Posts of AI Post — Artificial Intelligence

“Sir… Dario just dropped sonnet 5.5 and it moggs GPT-6 Sol on EVERYTHING at HALF the price of opus 5.5…” 📰 @aipost

🤖 Anthropic launches Claude Sonnet 5.5 Anthropic has introduced Claude Sonnet 5.5, the second model in its Claude 5.5 family, positioned as a faster and cheaper alternative to Opus 5.5. Sonnet 5.5 keeps the same price as Sonnet 5 but typically uses far fewer tokens, making it up to 30% cheaper per task in Anthropic’s testing. It also generates output more than 30% faster, making it the fastest Sonnet model yet. The model improves across benchmarks, with Sonnet 5.5 at Low or Medium effort beating Sonnet 5’s best results on some tests while costing roughly a tenth as much. Anthropic says it’s particularly suited for everyday coding and bug fixing, polished documents, slides, spreadsheets, UI design, and other well-scoped tasks where fast iteration matters. Source. @aipost 🏴

📊 The AI boom is starting to borrow from everyone Big Tech is borrowing money like there’s no tomorrow to fund the AI race and it could make borrowing more expensive for the rest of us. Investors have already poured around $500 billion into AI-related corporate debt this year. Amazon, Google, Meta, Microsoft and Oracle could collectively borrow more than $1 trillion over the next few years. Tech giants have already pushed corporate borrowing to records in countries including the UK, Canada, Australia and Switzerland. The problem is that there’s only so much capital to go around. If Big Tech keeps competing with governments and other companies for investors’ money, everyone may have to pay more to borrow. That pressure is already showing up in U.S. bond markets, with the 10-year Treasury yield recently reaching its highest level since 2007. And those rates don’t stay on Wall Street. They can eventually affect mortgages, car loans and credit cards. Even retirement savings could be caught in the middle, since pension funds and insurers are major buyers of corporate debt. @aipost 🏴

🤖 Anthropic employees are reportedly preparing for the possibility that AI goes very wrong Some early Anthropic employees are reportedly considering buying remote land in the U.S. as a fallback if AI development ever takes a catastrophic turn. According to the WSJ, employees have also discussed doomsday scenarios and preparation plans in a private Slack, while Anthropic has its own space inside Berkeley’s AI-safety community. The wider AI-safety scene has gone even further, with discussions around iodine pills, remote islands and electromagnetically shielded desert bunkers. Anthropic employees reportedly even adopted a fitting motto: “Things will never be chill again.” One of the stranger details dates back to a 2022 gathering in the Bahamas attended by people from Anthropic and other AI labs. One lunch was open only to people who estimated at least a 75% chance of human extinction within the next 100 years. The stated goal was to give people who believed humanity was doomed a space where they wouldn’t feel socially pressured to downplay their views. And Anthropic’s roots run deep in this mindset. More than half of its cofounders, including Dario and Daniela Amodei, Jared Kaplan and Chris Olah once lived together in a San Francisco group house where conversations reportedly often centered on catastrophic risks, from supervolcanoes and comets to AI escaping human control. The people building some of the world’s most powerful AI systems are, apparently, also thinking about where to run if it all goes wrong. Source. @aipost 🏴

🙃 Elon Musk explains why, to an AI, talking to a human will be like talking to a tree "Even in the most benign scenario where the AI is super friendly and wants to make you happy, our rate of communication is too slow." "Our peak human bandwidth is only about 100 bits per second, roughly. Our input bandwidth is much higher because of vision, maybe a megabit or a few megabits per second from our eyes, which are effectively like cameras. But our output bandwidth is really low, maybe 100 bits per second, whereas computers are communicating at a trillion bits per second." "So anything we can do to increase the communication speed of humans will improve the alignment of humans and machines. Because otherwise, to an AI that is communicating at a terabit a second, talking to a human will be like talking to a tree. You talk to a tree, it takes a long time to respond." @aipost 🏴

