Why Transformer-Based LLMs and Multi-Modal Models Are Not an Existential Threat?

The Real Danger Lies in AI Hype and the Absence of Governance

In recent years, transformer-based large language models (LLMs) and multi-modal AI systems have sparked debates over their potential risks to humanity. The argument often focuses on these models evolving into existential threats that could overpower human intelligence or operate beyond our control. However, this fear is largely misplaced. The true danger lies in the hype surrounding these potential risks, combined with the lack of comprehensive governance structures to regulate their misuse in areas such as fake news, deepfakes, and social media manipulation. By examining both the capabilities and limitations of current AI, we can understand why the models themselves are not inherently a threat—while also acknowledging the urgent need for responsible oversight in their application.

Understanding Transformer-Based LLMs and Multi-Modal Models

Transformer-based LLMs, such as GPT-4 and BERT, and multi-modal models that integrate text, images, and even video (like CLIP and DALL-E), represent significant advancements in AI. They have the ability to generate human-like text, understand context, and even create complex images and designs from mere descriptions. Despite these remarkable capabilities, these models operate strictly within the confines of their architecture—they are sophisticated pattern recognizers, not self-aware entities capable of independent reasoning or decision-making.

These models lack agency, intent, or understanding. They do not "know" what they are doing; they process massive datasets to predict the next word in a sentence or create images based on probabilistic patterns learned from their training data. While their outputs can be strikingly impressive, these AI systems are fundamentally bounded by the data and objectives with which they are trained. In short, they are tools that, like any technology, can be used for good or ill depending on the intentions of their human operators.

Hype and the Fear of AI

The portrayal of transformer-based models as existential threats is itself a dangerous form of hype. This narrative, often amplified by the media and public discourse, not only misrepresents the actual capabilities of AI but also diverts attention from the real issues that require immediate governance. Such alarmism can lead to public panic, uninformed regulatory decisions, and a loss of trust in AI technologies that hold immense potential for societal good—from healthcare innovations to climate change solutions.

The fear-based narrative around AI often fuels a kind of “techno-pessimism” that can stifle innovation. It is important to distinguish between science fiction and science. While AI researchers are pursuing AGI (Artificial General Intelligence), the idea that today's LLMs or multi-modal models are anywhere near achieving human-level consciousness or control is not grounded in the current state of technology. Moreover, presenting these systems as an existential threat often overlooks the tangible risks that exist today—risks that arise not from the AI itself but from how AI is used. The real threat is the fear of AI.

“the only thing we have to fear is fear itself.”

~ Franklin D. Roosevelt

The True Dangers

Misinformation, Deepfakes, and Social Media Manipulation

The immediate concern is not that AI will surpass human intelligence, but that it will be misused in ways that threaten societal stability. Generative AI has made it easier than ever to create fake news articles, deepfake videos, and misleading social media posts. The dangers of these AI-driven tools are already manifesting: false information can now be disseminated faster than ever before, eroding trust in institutions and undermining democracy.

Deepfakes, for instance, have become a powerful weapon for misinformation campaigns. AI-generated videos and images that convincingly mimic real people—whether celebrities, politicians, or public figures—can easily deceive viewers. These tools can be used to tarnish reputations, interfere in elections, or stoke social divisions. Without a robust governance framework to address the misuse of such technologies, the risks of disinformation and social manipulation are immense.

Social media platforms, already struggling with issues like fake news and echo chambers, now face an even greater challenge with the rise of generative AI. Models capable of producing an infinite stream of fake content make it easier for bad actors to flood the information ecosystem with disinformation. These platforms, however, often lack the resources or incentive to effectively combat these issues.

A Call for Responsible AI Regulation

While the technology underlying LLMs and multi-modal models is not an existential threat, the absence of proper governance structures presents a significant danger. Governments, tech companies, and civil society must come together to establish clear, enforceable guidelines on the ethical use of AI.

There must be legal frameworks that hold individuals and organizations accountable for the misuse of AI-generated content. Stricter penalties for the creation and dissemination of deepfakes, for example, are necessary to deter malicious actors. Similarly, robust fact-checking mechanisms and AI-based tools to detect fake news and manipulated media need to be more widely adopted across social media platforms.

Transparency is key. AI systems must be designed with explainability in mind, ensuring that users understand how and why certain outputs are generated. For instance, generative models should have clearly labeled outputs, so users are aware when content is AI-generated. This will help curb the spread of misinformation and maintain public trust.

Education is essential. The public must be informed about the capabilities and limitations of AI to avoid falling victim to the hype or fear surrounding these technologies. Understanding that LLMs and multi-modal models are tools, not autonomous agents, is critical to fostering a more informed discourse on AI.

We need to close the Governance Gap!

Focusing on the Real Threat

Transformer-based LLMs and multi-modal models are impressive technological achievements, but they are not the existential threat they are sometimes made out to be. The real danger lies in how these tools are used—specifically, their potential for misuse in generating fake news, deepfakes, and social media manipulation. Instead of succumbing to alarmist narratives, we must focus on establishing effective governance frameworks that promote responsible use, prevent abuse, and protect the integrity of information. Only through measured, informed approaches to AI governance can we harness the true potential of these technologies for societal good.