Meta is opening up its Muse AI platform to hardware enthusiasts, letting people connect the AI agent to ESP32 boards, Raspberry Pi devices, and other DIY setups. The company is also giving away 5,000 Muse Home Link dongles to US subscribers, which allow users to hook their homemade projects into the internet and have the AI control things like lights, TVs, and printers. Meta is providing SDKs and tools for building custom gadgets, though it cautions users to proceed at their own risk.
Google is making a strong push back into the AI frontier with Gemini 4 Argon, which the company calls its most powerful model to date. CEO Sundar Pichai and DeepMind chair Demis Hassabis helped unveil the model, which currently has limited access for selected cybersecurity partners through the Fairwind Program and for Google staff. Early benchmarks point to strengths in long-context tasks, coding, and enterprise-agent applications, signaling Google's renewed competitiveness in the space.
In the policy world, the White House's AI self-regulation accord is drawing scrutiny for its lack of enforcement teeth. The agreement calls for independent external auditors but does not identify who will select or fund them, nor does it require publishing audit findings. There are no explicit sanctions for non-compliance. Anthropic CEO Dario Amodei and other signatories agree that AI risk is real, but disagreements persist about how serious those risks are.
Meanwhile, a new study using analogy as a framework is exploring whether AI systems can be conscious. Researchers built a causal model to compare biological systems with AI, finding that behavioral similarity alone offers only limited evidence for artificial consciousness. The work clarifies what kind of evidence would be needed to make stronger claims about machine awareness.
Fossil fuel companies are leaning into AI to boost oil and gas discovery. Equinor used AI and new seismic technology to find oil off the Norwegian continental shelf, saving $100 million in 2022. Chevron says AI helped it discover 1.5 billion barrels of oil in the Permian Basin. A McKinsey report estimates that AI-driven efficiency gains have already saved the upstream oil and gas sector $100 billion. UN Secretary-General AntĂłnio Guterres warned that technology without accountability poses real dangers.
On the regulatory front, the EU's Artificial Intelligence Act requires all member states to set up AI Regulatory Sandboxes by August 2027. Researchers have introduced the AI Assessment Sandbox Configurator, an open-source framework with a catalogue of tests, a shared data model, and role-specific dashboards. An early pilot tested the tool in a live regulatory setting, aiming to make testing results easier to compare across countries.
Two new studies are questioning whether today's large language models truly reason like humans. Ten years after AlphaGo defeated Go champion Lee Sedol in 2016, researchers note that modern AI still lacks the search-based reasoning behind that achievement. Current models rely on repeated token prediction, which amounts to fast pattern matching rather than genuine thought. The studies point to key shortcomings, including the absence of a persistent, inspectable record of what the model is thinking.
The US and China are taking very different paths on AI policy. The US is emphasizing regulation and oversight, including a national AI strategy and a new AI safety board. China favors a hands-off approach that encourages innovation and entrepreneurship. These contrasting strategies carry significant implications for how AI develops and is deployed in each country and beyond.
AI in medicine is also under the microscope. Robert F. Kennedy Jr., Secretary of Health and Human Services, claimed AI can give patients a second opinion, drawing sharp pushback from physicians who question the accuracy and safety of medical AI tools. The debate reflects broader concerns about how much trust to place in AI-driven healthcare decisions.
Key Takeaways
- Meta is opening its Muse AI to DIY hardware, offering SDKs for ESP32 and Raspberry Pi and giving away 5,000 Home Link dongles to US subscribers.
- Google released Gemini 4 Argon, its most powerful AI model, with limited access through the Fairwind Program for cybersecurity partners and Google staff.
- The White House AI self-regulation accord lacks clear sanctions, auditor selection details, and publishing requirements, drawing criticism from signatories including Anthropic CEO Dario Amodei.
- A new analogy-based study finds that behavioral similarity alone provides limited evidence for AI consciousness, calling for stronger evidentiary standards.
- Equinor saved $100 million in 2022 using AI and seismic technology, while Chevron discovered 1.5 billion barrels of oil in the Permian Basin with AI assistance.
- McKinsey estimates AI-driven efficiency gains have saved the upstream oil and gas sector $100 billion overall.
- The EU's AI Assessment Sandbox Configurator is an open-source framework helping member states comply with the AI Act by August 2027.
- Researchers argue that large language models do not truly reason like humans, lacking persistent, inspectable thinking records compared to systems like AlphaGo.
- The US favors AI regulation and oversight while China takes a hands-off approach prioritizing innovation and entrepreneurship.
- RFK Jr.'s claim that AI can provide patients a second opinion has drawn physician backlash over concerns about accuracy and safety in medical AI.