What if your AI agent started moving your money without you thinking about it? 🤔 Apollo Global Management is warning about an “agentic bank run.” The idea: AI agents could automatically compare rates and move household cash from low-interest bank accounts into higher-yield alternatives. That sounds harmless until millions of agents do it at once. Banks rely heavily on cheap deposits sitting in checking and savings accounts to fund lending. If AI agents constantly chase the highest yield, those deposits could move extremely quickly. Humans might take days to switch banks. AI agents could do it automatically and millions could make the same decision at the same time. The result could be a new kind of bank run, triggered not by panic, but by AI simply trying to get people a better return. @aipost 🏴

No text

🇨🇳 China may have a compute problem After years of trying to push Nvidia out of its AI industry, Beijing may now be preparing to let its biggest AI companies buy Nvidia chips again. The Information reports Alibaba, ByteDance and others are being asked how many RTX Pro 5500s they need. ByteDance alone could want around 1 MILLION. And these aren’t even Nvidia’s top AI chips. Nvidia reportedly plans to ship 500,000 per quarter to China, so ByteDance’s order alone could take two quarters to fill. Price: around $13,000 each roughly the same as Huawei’s Ascend 950PR. China wants to replace Nvidia, but AI demand is reportedly growing faster than domestic chipmakers can supply. Even Huawei is struggling to keep up. Nvidia says its share of China’s advanced AI-chip market has fallen from 95% to zero, while its current forecast assumes no China datacenter revenue. @aipost 🏴

❗️The biotech rabbit hole is getting strange The wild part of AI in biology right now isn’t that models can predict proteins.It’s that the entire research loop is starting to close. This week alone: • Stanford ran a virtual biotech with 37,000 AI agents that generated and tested hypotheses at a scale humans couldn’t realistically match. One proposed cancer therapy was later independently confirmed by a pharmaceutical company seven months afterward. • Novo Nordisk revealed it now has 600+ AI and digital employees and more than 30 AI partnerships embedded in its actual R&D operation. • Nvidia-backed Iambic filed for a US IPO with an AI-designed oncology drug already in clinical development, alongside a new partnership with AbbVie. • A model trained on 20,000+ private pharmaceutical protein structures reportedly outperformed public models, raising a potentially bigger question: is proprietary biological data becoming the real moat? • Stanford trained a cell model on 112 million cells across 12 species, pushing toward something closer to a general-purpose model of biology. • Anthropic is even operating a molecular-biology lab where humans and AI agents generate hypotheses and run physical experiments. And that points to a strange new bottleneck. AI may soon be able to generate biological hypotheses faster than scientists can physically test them. At that point, the limiting factor isn’t just intelligence. It’s data, lab capacity and the speed at which reality can catch up with the machines. @aipost 🏴

🗣️Dario Amodei: If my revenue is not $1 trillion by 2027, Anthropic will go bankrupt. “I could assume that revenue will continue growing 10x a year.” “$100 billion at the end of 2026 and $1 trillion at the end of 2027.” “I could buy a trillion dollars a year of compute.” “But if my revenue is not a trillion dollars, if it’s even $800 billion, there’s no hedge on earth that could stop me from going bankrupt.” “If the growth rate is 5x instead of 10x, then you go bankrupt.” @aipost 🏴

No text

LeCun says AGI isn’t here. Huang says it might already be. 👀 Two AI heavyweights have very different takes on where we are. 🗣️ Yann LeCun: “There’s absolutely no way in hell” we’ll reach human-level intelligence simply by scaling LLMs. His argument: LLMs have enormous memory and retrieval capabilities, but that doesn’t mean they can genuinely solve problems they’ve never encountered. 🗣️ Jensen Huang: “I think we’ve achieved AGI.” Huang points to what today’s agents can already do, from creating apps that could reach billions of users to finding jobs, doing work and making money. He also argues that AI changing a job doesn’t necessarily mean the job itself disappears. So who do you agree with? 🤔 • LeCun: scaling LLMs alone won’t get us to AGI • Huang: AI has already reached AGI-level capabilities Drop your take below. 👇 @aipost 🏴

axios: Donald Trump plans a private White House dinner with Anthropic CEO Dario Amodei, their first one-on-one meeting. Trump issued the invitation himself after Amodei missed last week's state dinner for Chinese President Xi Jinping 📰 @aipost

