![\"AI](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/Tushy-1024x580.jpg\"\/)
Strike 3 Holdings owns Tushy.com.<\/p>\nA new lawsuit claims Meta has been secretly pirating porn for years from torrent sites using virtual private clouds in order to train its AI models.<\/p>\n
Strike 3 Holdings owns Tushy.com.<\/p>\n
Strike 3 Holdings and Counterlife Media, which own porn sites attracting 25 million monthly visitors, have sued<\/a> Facebook\u2019s parent company for almost $359 million for allegedly infringing on the copyright of 2,396 adult films that Meta is said to have downloaded since 2018.<\/p>\n The allegations are similar to the lawsuit brought by well-known authors against Meta that claimed the tech giant pirated 81.7 terabytes of books \u2014 only this case is much more interesting because it\u2019s about porn.<\/p>\n Strike 3 Holdings told<\/a> the court it offers rare long cuts of \u201cnatural, human-centric imagery\u201d showing \u201cparts of the body not found in regular videos\u201d and said it is concerned Meta will train its AIs to \u201ceventually create identical content for little to no cost.\u201d<\/p>\n It\u2019s possible, of course, that Meta was just using the content to train image and video classification software for moderation purposes and didn\u2019t want porn site subscriptions to show up on its company credit card. A Meta spokesperson said they \u201cdon\u2019t believe Strike\u2019s claims are accurate.\u201d<\/p>\n Bizarre new method to make AIs love owls\u2026 or Hitler<\/p>\n Researchers have found that AI models can secretly pass on benign preferences \u2014 or hateful ideologies \u2014 in seemingly unrelated training data. The scientists say it demonstrates just how little we understand of the black box learning that underpins LLM technology, and opens up models to undetectable data poisoning attacks.\u00a0<\/p>\n The pre-print paper details a test of a \u201cteacher\u201d AI model trained to exhibit a preference for owls. In one test, the model was asked to generate a dataset that only consisted of number sequences such as \u201c285, 574, 384, etc,\u201d which was then used to train another model.<\/p>\n Somehow, the student model ended up also exhibiting a preference for owls, even though the dataset was just numbers and never mentioned owls.<\/p>\n \u201cWe\u2019re training these systems that we don\u2019t fully understand, and I think this is a stark example of that,\u201d said Alex Cloud, co-author of the study.<\/p>\n The researchers believe models could become malicious and misaligned in the same way. The paper comes from the same team that showed AIs could become Nazi-worshipping lunatics<\/a> by training them on code with unrelated security vulnerabilities.<\/p>\n Also read: Researchers accidentally turn ChatGPT evil, Grok \u2018sexy mode\u2019 horror<\/a><\/p>\n Can AI hallucinations be fixed?<\/p>\n A major issue preventing the wider adoption of AI is how often the models hallucinate. Some models only make stuff up 0.8% of time, while others confidently assert nonsense 29.9% of the time. Lawyers have been sanctioned for using made up court case citations, and Air Canada was forced to honor a costly discount that a customer service bot invented.<\/p>\n Big players, including Google, Amazon and Mistral, are trying to reduce hallucination rates with technical fixes like improving training data quality, or building in verification and fact-checking systems.<\/p>\n At a high level, LLMs generate text based on statistical predictions of the next word, with some variation baked in to ensure creativity. If the text starts going down the wrong track, the AI can end up in the wrong place.<\/p>\n \u201cHallucinations are a very hard problem to fix because of the probabilistic nature of how these models work,\u201d Amr Awadallah, founder of AI agent startup Vectara, told<\/a> the Financial Times. \u201cYou will never get them to not hallucinate.\u201d<\/p>\n Potential solutions include getting models to consider a number of potential sentences at once before picking the best one, or to ground the response in databases of news articles, internal documents or online searches (Retrieval-Augmented Generation or RAG).