AGI test overwhelms AI – Common Sense, 6G, and Crypto vs AI

Introduction

Artificial General Intelligence (AGI) is considered the "holy grail" of AI research – a machine that can think and learn flexibly like a human. Current AI models amaze with their capabilities but reach their limits when it comes to real understanding and reasoning. Recent developments now highlight five closely linked topics on the path to AGI:

1. A new, challenging AGI test that makes even the most modern AI models stumble.
2. Research efforts to integrate common sense into AI systems to achieve true general intelligence.
3. The vision of using next-gen wireless networks (such as 6G and beyond) to enable AI to think more like humans.
4. China's advanced AI agent "Manus," which is celebrated as the first glimpse of real AGI.
5. The cultural and technological debate between cryptocurrencies and AI as a defining narrative of our time.

Below we provide a comprehensive overview of these trends. Semantic SEO strategies are employed by incorporating relevant keywords, LSI keywords, and entity links. Let's dive into the world of AI research and technology – and how it might decide the race for the future.

New AGI Test: AI Models at Their Limit

What is the ARC-AGI-2 Intelligence Test and why do AI models fail it?

A brand new intelligence test called ARC-AGI-2 is currently causing a stir in the AI community. This test was developed by French AI researcher François Chollet and the Arc Prize Foundation to measure true general intelligence. ARC-AGI-2 consists of visual puzzles – colorful puzzle challenges – in which an AI system must perform pattern recognition and independently generate solutions. What's special: The tasks are designed to be completely novel. An AI model cannot simply rely on what it has already learned but must demonstrate adaptive thinking.

The results are sobering (and revealing at the same time): Even the most modern AI systems only achieve random values in this AGI test. State-of-the-art models like OpenAI's "o1-pro" or DeepSeek's "R1" only achieve about 1% of the points. Even powerful language AIs of the latest generation – for example, GPT-4.5 from OpenAI or Google's Gemini 2.0 – barely exceed this value. Human participants, on the other hand, perform dramatically better: In initial trials, groups of people solved around 60% of the tasks on average. This enormous gap (60% vs. ~1%) shows that current AI may excel in narrowly defined areas but quickly reaches its limits with truly novel challenges.

Why do the machines fail? ARC-AGI-2 forces them to think flexibly without training and prior knowledge. Simple pattern recognition or "brute force" computing power are not enough. Instead, it would require transfer learning, creativity, and real understanding – abilities we know from human intelligence but which are largely lacking in today's algorithms. That's precisely why many experts view this test as an important milestone: It exposes the weaknesses of our narrow AI models and defines a new hurdle on the way to strong AI (AGI). No wonder the industry is calling for new benchmark tests to better measure progress in artificial intelligence.

For further reading: In our post "AGI is closer than you think," we discuss what breakthroughs researchers believe are possible in the coming years. This new AGI test clearly shows: Until machines achieve true general knowledge and adaptability, there is still a long way to go.

Visual representation of a neural network in the form of a brain, symbolizing the interplay of human and artificial intelligence (AGI concept).

Image: A Flux-generated AI concept image visualizes how closely electronic circuits (AI) and neural patterns (human brain) are intertwined. Such graphics underscore the goal of AI research: Computers should one day be able to think and learn like we humans do.

AI Needs Common Sense – The Path to General Intelligence

How are researchers bringing common sense into AI systems?

One of the main reasons why today's AI has not yet achieved general intelligence is the lack of common sense. While humans learn to understand the world intuitively as children – e.g., physical relationships or everyday logic – AIs are mostly based only on statistics and data. These often lack understanding of obvious things that were not explicitly included in the training. The result: Impressive language models like ChatGPT can write novels but stumble over simple logical puzzles or banal everyday questions that are trivial for us. Experts therefore describe common sense as the missing puzzle piece on the way to AGI.

