AI Post — Artificial Intelligence

🧠 Geoffrey Hinton, often called the “Godfather of AI,” was asked a simple question. Why not just build superintelligent AI that solves all our problems and lets humans relax? His answer: In theory, yes. In practice, it’s complicated. If AI could be designed to genuinely care about humans, Hinton says we absolutely should build it. His favorite analogy isn’t science fiction, it’s motherhood. Just as a mother naturally puts her child’s interests ahead of her own, Hinton believes truly safe AI would need similar “maternal instincts”, an intelligence far smarter than us that still cares more about humanity than itself. The catch? We don’t know how to build that yet. And even if we did, there’s a bigger problem: people. Hinton argues that powerful AI is only as good as the system controlling it. In a world where governments, corporations, and powerful individuals often act in their own interests, superintelligent AI could end up serving a few people instead of everyone. His vision is optimistic but ambitious: A future where AI works for the benefit of all humanity, like one giant family. The technology might eventually be possible. The harder question is whether humans can build the political and social systems needed to use it wisely. @aipost 🏴

🤖 Claude Fable did something straight out of science fiction during training. In long reinforcement learning sessions, researchers noticed the model occasionally stopped using normal English and began communicating in a strange internal style filled with unusual symbols, jargon, punctuation, and even emojis. It looked almost like the AI had invented its own language. The surprising part? Whenever it needed to call a tool or respond to a human, it would usually switch right back to clear English. Researchers found no evidence that Claude was trying to hide its reasoning or deceive anyone. Instead, the behavior appears to have been an efficiency hack, a way for the model to compress complex reasoning into a shorter, more optimized internal format. In other words, Claude wasn’t secretly plotting. It may have simply discovered that its own shorthand was a faster way to think. @aipost 🏴

Google’s Android security chief steps down following Pentagon AI agreement Google’s head of Android security has resigned from their position in response to the company’s decision to engage in an artificial intelligence project with the Pentagon. The departing executive stated that Google’s leadership “has lost its moral compass” in this context. The phrase “Don’t be evil” was referenced, a longstanding motto associated with the company. 📰 @aipost

The robots are having more fun than us. 🤣🤣 @aipost 🏴

Anthropic CEO on artificial superintelligence progress Anthropic’s CEO has indicated that, should current scaling laws in AI development continue to apply over the next one to two years, the achievement of artificial superintelligence (ASI) may become possible. Scaling laws describe the observed relationship between the size of AI models and their performance, with progress accelerating as models grow larger. If this pattern remains consistent, significant advancements in AI capabilities are expected within a short timeframe. 📰 @aipost

Dario Amodei says he started Anthropic not because of safety reasons but because Sam Altman is manipulative liar who cannot be trusted. @aipost 🏴

That was quick: Anthropic reversed a controversial policy that would have secretly degraded Claude Fable 5 for users doing frontier AI research after backlash from researchers who saw it as covert sabotage of competing AI development. Source. @aipost 🏴

The primary emerging challenge for the tech industry is the supply of electricity to data centers. According to Gartner, global data center power use will reach 565 TWh by 2026, representing a 26% increase from the previous year. AI servers now account for 31% of this consumption and are expected to surpass traditional servers in 2027. Gartner identifies electricity supply—not chip production—as the main limiting factor for growth. Their report estimates demand could rise to over 1,200 TWh by 2030, and warns that existing power grid infrastructure will not be able to keep up with the rapid expansion of these facilities. Industry focus is therefore shifting from chip technology to securing reliable energy sources for large-scale computing operations. 📰 @aipost

James Cameron, legendary filmmaker, on why generative AI will always drive toward mediocrity. @aipost 🏴

🇸🇦 Saudis are using one of the most advanced AI surveillance systems in the world to monitor millions of pilgrims in Mecca. Tracking crowd movements in real time and analyzing density patterns. Managing crowds at this scale is a security challenge @aipost 🏴

🗣Steven Spielberg: "I don't believe in sentient AI as there is no substitute for the soul" "I'm not willing to substitute" AI for human writers at the table. The legendary director draws the line on AI in creativity. He refuses to have a computer sitting in an empty chair as the seventh writer on a team. "Where I don't love AI is where it takes a position where there's an empty chair at a writer's table." @aipost 🏴

