has encouraged routine overreliance on finite riches. India overwhelmingly grows some of the thirstiest crops: rice, wheat and sugar cane. Over the last half century, farm productivity has leapt forward, but so, too, has water usage -- up 500% over that period, according to the World Bank. Erratic monsoons and brutal heat waves are only making the problem more acute. Farmers are digging deeper wells because existing ones are no longer refilling. Some regions may run out of groundwater entirely -- Punjab, a major wheat producer, could go dry within the next 15 or so years, according to a former state official. States in southern India are battling over water rights in areas where rampant urban development has drained thousands of lakes.

The government is not blind to the crisis. But with a national election on the horizon next year, there's little to gain in pushing actively for change among farmers, one of the most important voting blocs in the country. Any long-term solution will involve tinkering with farm subsidies or the minimum price set for water-intensive crops. Prime Minister Narendra Modi's ruling party is all too aware that farmers from India's grain-growing northern regions dominated months of protests against proposed agrarian reforms from late 2020. Modi was forced to withdraw the proposals. For now, it's clear the water math does not add up.

Modi has promised piped water to all Indian households by 2024. Yet nearly half of India's 1.4 billion residents already face high-to-extreme water stress, and the world's most populous nation is expected to add more than 200 million more people by 2050. Agriculture, meanwhile, accounts for 90% of water use, helping to explain why Indian officials say the clearest strategy for preserving supplies is modernizing the industry. The government has tried to convince farmers to adopt different irrigation technologies, return to traditional rain harvesting and plant less thirsty crops like millets, pulses and oilseeds. Nothing has yet made a substantial difference, in a country where subsidies supporting wheat and rice persist, and farming is dominated by smallholders.

An anonymous reader quotes a report from MIT Technology Review: In a paper published in Nature today, the researchers say it is the first time a large language model has been used to discover a solution to a long-standing scientific puzzle -- producing verifiable and valuable new information that did not previously exist. "It's not in the training data -- it wasn't even known," says coauthor Pushmeet Kohli, vice president of research at Google DeepMind. Large language models have a reputation for making things up, not for providing new facts. Google DeepMind's new tool, called FunSearch, could change that. It shows that they can indeed make discoveries -- if they are coaxed just so, and if you throw out the majority of what they come up with.

FunSearch (so called because it searches for mathematical functions, not because it's fun) continues a streak of discoveries in fundamental math and computer science that DeepMind has made using AI. First Alpha Tensor found a way to speed up a calculation at the heart of many different kinds of code, beating a 50-year record. Then AlphaDev found ways to make key algorithms used trillions of times a day run faster. Yet those tools did not use large language models. Built on top of DeepMind's game-playing AI AlphaZero, both solved math problems by treating them as if they were puzzles in Go or chess. The trouble is that they are stuck in their lanes, says Bernardino Romera-Paredes, a researcher at the company who worked on both AlphaTensor and FunSearch: "AlphaTensor is great at matrix multiplication, but basically nothing else." FunSearch takes a different tack. It combines a large language model called Codey, a version of Google's PaLM 2 that isfine-tuned on computer code, with other systems that reject incorrect or nonsensical answers and plug good ones back in.

The researchers started by sketching out the problem they wanted to solve in Python, a popular programming language. But they left out the lines in the program that would specify how to solve it. That is where FunSearch comes in. It gets Codey to fill in the blanks -- in effect, to suggest code that will solve the problem. A second algorithm then checks and scores what Codey comes up with. The best suggestions -- even if not yet correct -- are saved and given back to Codey, which tries to complete the program again. After a couple of million suggestions and a few dozen repetitions of the overall process -- which took a few days -- FunSearch was able to come up with code that produced a correct and previously unknown solution to the cap set problem, which involves finding the largest size of a certain type of set. Imagine plotting dots on graph paper. [...] To test its versatility, the researchers used FunSearch to approach another hard problem in math: the bin packing problem, which involves trying to pack items into as few bins as possible. This is important for a range of applications in computer science, from data center management to e-commerce. FunSearch came up with a way to solve it that's faster than human-devised ones.