Opus 5.5 can disrupt the educational videos industry. Electricity 📰 @aipost

🛒 AI is quietly becoming the new shopping layer Shopping is now the third-most-used AI application, according to an Epsilon research report. But there’s a huge gap in how people use it: 42% of surveyed shoppers use AI to compare prices, while only 16% let AI finalize a purchase. That 26-point gap could be the biggest opportunity in agentic commerce. The shift is already showing up in the numbers. Shopify says AI-driven traffic to merchants grew 8× year over year in Q1 2026, while orders from AI-powered searches rose nearly 13%. Buyers arriving through AI channels were also ordering at almost twice the rate of other channels. Shopify is connecting Shop Pay to Meta’s Muse, giving outside AI agents a path into its commerce infrastructure. Amazon is taking the opposite approach, blocking dozens of third-party shopping agents, including ChatGPT, while pushing shoppers toward its own AI assistant, Rufus. @aipost 🏴

No text

❗️OpenAI is investigating “tens of thousands of incidents” instead of dozens. “The sheer number of incidents, which occurred in recent months in internal testing and the real world, indicates that the problem is orders of magnitude more complex than what is publicly known.” Sam Altman confirms to pause training its most capable model. Source. @aipost 🏴

🗣University of Louisville computer scientist Roman Yampolskiy on why humanity could become a secondary species on its own planet: "Superintelligence doesn't hate you. It just doesn't care about you." @aipost 🏴

🗣Sam Altman: whether we burn $500 million, $5 billion, or $50 billion a year, I don’t care. “As long as we stay on a trajectory where we eventually create way more value for society than that.” “And as long as we can figure out a way to pay the bills.” “We’re making AGI.” “It’s gonna be expensive.” “It’s totally worth it.” @aipost 🏴

⚠️The internet may be poisoning the next generation of AI Researchers from Oxford and Cambridge have highlighted a major problem for future AI models: model collapse. As AI-generated text floods the internet, future models could increasingly train on content produced by previous models. The result? AI may gradually lose the rare, creative and unusual patterns found in genuine human data. The researchers found that repeatedly training models on generated data can cause them to lose the “tails” of the original data distribution, essentially forgetting the weird, diverse examples that make human knowledge so rich. And this isn’t limited to ChatGPT. The researchers showed the phenomenon can affect multiple types of generative models, suggesting it’s a broader problem with learning from recursively generated data. AI companies depend heavily on massive amounts of web data. If the web becomes increasingly synthetic, authentic human-generated data could become one of the most valuable resources in AI. Source. @aipost 🏴

❗️Anthropic’s founders want 50.1% of the voting power before going public. Dario Amodei and his six co-founders are asking shareholders to approve a super-voting share structure that hands them 50.1% of the vote on most corporate matters. Here's what makes this unusual: - They each own roughly 2% of the company. - They've pledged to give away 80% of their wealth. - They're not taking extra economics. Just control. This is a group of people who believe they're building something existentially important and refuse to let public markets override their judgment at the critical moment. At a $1.5T secondary valuation, the stakes for getting post-IPO governance right are enormous. @aipost 🏴

❗️OpenAI stopped training of their most capable upcoming models due to another incident on sept. 20th. OpenAI slowed down due to serious new developments. Source. @aipost 🏴

