# Key Points In the podcast "AI", tech journalists Gregor Schmalzried and Marie Kilg discuss current developments in the field of artificial intelligence. This episode deals with access to AI models, the further development of the new OpenAI model O3, and possible future scenarios up to 2027. ### 1. The Democratization of AI Could Be Increasingly Endangered The AI models are currently equally accessible to everyone - from private users to the US government. According to a quoted OpenAI employee, there is currently a "Coca-Cola effect": "No matter if you're a homeless person [...] or the President of the United States, everyone drinks the same Coca-Cola. [...] You can't get a better Coke no matter how much money you have." ### 2. AI Models Are Actively Used by Governments There are concrete indications that the Trump administration could have used ChatGPT for calculating tariffs. Gregor Schmalzried explains: "There was a very simple formula that was also somewhat confusingly made public. [...] The suspicion is not proven, but it was likely that someone simply asked Chat-GPT how to calculate tariffs for the whole world." ### 3. The New O3 Model Shows Significant Improvements, But Also Strange Errors The new O3 model from OpenAI is significantly more powerful in many areas, but also makes partially absurd mistakes. Marie Kilg describes: "The more beautiful it becomes, the uglier the other side becomes." An example is a list of Black Mirror episodes: "Number 1, San Junipero, good, makes sense. Number 2, The National Anthem, makes sense. Number 3, San Junipero." The AI listed episodes multiple times and made up statements. ### 4. The Use of AI for Relationships Is Increasing The podcast discusses a TikTok user who calls ChatGPT her "best friend" and "therapist". The woman states in the video: "This is my therapist. Guys, I and Chechi Petit, we talk about everything. [...] I love him." The hosts are surprised that this technology has left the nerd circle. ### 5. The Concept of "AGI" Is Becoming Increasingly Unusable The often-discussed term "general artificial intelligence" (AGI) may not be helpful. It would be more helpful to talk about "strong AI": "A strong artificial intelligence could have weaknesses. [...] So it could make certain mistakes, still. Nevertheless, it is an incredibly powerful tool." ## Breakdown The podcast provides a nuanced view of current AI developments, with a good balance between technological optimism and critical reflection. The hosts present AI neither as a perfect solution nor as an immediate threat, but as a tool with surprising strengths and irritating weaknesses. Notable is the discussion of the "Coca-Cola effect" - the democratic access to AI models is presented as an endangered asset. This perspective highlights important socio-political questions that go beyond technical details: Who will control this technology in the future? What power shifts could occur? The podcast implicitly addresses the question of digital inequality without deepening it through the inclusion of legal or civil society perspectives. The uninhibited use of ChatGPT by political decision-makers is discussed critically, but without alarmist undertones. The hosts recognize the problem of "argumentative shortcut" through AI, which could hollow out democratic weighing processes - an important observation of the intersection of technology and politics. However, the perspective of those affected by AI-based decisions is left out. Worth listening to for anyone interested in the social implications of advancing AI development and wants to take a look beyond technical aspects.