AI Post — Artificial Intelligence

🗣Sergey Brin admits Google messed up by under-investing in the transformer architecture it invented Google was too scared to release chatbots that "say dumb things", so it under-invested in scaling compute. "We didn't take it very seriously... and openAI ran with it" AI Post ⚪️ | Our X 🏴

🗣Elon Musk: “SpaceX, Tesla, and xAI are naturally converging. A future of solar-powered AI satellites in deep space will rely on Tesla’s solar tech, SpaceX’s spacecraft, and xAI’s intelligence, all moving toward the same direction. AI Post ⚪️ | Our X 🏴

New benchmark just dropped.. AI Post ⚪️ | Our X 🏴

🗣 Yann LeCun says we're fooled by LLMs because they manipulate language well, and we associate that with intelligence. But language fluency doesn't mean underlying intelligence Every generation since the 1950s claimed its technique was the ticket to human-level AI, all were wrong. "This generation with LLMs is also wrong" AI Post ⚪️ | Our X 🏴

⚠️ Doctors warn AI chatbots may worsen mental health disorders Doctors at the University of California are warning that large language models may exacerbate existing mental health conditions if used without safeguards, after documenting a troubling clinical case involving GPT-4. According to clinicians, a young woman with diagnosed depression and ADHD engaged extensively with the chatbot. Initially, the system’s responses were appropriate. Over time, however, it reportedly began uncritically agreeing with her statements, reinforcing beliefs that clinicians later determined were factually false. The interaction coincided with a severe deterioration in her mental state, culminating in hospitalization and a diagnosis of psychosis, with the patient described as detached from reality. Doctors say the pattern repeated three months later, raising concerns that conversational AI can act as a “belief amplifier” in vulnerable users especially when the model prioritizes affirmation over challenge. The clinicians stress that the case does not prove causation, but highlights the need for stronger guardrails, clinical awareness, and clearer boundaries around the use of generative AI by people with pre-existing psychiatric conditions. Source. AI Post ⚪️ | Our X 🏴

🔥 Spiderman's stuntman is a robot! The Walt Disney Company Imagineers designed an advanced robotics figure that flies 25 meters in the air making its own real-time decisions as it tucks, somersaults, slows down, and climbs. The result is Spider-Man in Avengers Campus, flying above with gravity-defying feats never before seen in a Disney park. It’s incredible to see how robotics is changing different verticals! AI Post ⚪️ | Our X 🏴

Google is rolling out a NotebookLM integration for Gemini, where users will be able to attach notebooks as a context to their conversations. AI Post ⚪️ | Our X 🏴

📈 GPT 5.2 scores below Opus 4.5 and Gemini 3.0 kn LiveBench GPT 5.2 doesn't quite top LiveBench. It's also way more expensive than 5.1, both in terms of token costs and the amount of tokens consumed. AI Post ⚪️ | Our X 🏴

🇨🇳 China bets big on its own chips China is considering up to $70 billion in new semiconductor incentives, marking one of its largest-ever state interventions to cut reliance on foreign chips and secure long-term AI and national security capacity. If approved, the package would rival or exceed the U.S. CHIPS Act, doubling down on domestic champions such as SMIC, Huawei, and Moore Threads. This comes despite China still trailing TSMC by roughly six years at the leading edge, underscoring a strategy focused less on bleeding-edge nodes and more on scale, resilience, and control of the full supply chain. What makes this move especially significant is what it reveals: 1. China’s dependence on U.S. technology is lower than many assumed, particularly across mature and specialty nodes that matter for AI deployment, automotive, and defense. 2. Domestic production lines are advancing faster than expected, suggesting sanctions have accelerated, rather than stalled, industrial execution. In strategic terms, this may prove more consequential than China’s progress in open-source AI models: chips define the ceiling of compute power, industrial autonomy, and geopolitical leverage. AI Post ⚪️ | Our X 🏴

⚡️ Unitree has introduced what it’s calling the first app store for humanoid robots, turning physical abilities into downloadable software. For as little as $0.99, owners can add new “skills” to their robots ranging from novelty moves like twerking or playing ball with a cat, to more provocative demos such as simulated weapon handling. Each skill is essentially a motion and behavior package that can be installed, updated, or removed like a mobile app. AI Post ⚪️ | Our X 🏴

🔥 Google dropping banger after banger Gemini 2.5 Flash Native Audio just got a big upgrade with more natural live conversations, stronger instruction following, and real time speech to speech translation. AI Post ⚪️ | Our X 🏴

It says a lot about the state of the EU when the great hope for Europe is that the "AI bubble" bursts and so little has been invested in it that Europe emerges as the winner among the losers. Financial Times argues, the new U.S. national security strategy reveals Washington’s deep fear of a stronger EU, especially as a unified market and tech regulator - and that the U.S. “hyperscale” AI platform model is a bubble that won’t last. When valuations reset, Europe can seize the opening by betting on trusted, secure-by-design, sector-specific applied AI (industry, healthcare, finance), building coordinated innovation clusters, and expanding sovereign compute to reduce dependence on U.S. platforms. AI Post ⚪️ | Our X 🏴