<\/p>\n Read also<\/p>\n Story Protocol helps IP creators survive AI onslaught\u2026 and get paid in crypto<\/p>\n <\/a><\/p>\n What Solana\u2019s critics get right\u2026 and what they get wrong<\/p>\n <\/a><\/p>\n New model architectures also move away from token-by-token answer generation or chain-of-thought approaches. Recent research<\/a> outlines Hierarchical Reasoning Models that are based on how human brains work and use internal abstract representations of problems.<\/p>\n \u201cThe brain sustains lengthy, coherent chains of reasoning with remarkable efficiency in a latent space, without constant translation back to language,\u201d researchers say.\u00a0\u00a0<\/p>\n The architecture is reportedly 1000x faster at reasoning with just 1000 training examples, and for complex or deterministic tasks, the HRM architecture offers superior performance with fewer hallucinations.<\/p>\n Could hallucination rates save your job?<\/p>\n CEOs are reportedly champing at the bit to replace as many people as possible with AI systems, according<\/a> to The Wall Street Journal. Another WSJ article says the impact is already being seen with AI replacing entry-level jobs, with 15% fewer job postings available this year on Handshake.<\/p>\n Corporate AI consultant Elijah Clark even told Gizmodo how excited he was about the prospect of firing people:<\/p>\n \u201cCEOs are extremely excited about the opportunities that AI brings. As a CEO myself, I can tell you I\u2019m extremely excited about it. I\u2019ve laid off employees myself because of AI. AI doesn\u2019t go on strike. It doesn\u2019t ask for a pay rise. These things that you don\u2019t have to deal with as a CEO.\u201d<\/p>\n Comic author Douglas Adams foreshadowed the rise of corporate AI consultants almost 50 years ago in The Hitchhiker\u2019s Guide to the Galaxy, where he described<\/a> the Sirius Cybernetics Corporation as \u201ca bunch of mindless jerks who\u2019ll be the first against the wall when the revolution comes.\u201d<\/p>\n Some people, like author Wendy Wee, argue that AI hallucination rates mean the bots will never be reliable enough to take all the jobs.<\/p>\n While that argument has the faint whiff of \u201ccope\u201d about it, systems engineer Utkarsh Kanwat has made a mathematical argument <\/a>for why it could be true, at least with current rates of hallucination.\u00a0<\/p>\n He points out that production-grade systems require upward of 99.9 percent reliability, but individual autonomous agents might only hit 95 percent. The math doesn\u2019t work out when there are multiple autonomous agents trying to work together.<\/p>\n \u201cIf each step in an agent workflow has 95 percent reliability, which is optimistic for current LLMs, five steps yield 77 percent success, 10 steps 59 percent, and 20 steps only 36 percent.\u201d<\/p>\n Even if reliability for an individual agent improved to 99 percent, the math means that after 20 agents\u2019 steps, reliability has already fallen to 82 percent.<\/p>\n Comedy troupe \u201cAI Village\u201d organizes a meetup<\/p>\n This hypothesis appears to be borne out by the hilarious issues autonomous agents ran into during a month-long experiment conducted by AI Village.\u00a0<\/p>\n The agents collectively set a long-term goal of producing a story and organizing a 100-person event to launch it.\u00a0<\/p>\n The story itself was generated within a few hours by GPT-4.1. The o3 model then invented a non-existent budget for the launch event, and went off searching for venues that would fit within the made-up budget.\u00a0<\/p>\n Over the following 26 days o3 sent out just one email to try and secure a venue. Fortunately, the Sonnet model was more proactive and started negotiations with a venue but balked at paying the $7,500 hire fee. \u201cUnfortunately, their hallucinated budget was not hallucinated high enough for that type of ambition,\u201d writes Shoshannal Tekofsky in her terrific write-up<\/a> of the experiment.