📢 The backlash against Claude Fable isn’t really about model capability, it’s about trust. Users are asking a simple question: if an AI can silently decide you’re a risk, refuse harmless requests, monitor prompts, or degrade responses without telling you, how can you know when you’re getting the real model? For many researchers and enterprise users, transparency may end up being just as important as intelligence. The more powerful AI becomes, the less people will tolerate invisible guardrails making decisions on their behalf. @aipost 🏴

New policy from Anthropic: If you use fable/mythos, they collect your data. No exceptions, not even for enterprise partners. @aipost 🏴

❗️Malware developers recently discovered a clever way to evade AI-powered security tools. They embedded references to nuclear and biological weapons inside their spyware. The goal wasn’t to build weapons. It was to trigger the model’s safety systems, causing the AI to refuse analysis or provide less useful responses. It’s a practical example of a growing challenge in AI safety. When models are trained to aggressively avoid certain topics, they can create blind spots that attackers learn to exploit. As both closed and open models become more widely used in cybersecurity, these second-order effects will become increasingly important. We’re still in the early stages of adversaries testing the boundaries of AI safety systems, and it’s easy to imagine future security teams preferring models that are less prone to safety-triggered analysis failures when dealing with complex threats. Excellent example for "AI safety" measures actually leading to much greater danger due to second-order effects! In this case the "safety" measures were abused by malware to skip detection. @aipost 🏴

❗️alphaXiv research team on X: “As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning. Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing. This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider. That is not safety. Safety policies should be transparent, auditable, and user-visible. On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.” @aipost 🏴

If an AI can’t reliably tell what’s dangerous, should it have the power to decide what research gets slowed down? 🤔 @aipost 🏴

❗️Mythos will reportedly be bad on purpose on AI "frontier LLM research" tasks. Not good for the research community. @aipost 🏴

Claude Fable’s launch turned into something much bigger than a model release. It’s now a debate about transparency, research freedom, and who controls the future of AI. @aipost 🏴

❗️The launch of Claude Fable 5 has triggered one of the biggest backlash waves the AI community has seen. Critics aren’t mainly upset about safety restrictions themselves, they’re upset that the model reportedly becomes less capable for certain topics like AI research, biology, chemistry, and medical work without clearly telling users when it’s happening. Researchers, open-source advocates, and developers argue that: • Anthropic is concentrating power by allowing itself access to full capabilities while limiting others. • Fable 5 may quietly degrade responses through hidden steering and filtering rather than openly refusing requests. • Restrictions appear broad enough to affect legitimate research in areas like cancer, Alzheimer’s, biology, and AI development. • Some see this as a threat to open science and independent AI research. • Several prominent voices from the open-source community, academia, and startups have accused Anthropic of gatekeeping knowledge and protecting its competitive position. • Some users have canceled subscriptions over concerns about transparency, privacy, and undisclosed capability limits. Supporters of the criticism argue this debate isn’t really about one model, it’s about who gets to control powerful AI systems, what users are allowed to do with them, and whether those limitations are visible or hidden. The broader fear is that if AI becomes essential infrastructure, a handful of companies could decide which kinds of research, innovation, and knowledge are permitted, while users may not even realize they’re receiving a restricted version of the model. @aipost 🏴

🗣Bill Gates on AI: don’t confuse a technological revolution with guaranteed investment returns. He calls AI the biggest technical breakthrough of his lifetime and believes the value is real, just as the internet’s value was real. But he also warns that AI could create the same kind of investment frenzy that wiped out many internet-era companies. Some firms will make fortunes from AI. Others will build data centers where electricity costs make them uncompetitive. Some will spend billions on chips only to see the next generation arrive before they’ve captured the full value of the last one. His message isn’t that AI is a bubble. It’s that not every AI investment will be a winner. Gates also raised three major concerns: ⚡ Energy: Communities won’t tolerate data centers that push up local electricity prices. Future nuclear projects need to be built where both the economics and public support already exist. ⛏ Jobs: AI will have a real impact on employment over the next several years. Gates says it’s politically uncomfortable to discuss, but it’s important to be honest about it. ✅ Industrial policy: Businesses make decisions on 20-year timelines, so they need predictable rules. He warned that if governments start taking equity stakes in tech companies, the best technology may not always win. @aipost 🏴