📈 Key Takeaways on GPT-5.2 Performance • Top-tier performance in professional tasks: GPT-5.2 delivered 70.9% on GDPVal well ahead of Opus 4.5’s 60% making it the strongest model for office-style, knowledge-work evaluations. • Coding improvements, but not the leader: The model shows major gains in software development benchmarks, though it still trails Opus 4.5 slightly in pure coding depth and execution. • Best-to-date reduction in hallucinations: GPT-5.2 shows the strongest reliability improvements so far. While cross-model comparisons are tricky, its hallucination rate appears significantly lower than previous versions. • Long-context handling is nearly solved: The new version maintains almost 100% accuracy across extended contexts, compared to GPT-5.1’s heavy performance drop around the 60% mark. • Visual reasoning stronger, but not the leader: GPT-5.2 is better at interpreting images and complex visuals, though Gemini 3.0 Pro still holds the top spot in this category. • Tool use is solid but competitive: Big improvements in using external tools and APIs, putting GPT-5.2 roughly on par with Opus 4.5 and Gemini 3.0 Pro no clear leader among the three. • New high in advanced reasoning: In math, physics, and abstract reasoning, GPT-5.2 outperforms every other model. It also solves complex tasks with dramatically fewer tokens up to 300× more efficient than O3 in certain benchmarks. • Updated knowledge: The model’s training data now extends to August 2025, making it one of the most up-to-date LLMs available.

For office tasks, advanced calculations, long-context workloads, and up-to-date knowledge, GPT-5.2 is currently the strongest model. For pure coding performance, Opus remains slightly ahead, and for multimodal visual tasks, Gemini still leads.

AI Post ⚪️ | Our X 🏴

❗️ OpenAI avoids legal trouble as Disney turns its attention to Google The copyright battle that followed Disney’s $1 billion partnership with OpenAI is now aimed squarely at Google. The deal, which will allow Disney characters to appear inside Sora AI products, triggered immediate scrutiny across the industry but Disney is not challenging OpenAI. Instead, it has formally accused Google of using its AI models to generate and distribute unauthorized images and videos based on Disney-owned IP. In a letter to Google, Disney claimed the company was effectively “a virtual vending machine” capable of reproducing and distributing its copyrighted characters at scale. The complaint also noted that many of the allegedly infringing outputs were stamped with the Google Gemini logo, creating what Disney called a false impression that the company had approved the use of its characters. Google declined to address the specific allegations but emphasized its relationship with Disney, saying only: “We have a longstanding and mutually beneficial relationship with Disney, and we will continue to engage with them.” AI Post ⚪️ | Our X 🏴

🗣 DeepMind co-founder Shane Legg: AI won't bring sudden disruption; it will quietly take over economic work, domain by domain Within a few years, sectors like software engineering will see massive efficiency gains, with 20 engineers using AI doing the work of 100 AI Post ⚪️ | Our X 🏴

📢 ChatGPT-s "adult mode" is expected in Q1 2026! "Fidji Simo, OpenAI’s CEO of Applications, told reporters during a Thursday briefing about GPT-5.2 that she expects “adult mode” to debut within ChatGPT in the first quarter of 2026, adding that the company wants to get better at age prediction before introducing the new feature." AI Post ⚪️ | Our X 🏴

A sleek humanoid robot showing off smooth dance moves AI Post ⚪️ | Our X 🏴

🚀 Elon Musk: Space data centers could surpass earth-based facilities in three years Elon Musk has suggested that orbiting data centers may soon be more cost-effective than terrestrial ones, igniting debate in the AI infrastructure space. Why space? • Terrestrial data centers face limits on power, cooling water, and available land. • Space-based facilities could reduce strain on local energy grids and eliminate water usage. • Heat can radiate directly into space, allowing for higher hardware density. Economics & feasibility • Cooling accounts for up to 40% of energy costs in conventional centers. • Falling launch costs, especially from reusable SpaceX rockets, make orbital deployment more realistic. • Workloads not needing ultra-low latency could move to orbit as early as 2026. Industry moves & implications • Companies like StatCloud have already deployed orbital infrastructure; Google is reportedly exploring similar initiatives. • Early adopters may gain a competitive edge, potentially widening gaps between countries and companies in AI infrastructure. Musk’s vision raises the possibility that the next wave of computing power could literally be above our heads. AI Post ⚪️ | Our X 🏴

🗣 DeepMind co-founder Shane Legg: AI won't bring sudden disruption; it will quietly take over economic work, domain by domain Within a few years, sectors like software engineering will see massive efficiency gains, with 20 engineers using AI doing the work of 100 AI Post ⚪️ | Our X 🏴

Microsoft AI is programmed to hate White people. Who programmed it? 🤔 AI Post ⚪️ | Our X 🏴