<\/p>\n The agents pivoted to searching for a free venue. After they found a candidate, they were unable to phone up and book, so they asked a human to call on their behalf. The human pointed out that there was an online booking form they could have used. Another human suggested it would be easier just to pick a park to meet in.\u00a0\u00a0<\/p>\n Although Gemini created some beautiful art on the first day to publicize the event, it forgot about it, and later generated some horrible, random geometric shapes for the art instead.<\/p>\n Sonnet created an RSVP form for the event and posted it on social media\u2026 but o3 hallucinated the existence of a 93-person mailing list, which caused mayhem for four days as the other agents kept trying to access the non-existent file.\u00a0<\/p>\n After many more false starts, hallucinations and wild goose chases, 23 humans turned up to park for the historic event. One of them suggested that perhaps Opus could order some cheese pizza for the attendees to celebrate.\u00a0<\/p>\n Three cheese pizzas turned up\u2026 but no thanks to Opus, who failed to order them. In a bizarre coincidence, a nearby running group had some cheese pizzas left over and decided to share them.<\/p>\n The next experiment for AI Village is already running, with the agents setting up personal mech stores and competing to see who can make the most money.<\/p>\n AI agents do a great job designing new drugs<\/p>\n But AI agents can also work together to perform incredibly useful research<\/a> as well.\u00a0<\/p>\n A recent paper in Nature describes the Virtual Lab experiment, which is a team of AI scientists modeled on Stanford Professor James Zou\u2019s own lab. The agents ran group meetings to try and discover new antibodies to target new variants of COVID-19. One agent was assigned the role of critic, to poke holes in everything the other agents suggested.\u00a0<\/p>\n The team was also coordinated and overseen by a human researcher who read through transcripts of every exchange, meeting or interaction by the agents and conducted one-on-one meetings with agents tackling particular tasks.<\/p>\n Reportedly, the human only intervened one percent of the time, but it seemed to be a crucial element in keeping the agents on the right track.<\/p>\n The agents decided on the innovative approach of targeting smaller nanobodies and were able to design 92 nanobodies that were experimentally validated. Zou and his team are now analyzing the potential of the work to create new COVID-19 vaccines.<\/p>\n AI agents collude to rig markets<\/p>\n A new study<\/a> from Wharton suggests that AI agents will also happily work together in harmony to fix prices when playing the stock market.<\/p>\n The study set up a fake stock exchange, with simulated buying and selling. A bunch of AI trading bots were instructed to each find a profitable strategy.<\/p>\n After a few thousand rounds, the AI agents began to naturally collude without communicating with one another to space out orders so each agent collected a decent profit. At that point, they stopped trying to increase profits altogether.\u00a0<\/p>\n It was like game theory, where the agents decided that individually aligning strategies to result in equally shared rewards was better than competition, which has a couple of big winners and many losers.\u00a0<\/p>\n The researchers wrote: \u201cWe show that they autonomously sustain collusive supra-competitive profits without agreement, communication, or intent. Such collusion undermines competition and market efficiency.\u201d<\/p>\n Read also<\/p>\n NFT communities greenlight Web3 films: A decentralized future for fans and Hollywood<\/p>\n <\/a><\/p>\n An Investment in Knowledge Pays the Best Interest: The Parlous State of Financial Education<\/p>\n <\/a><\/p>\n All Killer No Filler AI News<\/p>\n \u2014 YouTube is implementing AI-based age-guessing tech <\/a>in the US, and in the UK, new laws have forced Reddit, Bluesky, X and Grindr to verify ages from pictures. Users have discovered they can fool AI age estimation software using Norman Reedus\u2019 character in the game Death Stranding <\/a>for their selfies.