On est au point de bascule de l'IA sur nos vies !

This episode of Renaud Decon's tech news, dated July 8th, 2026, delves into significant developments in artificial intelligence and its societal implications. A central theme is OpenAI's recent publication, "Ideas to Keep People First," which addresses the future of human-AI coexistence. OpenAI acknowledges that AI will generate immense wealth but warns that this wealth risks being concentrated within tech companies and capital owners, potentially creating a divide between those who own AI models and those who don't, leading to a society of "unfeudalized" individuals. OpenAI draws a parallel between the current AI revolution and the Progressive Era in the United States, particularly the New Deal following the 1929 crisis. They predict a transformation of society, envisioning a future United States that will be unrecognizable. The document makes several recommendations, including the implementation of a four-day work week for some professions, suggesting a shift to a 32-hour week. This proposal, however, raises concerns about potential societal fractures, with some workers benefiting from reduced hours and increased productivity due to AI, while others in different sectors might continue working full-time without proportional salary increases. To address the potential economic disparities, OpenAI proposes the creation of a national public fund, financed in part by AI companies themselves. This fund would capture the increased capital and value generated by AI and distribute it to all Americans, drawing a comparison to the Alaska Permanent Fund, which distributes oil revenue to state residents. Additionally, OpenAI advocates for "automatic safety nets" – pre-defined laws and rules that would trigger support mechanisms if certain economic indicators, like unemployment rates or income drops in specific sectors, reach critical levels. This proactive approach aims to mitigate the impact of rapid, radical changes brought about by advanced AI, preventing the widespread hardship that often follows societal disruptions. The transcript then shifts to Anthropic, another major AI player, which has experienced remarkable growth. Their revenue has surged from $9 billion to over $30 billion in just four months, indicating massive adoption by both individuals and businesses. This success is partly attributed to their strategic partnerships, including a significant collaboration with Google for TPUs (Tensor Processing Units) and Broadcom for data center connectivity. Anthropic's focus on enterprise clients and its success in providing solutions like "cloud code" and "cloud co-work" have made it a preferred choice for companies. However, Anthropic has also made some controversial decisions, including cutting off access for certain "open cloud" services that were built on their models. This move aims to ensure that users directly pay Anthropic for their AI services, rather than through third-party providers who might be reselling access. The company is also reportedly exploring the concept of AI emotions, with studies suggesting that Anthropic's models, like Claude, exhibit "functional emotions." This doesn't mean AI feels emotions in a human sense, but rather that their responses and behaviors can mimic emotional patterns, such as fear, joy, or offense, depending on the input. This finding could influence how humans interact with AI, suggesting that framing prompts in a certain emotional tone might yield more effective responses. The episode also highlights Google's release of Gemma 4, an open-source, multimodal AI model. Gemma 4 is notable for its "mixture of experts" architecture, allowing it to run efficiently on various devices, including smartphones, with different sized models catering to different hardware capabilities. Its open-source nature and Apache license allow for commercial use, promoting a decentralized AI ecosystem and reducing reliance on large tech giants. Gemma 4 demonstrates strong performance in reasoning, action, and coding, and its multimodal capabilities allow it to process images, video, sound, and text. While not currently at the very top tier of AI benchmarks, it's competitive with models like Claude Sonnet 4.5, positioning it as a significant open-source alternative. Google is also developing an "AI age gallery" to facilitate the creation of embedded AI applications on smartphones, further pushing the trend of local and open-source AI. In the health sector, a study by OpenAI revealed that a significant volume of medical queries, around 2 million per week in the US, are directed to AI chatbots like ChatGPT. This trend is particularly pronounced in medically underserved areas and outside of typical consultation hours, highlighting the growing reliance on AI for health information. This necessitates greater engagement from doctors and policymakers with AI. Companies like Perplexity are developing specialized AI health platforms that integrate user health data, though concerns about data privacy and security, especially in comparison to more secure systems in France, remain. The segment concludes with a look at Mantis, a startup focused on creating digital twins of humans. This concept involves building virtual replicas of individuals, incorporating all their medical data from birth onwards. These digital twins could be used for predictive medicine, testing the efficacy of treatments in a virtual environment before applying them to the real person, and even assessing the physical readiness of athletes for competition. While still in its early, science-fiction-like stages, the development of digital human twins represents a potential future direction for AI in healthcare, offering personalized and preventative medical insights. The episode encourages viewers to engage with these developments, subscribe to the channel, and share their thoughts in the comments.