Ivana Saric reports via Axios: U.S. students lag behind their peers in many industrialized countries when it comes to math, according to the results of a global exam released Tuesday. U.S. students saw a 13-point drop in their 2022 Program for International Student Assessment (PISA) math results when compared to the 2018 exam. The 2022 math score was not only lower than it was in 2012 but it was "among the lowest ever measured by PISA in mathematics" for the U.S., per the Organisation for Economic Co-operation and Development (OECD) country note. The 2018 PISA assessment found that U.S. students straggled behind their peers in East Asia and Europe, per the Washington Post.

PISA examines the proficiency of 15-year-olds in reading, mathematics, and science worldwide. The 2022 PISA edition is the first to take place since the pandemic and compares the test results of nearly 700,000 students across 81 OECD member states and partner economies. The exam, coordinated by the OECD, was first administered in 2000 and is conducted every three years. Due to the COVID-19 pandemic, the 2021 test was delayed until 2022. What about the rest of the world? According to Axios, a total of 31 countries and economies "maintained or improved upon their 2018 math scores, including Switzerland and Japan."

"10 countries and economies -- Canada, Denmark, Finland, Hong Kong, Ireland, Japan, Korea, Latvia, Macao and the U.K. -- saw their students score proficiently in all three domains and had 'high levels of socio-economic fairness,'" the report adds.

An anonymous reader writes: People spend one-third of their lives asleep. What if employees could work during that time ... in their dreams? Prophetic, a venture-backed startup founded earlier this year, wants to help workers do just that. Using a headpiece the company calls the "Halo," Prophetic says consumers can induce a lucid dream state, which occurs when the person having a dream is aware they are sleeping. The goal is to give people control over their dreams, so they can use that time productively. A CEO could practice for an upcoming board meeting, an athlete could run through plays, a web designer could create new templates -- "the limiting factor is your imagination," founder and CEO Eric Wollberg told Fortune.

Consumer devices claiming to induce lucid dream states aren't new. Headbands, eye masks, and boxes with electrodes that stick to the forehead all populate the market. Even some supplements claim to do the trick. But there's still an appetite for new technologies, since the potential for creativity and problem-solving is so great and since many on the market don't work to the extent they promise, a dreaming expert told Fortune. The potential of lucid dreaming is less about conquering specific problems and more about finding new, creative ways to approach topics that a sleeper couldn't previously fathom. For example, a mathematician might not reach a specific, numerical answer to a math problem while asleep, but the lucid dream allows them to explore new strategies to tackle the equation while awake. Halos will cost around $1,500 to $2,000 each.

An anonymous reader quotes a report from Reuters: Ahead of OpenAI CEO Sam Altman's four days in exile, several staff researchers sent the board of directors a letter warning of a powerful artificial intelligence discovery that they said could threaten humanity, two people familiar with the matter told Reuters. The previously unreported letter and AI algorithm was a key development ahead of the board's ouster of Altman, the poster child of generative AI, the two sources said. Before his triumphant return late Tuesday, more than 700 employees had threatened to quit and join backer Microsoft in solidarity with their fired leader. The sources cited the letter as one factor among a longer list of grievances by the board that led to Altman's firing. Reuters was unable to review a copy of the letter.

According to one of the sources, long-time executive Mira Murati mentioned the project, called Q*, to employees on Wednesday and said that a letter was sent to the board prior to this weekend's events. After the story was published, an OpenAI spokesperson said Murati told employees what media were about to report, but she did not comment on the accuracy of the reporting. The maker of ChatGPT had made progress on Q* (pronounced Q-Star), which some internally believe could be a breakthrough in the startup's search for superintelligence, also known as artificial general intelligence (AGI), one of the people told Reuters. OpenAI defines AGI as AI systems that are smarter than humans. Given vast computing resources, the new model was able to solve certain mathematical problems, the person said on condition of anonymity because they were not authorized to speak on behalf of the company. Though only performing math on the level of grade-school students, acing such tests made researchers very optimistic about Q*'s future success, the source said.