Meta opens source tools for building Muse AI gadgets
Meta now lets people connect Muse AI to their own hardware like displays or a Raspberry Pi. Users can program ESP32 boards or set up Raspberry Pi with Meta's SDKs to build their own devices. Meta is also giving away 5,000 Muse Home Link gadgets that let the AI agent control lights, TVs, and printers. The company warns users to proceed at their own risk.
Muse Gadgets opens AI agent for DIY hardware fans
Muse Gadgets is opening its AI agent to the public for use with ESP32 and Linux hardware. The company is offering 5,000 free Home Link dongles to US subscribers. The Home Link dongle lets users connect DIY hardware projects to the internet. This move is a big step for the DIY hardware community to bring AI technology into their projects.
Analogy helps explore artificial consciousness
A new study uses analogy to explore whether AI can be conscious. The researchers built a causal framework to weigh similarities between biological systems and AI. The framework shows that behavioral similarity gives only limited evidence for consciousness in AI. It also clarifies what kind of evidence would strengthen claims of artificial consciousness.
Google returns to AI frontier with Gemini 4 Argon
Google CEO Sundar Pichai and DeepMind chair Demis Hassabis helped release Gemini 4 Argon. Google calls it its most powerful AI model yet. Access is limited to selected cybersecurity partners through the Fairwind Program and Google staff. Benchmarks show strength in long-context, coding, and enterprise-agent tasks. This puts Google back in the AI race.
RFK Jr.'s AI claims draw physician backlash
Robert F. Kennedy Jr., Secretary of Health and Human Services, said AI can give patients a second opinion. His claims sparked pushback from physicians. The debate over AI's role in medicine is growing more intense. Doctors are questioning the accuracy and safety of medical AI tools.
White House AI accord lacks clear sanctions and accountability
The White House announced an AI self-regulation accord that aims to prevent harm before it occurs. The agreement calls for independent external auditors but does not name who will select or pay them. It also does not specify whether audit findings will be published. The accord has no explicit sanctions for failed compliance. Signatories including Anthropic CEO Dario Amodei agree that AI risk is real, but disagreements remain about the nature of that risk.
US and China take different paths on AI policy and anxiety
The United States and China have very different approaches to artificial intelligence. The US emphasizes regulation and oversight, including a national AI strategy and a new AI safety board. China favors a hands-off approach that encourages innovation and entrepreneurship. These differing policies have implications for how AI technologies develop and are used. The article explores the potential consequences for both nations and the global community.
New framework helps test AI systems in regulatory sandboxes
The EU's Artificial Intelligence Act requires all Member States to establish AI Regulatory Sandboxes by August 2027. Researchers introduced the AI Assessment Sandbox Configurator, an open-source framework designed to support technical testing within these sandboxes. It includes a catalogue of tests, a shared data model, and role-specific dashboards. An early-stage pilot tested the framework in a live regulatory engagement. The tool aims to make testing outputs easier to compare and reuse across Member States.
AI tools help fossil fuel companies find more oil and gas
Fossil fuel companies are using AI to improve drilling efficiency and discover previously unreachable oil and gas. Equinor used AI and new seismic technology to find oil off the Norwegian continental shelf, saving $100 million in 2022. Chevron discovered 1.5 billion barrels of oil in the Permian Basin using AI. A McKinsey report estimated AI-driven efficiency gains have already saved the upstream oil and gas sector $100 billion. UN Secretary-General AntĂłnio Guterres warned that technology without accountability is the real danger.
Large language models do not truly reason like humans
Ten years after AlphaGo defeated Go champion Lee Sedol in 2016, today's AI still lacks the reasoning machinery that made that victory possible. AlphaGo used a search system that explored future consequences of moves, a form of genuine reasoning. Today's large language models pick the next token repeatedly, which is fast pattern matching but not true reasoning. The article identifies three key shortcomings, including the lack of a persistent and inspectable record of what the model is thinking. The author argues that future AI systems need genuine reasoning to be trustworthy in fields like science and medicine.
Sources
- Meta open sources code to let you make Muse AI gadgets
- Muse Gadgets: Meta Opens Its AI Agent to DIY Hardware
- What Can Analogy Tell Us About Artificial Consciousness?
- Google’s journey back to the AI frontier
- RFK Jr.'s AI claims spark physician backlash as debate over medical AI intensifies
- What could make AI self-regulation work?
- How Policy Shapes AI Anxiety and Extremism in the United States and China
- The AI Assessment Sandbox Configurator: A Framework to Support Technical Assessment in AI Regulatory Sandboxes
- AI was supposed to cut emissions. But fossil fuel companies are using it to find more oil and gas.
- Don’t be fooled—LLMs don’t reason
Comments
Please log in to post a comment.