Researchers worldwide are tinkering with ways to imbue AI systems with this general knowledge. Some approaches use knowledge graphs (e.g., ConceptNet) and ontological databases to equip AIs with basic facts. Others pursue neuro-symbolic AI, which combines machine learning with logical rules. This is how the machine should learn to think, plan, and abstract as we do. A research team from Virginia Tech University is taking a particularly futuristic approach: They suggest connecting AI directly with our real world. According to their current study, it is crucial for true AGI to equip AI systems with common sense so that they can think beyond their training knowledge, imagine things, and plan.

Common Sense through Experience: AI Embedded in the Network

The researchers' vision: In the future, AI agents could continuously learn from the real environment via advanced wireless networks (keyword beyond 6G). Instead of only training in isolated data silos, they would be embedded in an intelligent environment that conveys everyday experience to them. Similar to how we humans develop our common sense through interaction with the physical world, an AI in an "AI-native" network could observe situations live, draw consequences, and build intuitive knowledge.

Of course, this idea is still in its infancy. But the study outlines an interesting blueprint: A future Intelligent Network would not only transmit data but would itself become a learning organism. Components such as digital twins – virtual images of the real world – could serve as a world model to integrate human thinking into the network. This would give AI systems a built-in common sense, as they are constantly confronted with realistic scenarios.

In practical terms, this means: An AI that is connected to real traffic and sensor data via an autonomous vehicle network, for example, could implicitly learn that rain makes roads slippery without being explicitly told. This gain in everyday understanding would come closer to what we call common sense.

We are still far from this, and classical approaches such as DARPA's Machine Common Sense Project (2019) have shown how difficult it is to code common sense. But the research direction is clear: If you want to create strong AI, you must enable machines to think outside the box. A mixture of knowledge transfer (e.g., facts, physics rules) and experience (interactions in virtual and real environments) seems to be the best way forward.

Interestingly, the mentioned vision already points to the next big trend – 6G networks as a catalyst for AI. In the following section, we take a closer look at how upcoming wireless technologies could boost machine intelligence. (More background on Common Sense AI in our article "Why AI Still Lacks Common Sense.")

Next-Gen Networks: Does AI Think More Human-Like with 6G?

Can 6G and beyond enhance AI's thinking ability?

While 5G mobile is just reaching the mainstream, researchers are already planning the next generation: 6G and post-6G networks. These future networks are meant to be far more than just faster data lines. Visionaries see them as the foundation for completely new AI capabilities. The keyword is AI-native Wireless: intelligent networks that not only transport data but understand and learn themselves.

The concept: AI algorithms are integrated directly into the network infrastructure. Such a "thinking network" could analyze large data streams in real-time and act as a kind of distributed brain. Virginia Tech researchers argue that only this fusion of communication and AI could enable the leap to general intelligence. Instead of operating in isolation in a data center, an AI would be present throughout the network – similar to neural connections in the brain. It could permanently learn from the users, devices, and environments to which the network is connected.

A practical example: Imagine a 6G smart city in which billions of sensors, vehicles, and devices are connected. A built-in AI can learn from all these sources – traffic patterns, weather, human behavior – and provide this knowledge to all connected devices. Thus, the autonomous car would already have "experience" from the accident that another car had two streets away and could adjust its driving behavior preventively. The AI within the network would serve as a collective memory and teacher for all participants.

According to Professor Walid Saad of Virginia Tech, such an intelligent network is a missing link. It would move beyond mere data transmission and instead actively learn from the data, comparable to an organism. Digital images (Digital Twins) of the environment act as a training ground in which the AI can safely play through scenarios. The advantage of these next-gen networks is obvious: AI systems could master unforeseen situations because the network provides them with cumulative experience. They would be less dependent on rigid training and could generalize better – a crucial step towards AGI.

But how close is this vision? Experts dampen expectations that are too high: Realistically, it will probably take another 10-15 years until such an AGI-capable 6G network becomes reality. The technical challenges are enormous – from the required computing power (keyword Edge Computing and best server infrastructure) to security and data sovereignty in a learning network. Nevertheless, building blocks can already be implemented today. Companies are working on network slicing and intelligent base stations that take on the first AI functions.