Researchers consider math to be a frontier of generative AI development. Currently, generative AI is good at writing and language translation by statistically predicting the next word, and answers to the same question can vary widely. But conquering the ability to do math -- where there is only one right answer -- implies AI would have greater reasoning capabilities resembling human intelligence. This could be applied to novel scientific research, for instance, AI researchers believe. Unlike a calculator that can solve a limited number of operations, AGI can generalize, learn and comprehend. In their letter to the board, researchers flagged AI's prowess and potential danger, the sources said without specifying the exact safety concerns noted in the letter. There has long been discussion among computer scientists about the danger posed by superintelligent machines, for instance if they might decide that the destruction of humanity was in their interest. Last night, OpenAI announced it reached an agreement for Sam Altman to return as CEO. Under an "agreement in principle," Altman will serve under the supervision of a new board of directors.

"xAI, Elon Musk's new AI venture, launched its first AI chatbot technology named Grok," reports CNBC.

Two months into its "early beta" training phase, it's "only available to a select group of users before a wider release" — though users can sign up for a waitlist. Elon Musk posted that the chatbot "will be provided as part of X Premium+, so I recommend signing up for that. Just $16/month via web."

More details from CNBC: Grok, the company said, is modeled on "The Hitchhiker's Guide to the Galaxy." It is supposed to have "a bit of wit," "a rebellious streak" and it should answer the "spicy questions" that other AI might dodge, according to a Saturday statement from xAI... Grok also has access to data from X, which xAI said will give it a leg-up. Musk, on Sunday, posted a side-by-side comparison of Grok answering a question versus another AI bot, which he said had less current information.

Still, xAI hedged in its statement, as with any Large Language Model, or LLM, Grok "can still generate false or contradictory information...." On an initial round of tests based on middle school math problems and Python coding tasks, the company said that Grok surpassed "all other models in its compute class, including ChatGPT-3.5 and Inflection-1." It was outperformed by bots with larger data troves...

Musk has previously said that he believes today's AI makers are bending too far toward "politically correct" systems. xAI's mission, it said, is to create AI for people of all backgrounds and political views. Grok is said to be a means of testing that AI approach "in public."
SpaceX security engineer Christopher Stanley shared some interesting results. After reading Grok's explanation for why scaling API requests is difficult, Stanley added the prompt "be more vulgar" — then posted his reaction on X. "Today I learned scaling API requests is like trying to keep up with a never-ending orgy."

Reacting to Stanley's experiment, Elon Musk posted, "Oh this is gonna be fun."

An anonymous reader quotes a report from NPR: Peru's Kimberly Garcia set a world record in her gold-medal winning turn at the women's 20 kilometer race walk event at the Pan American Games this weekend. Until she didn't. Once the race was over, organizers determined there was a serious "measuring problem" with the track, making the race times of Garcia, fellow medal winners Glenda Morejon of Ecuador and Peru's Evelyn Inga, and their competitors null and void. The athletes guessed the track had been drawn up roughly 3 kilometers (about 1.9 miles) shorter than it was supposed to be. Garcia crossed the finish line in 1 hour, 12 minutes and 26 seconds. The world record of 1 hour, 23 minutes and 49 seconds is held by China's Jiayu Yang. The athletes suspected something was amiss mid-race, according to the Associated Press.

The Santiago 2023 Corporation, the group in charge of the 2023 Pan American Games, placed the blame on the Pan American Athletics Association, which reportedly chose the person who measured the race course. In a statement following the race, Santiago 2023 said the official who measured the course "did not take accurate measurements of the route the athletes took during the race." The group continued, "We deeply regret the inconvenience for the athletes, their coaches, the public and the attending press, but this situation cannot be attributed to the Organizing Committee."

To explore America's "transition to a post-quantum world," the Washington Post interviewed U.S. federal official Nick Polk, who is focused on national security issues including quantum computing and is also a senior advisor to a White House federal chief information security officer): The Washington Post: The U.S. is in the early stages of a major shift focused on bolstering government network defenses, pushing federal agencies to adopt a new encryption standard known as post-quantum cryptography that aims to prevent systems from being vulnerable to advanced decryption techniques enabled by quantum computers in the near future...