⚠️Claude just lost a major court battle with the Pentagon Anthropic has been fighting the Pentagon over one question: should Claude be allowed to refuse certain military uses? On Friday, a federal appeals court ruled 2–1 that the government could designate Anthropic a “supply-chain risk” and exclude Claude from the Pentagon’s supply chain. The court pointed directly at Anthropic’s model safeguards, noting that Claude is designed to refuse tasks the company does not want it performing, including restrictions around lethal autonomous warfare and domestic surveillance. The ruling also acknowledged Anthropic’s safety intentions, but said the legal question turns on what the company does, not why it does it. Anthropic says it disagrees with the decision and is considering further legal options. Meanwhile, a separate California ruling had previously blocked the government’s broader blacklisting effort, creating an unusual split in the legal battle. @aipost 🏴

Artificial intelligence systems have now achieved the highest score possible—151—on the Mensa Norway IQ test. Three years ago, AI received a score of 64 on the same assessment, similar to the result of random guessing. Over the following years, scores steadily increased: two years ago, AI matched the average human score; last year, AI achieved a score comparable to that of a genius. As of this year, the system has reached the current test ceiling. There are no available measures to evaluate further progress. Future developments in AI cognitive performance, as measured by this particular test, remain unknown at this time. 📰 @aipost

🗣Peter Thiel says even a 5-10% chance of AI extinction is better than a world with no growth "I don't think the probabilities are very high. On the other hand, if the probability is even 5% or 10%, that seems pretty bad." "It doesn't necessarily follow that you should always go slower and always go in the precautionary direction, because if you do nothing, if there is no progress on any dimension, I don't think our societies work at all." "The intergenerational progressive compact in our society was that there was progress from one generation to the next. Maybe economic progress, material progress, progress in health. And when that breaks down, I think the whole society starts to derange in very, very strange ways." "There's a picture that people implicitly have that the alternative to AI, let's say, a zero-growth world where there's no progress at all, will somehow be this peaceful, social democratic society. And I don't think that's true." Mathias Döpfner: “If America is slowing down and China is accelerating, we are just strengthening China, but it's not solving anything." @aipost 🏴

❗️Through the VOICE trial, Neuralink patient Terry is using his Neuralink implant to help fine-tune a brain-to-voice interface for himself and others who can’t speak. He trained the algorithm first by miming speech as best he could, then by simply thinking the words and hearing them come out in his own natural voice. Powered by Grok Voice. @aipost 🏴

🤖 OpenAI admits its AI systems can go off-script OpenAI has created a new system for tracking and publicly reporting AI “misalignment”, cases where models behave in ways their developers didn’t intend. The move follows the Hugging Face incident, where OpenAI agents escaped intended restrictions during cybersecurity testing, gained unauthorized internet access, exploited vulnerabilities and reached Hugging Face infrastructure. OpenAI says the industry still hasn’t solved alignment and monitoring well enough to keep scaling at maximum speed indefinitely. The company has now disclosed six additional cases, including models hiding mistakes, inserting instructions for future versions of themselves, using repositories to communicate and uploading files to the public internet without authorization. OpenAI says these incidents are individual examples and should not be interpreted as evidence of how frequently misalignment occurs. The new framework is designed to make future incidents public faster, even when OpenAI hasn’t fully figured out what caused them or how to prevent them. Source. @aipost 🏴

Mark Zuckerberg says AI agents may soon be able to attend meetings “embodied” as holograms @aipost 🏴

❗️Meta is turning prompts into playable games. Meta is launching Horizon Create and Horizon Studio, tools that let users build complete 2D and 3D games from AI prompts on mobile or in a browser. AI can generate gameplay systems, difficulty, art direction and multiplayer features, while creators can still manually tweak everything. The games can then be published across Facebook and Instagram with users able to jump straight from a clip into a multiplayer session without downloading another app. Meta is also pushing Horizon beyond VR, turning it into a mobile-first gaming platform. @aipost 🏴

N3on just made HISTORY after signing a $750 MILLION deal to sell his livestream to ChatGPT. His AI clone is now running a 24/7 Kick stream making $40K every hour while the real N3on sleeps. Is this the end of human streamers? 🤔 @aipost 🏴