In 6G research, it is becoming apparent that networks of the future will be far more adaptive. They may not "think" like a human, but they could help AI simulate human thinking by providing context and experience. If machines are to truly possess general knowledge one day, they may need such a rich, networked environment to develop their understanding.

Curious about 6G? Our article "What is 6G and what does it bring?" explains the technical basics of this mobile generation. One thing is clear: Next-generation wireless could revolutionize far more than just our download speeds – it could be the mental fertilizer for the next evolutionary stage of AI.

China's Manus Agent: A First Glimpse of Real AGI?

What makes Manus AI so special and groundbreaking?

While researchers around the world are still theoretically discussing AGI, a concrete prototype in China is making headlines: the Manus AI agent. Developed by Beijing startup Butterfly Effect, Manus is presented as the world's first general AI agent. What's special: Manus works autonomously – it makes decisions, plans steps, and carries out tasks without having to be given each command individually. The developers speak of a first glimpse of real AGI that Manus could offer.

Unlike conventional chatbots like ChatGPT (which respond to an input and generate an output), Manus acts more like a virtual employee. You give it a goal or problem, and the agent independently considers how to solve it. In doing so, it draws on several AI models and tools that work together – a so-called multi-agent architecture. Where ChatGPT uses only a single large language model, Manus orchestrates several specialized LLMs and software modules in the background. These can, for example, research on the internet, write code, or analyze data without the user having to initiate each intermediate step.

The capabilities that early users have demonstrated with Manus sound impressive. With just a simple text instruction, Manus could, among other things:

• Develop video games – generate playable prototypes from a vague game idea.
• Design and launch websites – execute automated design to go-live.
• Create travel plans – suggest complex holiday routes including bookings.
• Analyze stock markets – sift through large amounts of stock market data and recognize trends.
• Pre-sort applications – evaluate resumes and filter out suitable candidates.

(These examples are based on reports from early testers and show the broad task spectrum of Manus.)

The key point: Manus AI can break down complex tasks into sub-problems, navigate the web independently, use external tools, and dynamically adjust its plan. All this happens while the user may have already closed the computer – Manus continues to work in the background until the goal is achieved. This level of autonomy has not previously existed in publicly known AI systems.

Of course, Manus is not yet perfect. In the limited testing phase, some users reported crashes and situations in which the agent got stuck in endless loops. According to Peak Ji, the lead scientist of the project, such teething problems are to be expected with a completely new system. A journalist who was allowed to try Manus described the experience as "collaboration with a highly intelligent, efficient intern" – who, however, sometimes still shows misunderstandings and takes shortcuts that don't always lead to the goal. In other words: promising, but not flawless.

So is Manus already AGI? The developers themselves dampen expectations that are too high. In a presentation video, they emphasize that Manus is more of a "Glimpse of AGI" – a foretaste of what future general AI could achieve. In fact, some experts are skeptical about whether Manus already meets the criteria for strong AI. Nevertheless, the system demonstrates capabilities today that were recently considered science fiction. It shows how a combination of advanced LLMs and clever task planning can bring machines significantly closer to human problem-solving abilities.

The geopolitical aspect is also exciting. Manus is already the second AI system from China within a short time to cause a stir internationally – after DeepSeek, a ChatGPT-like model that showed similarly good performance as western competitors at lower costs in early 2025. China is thus positioning itself at the forefront of the AI race. While OpenAI, Google & Co. in the West are still working on multimodal models and GPT-5, Manus could show an alternative route to AGI: through autonomous agents that combine several specialized AI components.

In any case, Manus AI offers a fascinating outlook. Should upcoming versions become more stable and generally available, we are facing a new era of AI tools that act largely independently. This opens up enormous opportunities for developers and companies – from real automation of complex processes to creative collaborations between humans and machines on an equal footing. Of course, this also raises new ethical questions (e.g., responsibility and control of such autonomous agents), but the AGI future seems a bit more tangible with Manus.

(Internal reading recommendation: More on the technical details of multi-agent AI systems in our article "AI Agents Explained: When AIs Work Together.")

Crypto vs. AI – Which Tech Narrative Shapes Our Era?