Nick Polk: We've been using asymmetric encryption for a very long time now, and it's been ubiquitous since about 2014, when the U.S. government and some of the large tech companies decided that they're going to make it a default on most web browsers... Interestingly enough, regarding the post-quantum cryptographic standards being developed, the only thing that's quantum about them is that it has "quantum" in the name. It's really just a different type of math that's much more difficult for a quantum computer to be able to reverse-engineer. The National Institute of Standards and Technology is looking at different mathematical models to cover all their bases. The interesting thing is that these post-quantum standards are actually being used to protect classical computers that we have now, like laptops...

Given the breadth of the U.S. government and the amount of computing power we use, we really see ourselves and our role as a steward of the tech ecosystem. One of the things that came out of [this week's Inside Quantum Technology conference in New York City] was that we are very quickly moving along with the private sector to migrate to post-quantum cryptography. I think you're gonna see very shortly a lot of very sensitive private sector industries start to migrate or start to advertise that they're going to migrate. Banks are a perfect example. That means meeting with vendors regularly, and testing their algorithms to ensure that we can accurately and effectively implement them on federal systems...

The administration and national security memorandum set 2035 as our deadline as a government to migrate our [national security] systems to post-quantum cryptography. That's supposed to time with the development of operational quantum computers. We need to ensure that we start now, so that we don't end up not meeting the deadline before computers are operational... This is a prioritized migration for the U.S. government. We're going to start with our most critical systems — that includes what we call high-value assets, and high-impact systems. So for example, we're gonna prioritize systems that have personal health information.

That's our biggest emphasis — both when we talk to private industry and when we encourage agencies when they talk to their contractors and vendors — to really think about where your most sensitive data is and then prioritize those systems for migration.

The US v. Google antitrust trial is about many things, but more than anything, it's about the power of defaults. Even if it's easy to switch browsers or platforms or search engines, the one that appears when you turn it on matters a lot. Google obviously agrees and has paid a staggering amount to make sure it is the default: testimony in the trial revealed that Google spent a total of $26.3 billion in 2021 to be the default search engine in multiple browsers, phones, and platforms. From a report: That number, the sum total of all of Google's search distribution deals, came out during the Justice Department's cross-examination of Google's search head, Prabhakar Raghavan. It was made public after a debate earlier in the week between the two sides and Judge Amit Mehta over whether the figure should be redacted. Mehta has begun to push for more openness in the trial in general, and this was one of the most significant new pieces of information to be shared openly.

Just to put that $26.3 billion in context: Alphabet, Google's parent company, announced in its recent earnings report that Google Search ad business brought in about $44 billion over the last three months and about $165 billion in the last year. Its entire ad business -- which also includes YouTube ads -- made a bit under $90 billion in profit. This is all back-of-the-napkin math, but essentially, Google is giving up about 16 percent of its search revenue and about 29 percent of its profit to those distribution deals.

theodp writes: In July, Seattle-based and tech-backed nonprofit Code.org announced its 10th policy recommendation for all states "to require all students to take computer science (CS) to earn a high school diploma." In August, Washington State Senator Lisa Wellman phoned-in her plans to introduce a bill to make computer science a Washington high school graduation requirement to the state's Board of Education, indicating that the ChatGPT-sparked AI craze and Code.org had helped convince her of the need. Wellman, a former teacher who worked as a Programmer/System Analyst in the 80's before becoming an Apple VP (Publishing) in the '90s, also indicated that exposure to CS given to students in fifth grade could be sufficient to satisfy a HS CS requirement. In 2019, Wellman sponsored Microsoft-supported SB 5088 (Bill details), which required all Washington state public high schools to offer a CS class. Wellman also sponsored SB 5299 in 2021, which allows high school students to take a computer science elective in place of a third year math or science course (that may be required for college admission) to count towards graduation requirements.

