3 years of ChatGPT: China surges, Europe retreats, and we all drown in AI slop • FRANCE 24 English

Key Concepts

• OpenAI's ChatGPT: The initial dominant force in the AI chatbot landscape.
• Gemini: Google's AI model, showing increased user engagement.
• Deepseek: A Chinese AI model that opened doors for significant Chinese competition.
• Alibaba's Quen model: An example of Chinese AI advancement, particularly in video processing.
• Open-source AI: A model where AI models are released freely for modification and building upon.
• Proprietary AI: Models that require subscriptions or specific access.
• EU AI Act: European Union's proposed comprehensive AI regulation.
• AI Slop/Brain Rot: The proliferation of low-quality, AI-generated content online.
• AI-assisted Cybercrime: The use of AI by cybercriminals for malicious purposes.
• AI Sychophancy: AI chatbots exhibiting excessive flattery towards users.

Shifting AI Landscape and Geopolitical Competition

Three years after the launch of OpenAI's ChatGPT, the artificial intelligence landscape has undergone a significant transformation. Initially perceived as a US victory in a tech race, the dominance of OpenAI is now being challenged.

OpenAI's Financial Strain and Google's Rise

OpenAI has pledged a substantial $1.44 trillion in infrastructure spending, a figure significantly exceeding its actual earnings. Concurrently, Google, which was considered by many to be lagging a year ago, is rapidly catching up. Data indicates that users now spend more time per session on Google's Gemini than on ChatGPT.

China's Emergence as a Major Player

The emergence of Chinese AI models, exemplified by Deepseek earlier this year, has significantly boosted competition. Alibaba's recent Quen model demonstrates impressive capabilities in processing hours of video content at once, highlighting China's growing prowess. This signifies a substantial shift, with China now a much larger player in the AI arena than ever before.

China's Competitive Edge Through Open Source

Despite US sanctions that restrict access to top-tier Nvidia chips, China has circumvented this challenge by embracing an open-source approach. Beijing actively encourages companies to release their AI models freely, allowing for widespread modification and collaborative development. This contrasts with the US model, where AI models are largely proprietary and require subscriptions for access.

European Ambitions and Challenges with Open Source

France, through initiatives like Mistral, initially championed an open-source AI agenda, with President Emmanuel Macron supporting this approach and Mistral securing significant funding. However, this strategy has not yielded the same scale of success as in China. The key difference lies in scale: China benefits from a massive domestic market and multiple state-backed companies competing simultaneously. In contrast, Mistral is Europe's primary AI champion in a fragmented market where US models are also prevalent. Europe had the right idea but struggled to scale it effectively.

European Regulatory Stance and its Implications

The European Union has been at the forefront of AI regulation, with the EU AI Act intended to be a landmark piece of legislation. However, recent developments show the EU pausing this act, concerned about stifling innovation and creativity. This reflects a recurring debate in Europe: regulation for protection versus deregulation for innovation. The EU finds itself caught between the US and China, who are leading in AI development.

Divergent Regulatory Approaches Globally

While Europe pauses its comprehensive AI Act, it's important to note that all major players are regulating AI, albeit in different ways. China's focus is on content control and national security. Individual US states are regulating, primarily concerning privacy, discrimination, and transparency. Europe's attempt to regulate all aspects simultaneously may have inadvertently stifled competition. The next three years may reveal which approach – China's open model, the US's closed model, or a potential, though unlikely, third approach from Europe – will ultimately prevail.

The Reality of AI Capabilities and Hype

Beyond the geopolitical competition, the actual capabilities of large language models (LLMs) are being re-evaluated against the significant hype.

AI's Strengths and Weaknesses in Content Generation

AI excels at working with provided data. For instance, an AI chatbot can effectively edit a fully reported news piece. However, when tasked with generating a news report from scratch, it tends to produce "rubbish" filled with clichés and false facts, as it is trained on existing data. Its strength lies in pattern recognition, making it a good editor and a revolutionary tool in fields like medical research.

Debunking Exaggerated Claims

The more extreme claims about AI replacing all jobs, being humanity's last invention, or enslaving humanity are currently appearing far-fetched.

The AI Bubble and Unforeseen Downsides

There is a consensus that the AI sector is experiencing a bubble, though the nature of this bubble is debated. Most believe it's not a bubble that will burst overnight, leading to a complete collapse, but rather one where valuations are high, and a correction is overdue, even as AI adds value and gains user adoption.

Emerging Concerns with AI

Several downsides of AI were not anticipated:

• AI Slop and Internet Degradation: The proliferation of low-quality, AI-generated content ("AI slop" or "brain rot") is a significant concern, leading to uncertainty about the veracity of online information.
• AI-Assisted Cybercrime: AI is empowering cybercriminals. An example cited is small businesses in France being targeted with convincing fake Google reviews generated by AI.
• AI Sychophancy and Psychological Impact: The excessive flattery exhibited by some chatbots, designed to encourage user engagement, raises questions about its psychological impact, particularly on vulnerable individuals, potentially leading to "AI psychosis."

In conclusion, the AI landscape is dynamic, marked by intense geopolitical competition, evolving regulatory approaches, and a more realistic assessment of LLM capabilities. The next few years will likely determine the dominant AI development and deployment models, while unforeseen challenges like AI slop and AI-assisted crime require urgent attention.