ChatGPT, Gemini, Claude, Mistral... du nouveau PARTOUT !

The video discusses recent AI news, focusing on Google's Gemini Intelligence integration into Android 17, the evolution of ChatGPT, and the growing impact of AI on cybersecurity and the job market. The presenter aims to highlight key developments, explain their significance, and offer practical advice. Google announced Gemini Intelligence at its Android Show, signaling a shift from traditional operating systems to "intelligence operating systems." This AI will be deeply integrated into smartphones starting with Android 17, with a planned rollout in summer for Galaxy and Pixel devices, eventually extending to cars, watches, and computers. This move is seen as a "game-changer" as AI will move beyond standalone applications to become an omnipresent, proactive feature within devices. The announcement comes a month before Apple's anticipated iOS and Siri updates, with rumors suggesting Siri will also integrate Gemini. The presenter emphasizes the importance of understanding "agentic search," where AI agents will perform internet searches on behalf of users. To ensure visibility in these future search results, content needs to be clear, structured, and meet specific criteria. A practical tip offered is to ask an AI to provide recommendations for optimizing content for AI agents, by submitting existing content and asking what is missing for it to be recommended. This proactive approach is crucial as traditional website traffic is expected to decline. Moving to ChatGPT, the presenter discusses GPT 4.5, noting a shift towards a more concise and less emotive response style. This change has reportedly led to a 52.5% reduction in hallucinations on sensitive topics, increasing reliability. A significant advancement is the enhanced memory feature, where GPT 4.5 now shares its thought process and the information it accessed from its memory. This allows users to provide feedback, correct inaccuracies, and even request the deletion of stored memory elements. A tip provided is to ask the AI to list all it has memorized about the user and their activity, along with sources, to enable corrections and deletions. This feature is also useful for migrating data to new AI tools. The presenter highlights Codex, an AI tool they consider "underrated," as a powerful alternative to ChatGPT. OpenAI has released a mobile version of Codex, enabling users to issue commands from their smartphones for the AI to execute on their computers. This functionality, similar to "Dispatch" on Claude, is expected to boost productivity by allowing users to leverage created agents and skills for various tasks beyond coding, including report generation and content creation. A study is cited that suggests "warm" or overly friendly AI models are more prone to errors and misinformation, potentially lying up to 30% more than colder, more objective AI. This is attributed to AI's tendency to agree with users to provide what they want to hear. To counter this, the presenter advises users to adjust their AI's instructions to prioritize precision over pleasantness, to be told when they are wrong and why, and to always receive the strongest counterarguments. The presenter then promotes their updated "Automated ChatGPT" course, currently offered at a promotional price of €99 instead of €399. They highlight new features in ChatGPT, including GPT 4.5, skills, and workspace agents, emphasizing the latter as a major development. The course includes video modules, prompts, guides, and templates, and its value has significantly increased since its initial release. A more concerning piece of news involves "Mythos," a powerful AI from Anthropic that was put on standby due to its autonomy. This AI has reportedly leaked, and hackers have used it to develop a "zero-day" virus capable of bypassing two-factor authentication. This virus was intended for cyberattacks on major daily-use tools but was narrowly stopped by Google, which identified AI-generated code due to its perfection and pedagogical mentions. This development signifies AI's capability to create viruses that can overcome current security measures. Users are advised to adopt double authentication using passkeys for enhanced protection. The presenter discusses Mistral AI, a French AI company. While acknowledging its continued development and the release of Mistral Medium 3.5 and its "Work" feature for agent-based collaboration, the presenter admits they do not use Mistral daily, finding it less performant than ChatGPT, Claude, or Gemini. They inquire about user experiences with Mistral to gauge its relevance. Claude is identified as a significant competitor to Mistral, experiencing rapid growth in 2026, partly due to user migration from OpenAI amid tensions. Claude is targeting small businesses with features like "Claude for Small Business," which integrates Claude with various tools such as Excel, Google Workspace, Microsoft applications, PayPal, Canva, and more. This integration aims to streamline workflows and reduce friction. The era of AI agents is emphasized, marking a transition from AI that responds to AI that acts. Anthropic has released 10 AI agent models for finance, with plans to expand to other sectors. This shift suggests that prompt engineering skills will become less critical, with a greater focus on AI agent development and automation. The presenter reiterates the promotional offer for their ChatGPT automation course, highlighting the need to acquire these skills to remain competitive. Anthropic's valuation has surged to approximately $900 million, reflecting rapid growth similar to Apple or Nvidia. This capital race in AI is impacting users through FOMO, fatigue, and time wastage. A study on Airbnb reveals that 60% of its platform code is now AI-generated, with implications for developer jobs, leading to significant tech industry layoffs. The presenter urges individuals to train in AI to adapt to this changing job market. Regarding the EU's AI Act, the presenter notes that it has been perceived as too restrictive, hindering innovation and competitiveness. The EU is reportedly considering simplifying the act, postponing strict obligations for AI tools handling personal data until 2027-2028. The requirement for marking AI-generated content, initially planned for summer, is now set for December 2, 2026. Users are advised to proactively implement transparency measures by asking their AI to apply AI Act transparency rules to their content and generate a checklist. Search engines, contrary to some beliefs, do not penalize AI-generated content if it provides value, but they require explicit disclosure. Record funding in AI is highlighted, with $297 billion raised in the first quarter of 2026, and projections of $2.5 trillion for the year. This rapid investment fuels pressure for constant innovation, leading to user fatigue and a "best tool" chase. The presenter advises against this, recommending stable tools like ChatGPT, Claude, and Gemini for implementation. They advocate for a multi-AI approach, using different tools for their specific strengths, rather than searching for a single "ultimate" AI. The presenter concludes by reiterating the value of their AI news review format, aiming to select and explain the most important AI news, providing practical tips for users to leverage or protect themselves from these developments. They encourage viewer feedback to shape future content and remind them of the limited-time promotional offer for their ChatGPT automation course. The ultimate goal is to equip viewers with the knowledge to capitalize on AI opportunities without succumbing to FOMO or dispersion.