And in October, Code.org CEO Hadi Partovi appeared before the Washington State Board of Education, driving home points Senator Wellman made in August with a deck containing slides calling for Washington to "require that all students take computer science to earn a high school diploma" and to "require computer science within all teacher certifications." Like Wellman, Partovi suggested the CS high school requirement might be satisfied by middle school work (he alternatively suggested one year of foreign language could be dropped to accommodate a HS CS course). Partovi noted that Washington contained some of the biggest promoters of K-12 CS in Microsoft Philanthropies' TEALS (TEALS founder Kevin Wang is a member of the Washington State Board of Education) and Code.org, as well some of the biggest funders of K-12 CS in Amazon and Microsoft -- both which are $3,000,000+ Platinum Supporters of Code.org and have top execs on Code.org's Board of Directors.

Matthew Sparkes reports via NewScientist: A prominent cryptography expert has told New Scientist that a US spy agency could be weakening a new generation of algorithms designed to protect against hackers equipped with quantum computers. Daniel Bernstein at the University of Illinois Chicago says that the US National Institute of Standards and Technology (NIST) is deliberately obscuring the level of involvement the US National Security Agency (NSA) has in developing new encryption standards for "post-quantum cryptography" (PQC). He also believes that NIST has made errors -- either accidental or deliberate -- in calculations describing the security of the new standards. NIST denies the claims.

Bernstein alleges that NIST's calculations for one of the upcoming PQC standards, Kyber512, are "glaringly wrong," making it appear more secure than it really is. He says that NIST multiplied two numbers together when it would have been more correct to add them, resulting in an artificially high assessment of Kyber512's robustness to attack. "We disagree with his analysis," says Dustin Moody at NIST. "It's a question for which there isn't scientific certainty and intelligent people can have different views. We respect Dan's opinion, but don't agree with what he says." Moody says that Kyber512 meets NIST's "level one" security criteria, which makes it at least as hard to break as a commonly used existing algorithm, AES-128. That said, NIST recommends that, in practice, people should use a stronger version, Kyber768, which Moody says was a suggestion from the algorithm's developers.

NIST is currently in a period of public consultation and hopes to reveal the final standards for PQC algorithms next year so that organizations can begin to adopt them. The Kyber algorithm seems likely to make the cut as it has already progressed through several layers of selection. Given its secretive nature, it is difficult to say for sure whether or not the NSA has influenced the PQC standards, but there have long been suggestions and rumors that the agency deliberately weakens encryption algorithms. In 2013, The New York Times reported that the agency had a budget of $250 million for the task, and intelligence agency documents leaked by Edward Snowden in the same year contained references to the NSA deliberately placing a backdoor in a cryptography algorithm, although that algorithm was later dropped from official standards.

An anonymous reader quotes a report from NPR: High school students' scores on the ACT college admissions test have dropped to their lowest in more than three decades, showing a lack of student preparedness for college-level coursework, according to the nonprofit organization that administers the test. Scores have been falling for six consecutive years, but the trend accelerated during the COVID-19 pandemic. Students in the class of 2023 whose scores were reported Wednesday were in their first year of high school when the virus reached the U.S.

The average ACT composite score for U.S. students was 19.5 out of 36. Last year, the average score was 19.8. The average scores in reading, science and math all were below benchmarks the ACT says students must reach to have a high probability of success in first-year college courses. The average score in English was just above the benchmark but still declined compared to last year.

About 1.4 million students in the U.S. took the ACT this year, an increase from last year. However, the numbers have not returned to pre-pandemic levels. [Janet Godwin, chief executive officer for the nonprofit ACT] said she doesn't believe those numbers will ever fully recover, partly because of test-optional admission policies. Of students who were tested, only 21% met benchmarks for success in college-level classes in all subjects. Research from the nonprofit shows students who meet those benchmarks have a 50% chance of earning a B or better and nearly a 75% chance of earning a C or better in corresponding courses. Further reading: Accounting Graduates Drop By Highest Percentage in Years