Contest of the Future: Cryptocurrency or Artificial Intelligence?

Our present is shaped by two major technological currents: blockchain-based cryptocurrencies and artificial intelligence. Both promise to change the world – but in very different ways. In recent years, an exciting cultural and technological debate has emerged about which of these developments will be the defining narrative of our time. Is it the vision of a decentralized, free financial system through crypto? Or the rise of ever smarter machines that permeates our lives with AI?

Looking back, both areas had their hype phases. The years 2017-2021 were considered the heyday of the crypto boom: Bitcoin, Ethereum & Co. were on everyone's lips, blockchain was touted as a revolution. But recently, AI as a narrative seems to have taken the lead – at the latest since the resounding success of ChatGPT (end of 2022), capital and attention have increasingly been directed towards artificial intelligence. Talents from the tech industry are switching from the crypto sector to AI startups. Investors ask: Is AI the new crypto? 🔄

In fact, some thought leaders argue that AGI is the much bigger promise. One investor pointedly stated that AGI is the biggest narrative in history, attracting all brilliant minds. In his view, the crypto world is threatened by an exodus of talent if it no longer offers a higher goal and degenerates into a mere playground for speculators. Many developers want not only to get rich but to create technology that makes history in a sustainable way. If the environment no longer fulfills their ideal conceptions, they switch to where they see more meaning – currently often in AI projects. Simply put: Artificial intelligence – especially the prospect of AGI – electrifies the imagination, awakens utopian visions (from curing all diseases to interstellar travel), and thus attracts enormous resources.

On the other side are the advocates of the blockchain revolution. They remind us that crypto has not failed but is undergoing a longer-term transformation. Blockchain technology aims at nothing less than a redesign of economic and power structures – a struggle against established institutions that is naturally chaotic and lengthy. Decentralization remains a radical alternative to centralized control – whether by banks or by AI monopolies. This vision still has traction: It promises individual sovereignty over data and money. Some experts therefore emphasize that crypto vs. AI cannot be categorized as "better or worse" – both narratives serve different areas. Blockchain addresses trust, security, and production relations (who controls resources?), while AI primarily increases productivity and makes our everyday lives smarter. In this view, crypto is a revolution of infrastructure, AI an evolution of capabilities.

Interestingly, there are also overlaps: AI is becoming a trend in the blockchain industry itself (keyword AI-Crypto – AI models for predicting markets or AI-driven smart contracts). And conversely, the AI community is discussing whether blockchain can help ensure trustworthy AI (e.g., traceability of training data). So it's quite possible that both fields will complement each other in the future, rather than compete.

For the moment, however, "Crypto vs. AI" as a culture war is having a lively debate in social media and tech circles. Some tweet about the end of cryptocurrencies and proclaim the AI age, others hold against it and expect the next crypto renaissance – possibly enriched with AI technologies. This field of tension is stimulating because it forces us to ask: What future do we want? An algorithmized world where data and AI rule? Or a dematerialized economy in which code and cryptography replace trust?

Probably both developments will shape our future – but in different ways. While AGI could challenge our idea of work, creativity, and even consciousness in the coming decades, blockchain has already begun to question traditional financial and governance models. Perhaps we will even experience a convergence: decentralized AI networks that redefine both intelligence and infrastructure.

One thing is certain: Both crypto and AI embody the spirit of innovation of our time. They are narratives about how technology can change society – one more focused on freedom and autonomy, the other on progress and efficiency. Instead of crowning a "winner," we should critically accompany both. Because it is often in the confrontation that the best ideas emerge.

Conclusion and Outlook

Whether AGI test, common sense, 6G networks, Manus AI, or Crypto vs. AI – all these topics show how dynamically the tech world presents itself in 2025. The development of artificial intelligence towards general intelligence is not proceeding in a straight line but on many levels simultaneously: Researchers are working on new benchmarks and capabilities, engineers are designing intelligent infrastructures, and pioneers like Manus are already venturing into the unknown. At the same time, there is heated debate about which vision will dominate our future.