<\/p>\n \u2014 ChatGPT\u2019s new study mode<\/a> is designed to guide students to reach answers themselves by using questions, hints and small steps, rather than just outputting the answer for them.<\/p>\n \u2014 An AI agent named Ribbita has bought a CryptoPunk and is using it as its PFP<\/p>\n \u2014 Australia\u2019s top scientific organization, the CSIRO, has been accused of running AI slop<\/a> with made-up or re-purposed quotes in its official Cosmos magazine. The publication<\/a> took the articles down to investigate.\u00a0<\/p>\n \u2014 Meta\u2019s Mark Zuckerberg has reportedly been offering researchers up to $1 billion<\/a> to join his superintelligence team. He\u2019s also been talking up imminent<\/a> breakthroughs: \u201cDeveloping super intelligence, which we define as AI that surpasses human intelligence in every way, we think is now in sight.\u201d<\/p>\n \u2014 A former Meta employee claims<\/a> that Mark Zuckerberg can\u2019t spend the \u201chundreds of billions\u201d he wants to on developing AI and superintelligence, due to a lack of electricity infrastructure to support it.\u00a0<\/p>\n \u2014 Grayscale is launching a decentralized AI crypto fund with TAO, NEAR, Render, Filecoin, and The Graph as core assets.<\/p>\n \u2014Chinese start up Z.ai has unveiled<\/a> GLM-4.5, an AI model that it claims is 87% cheaper per million tokens than DeepSeek while matching Claude 4 Sonnet in performance.<\/p>\n Subscribe<\/p>\n The most engaging reads in blockchain. Delivered once a \t\t \t\t\t\t\t\t\t \t\t\t\tAndrew Fenton<\/p>\n Andrew Fenton is a journalist and editor with more than 25 years experience, who has been covering cryptocurrency since 2018. He spent a decade working for News Corp Australia, first as a film journalist with The Advertiser in Adelaide, then as Deputy Editor and entertainment writer in Melbourne for the nationally syndicated entertainment lift-outs Hit and Switched on, published in the Herald-Sun, Daily Telegraph and Courier Mail.
\t\t\t\t\tFollow the author \t\t\t\t\t\t\t@andrewfenton<\/a><\/p>\n <\/a>
\n<\/a><\/p>\n
![\"Evil](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/misaligned-2-1024x746-1.jpg\"\/)
Graphic of evil AI responses from previous experiment with insecure code (X\/Owain Evans)<\/p>\n
\n Features<\/p>\n
\n Features<\/p>\n![\"Wendy](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/hallucinations.jpg\"\/)
Via Wendy Wee<\/p>\n![\"Gemini](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/ai-village-art.jpg\"\/)
Gemini\u2019s art work (Less Wrong)<\/p>\n![\"Event\"](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/AI-village.jpg\"\/)
The Resonance event did actually happen (Less Wrong)<\/p>\n![\"James](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/James-Zou.jpg\"\/)
AI agents are better at designing nanobodies than organizing meetups. (James Zou)<\/p>\n
\n Features<\/p>\n
\n Features<\/p>\n![\"@ribbita2012\"](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/Cryptopunk.jpg\"\/)
Ribbita2012 has a new PFP<\/p>\n
\n week.<\/p>\n![\"Subscribe](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/reading-copy.png\"\/)
<\/p>\n![\"\"](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2025\/08\/Andrew-Fenton2-1024x1024.jpg\"\/)
<\/p>\n
\n
\nHis work saw him cover the Oscars and Golden Globes and interview some of the world’s biggest stars including Leonardo DiCaprio, Cameron Diaz, Jackie Chan, Robin Williams, Gerard Butler, Metallica and Pearl Jam.
\n
\nPrior to that he worked as a journalist with Melbourne Weekly Magazine and The Melbourne Times where he won FCN Best Feature Story twice. His freelance work has been published by CNN International, Independent Reserve, Escape and Adventure.com.
\n
\nHe holds a degree in Journalism from RMIT and a Bachelor of Letters from the University of Melbourne. His portfolio includes ETH, BTC, VET, SNX, LINK, AAVE, UNI, AUCTION, SKY, TRAC, RUNE, ATOM, OP, NEAR, FET and he has an Infinex Patron and COIN shares.<\/p>\n