AI blog

According to Sam Altman, CEO of OpenAI, saying “thank you” or “please” to ChatGPT represents significant energy consumption. Each additional word is treated as a token, requiring more computing power. While the individual impact seems negligible, the cumulative effect of millions of daily interactions could cost tens of millions of dollars in electricity globally. This paradox places us in the face of a dilemma: should we prioritize our habits of politeness or adopt digital sobriety? A balanced approach would be to be concise with AIs while maintaining our courtesy for human interactions. Ironically, in this technological context, politeness comes at a very real price.

The new OpenAI models, o3 and o4-mini, mark a breakthrough: AI is no longer content with responding, it reasons and acts independently. These advances make it possible to deal with complex business issues from end to end — financial modeling, anomaly detection, opportunity scoring, logistics simulations — without human supervision. Thanks to their integration via API, these AIs are transforming business tools.

La galerie de prompts Copilot proposée par Microsoft est une ressource pratique pour tous ceux qui utilisent Microsoft 365 et souhaitent optimiser leur usage de l’IA. Elle regroupe des exemples de requêtes prêtes à l’emploi, filtrables par application (Word, Excel, Outlook…), type de tâche (rédiger, organiser, résumer…) et type de travail (communication, gestion de projet, analyse…). Cette galerie permet de gagner du temps, de structurer plus efficacement ses demandes à Copilot et de découvrir des cas d’usage concrets facilement adaptables.

Knowledge Graphs allow data to be linked together in an explicit and logical way, unlike SQL databases (rigid) or NoSQL databases (flexible but not connected). They offer an ideal framework for AIs like ChatGPT, providing context, accuracy, and reducing hallucinations. Thanks to their flexibility, they perfectly complement existing systems and pave the way for advanced use cases: AI assistants, internal search engine, smart recommendation, etc.

OpenAI has just launched the OpenAI Academy, a free platform dedicated to those who want to use AI to meet concrete challenges. Training, API credits, technical support and the international community: this initiative marks a key step towards a more accessible, field-oriented AI. In this article, Strat37 looks back at this launch and what it reveals about current trends in the adoption of artificial intelligence in businesses.

OpenAI has just launched Sora, an image generator integrated into ChatGPT via the Gpt-4o model. This new tool significantly improves image quality compared to the old DALL-E, with more accurate visuals, better text readability, and an increased ability to handle complex requests. Even better, it is available for free to all ChatGPT users.

The emergence of generative AI, especially since the launch of ChatGPT, is profoundly transforming the developer market. According to data from the Bureau of Labor Statistics, traditional programmer positions fell by -27.5%, while senior developer and architect positions fell by only -0.3%. This trend shows that automation mostly impacts basic programming tasks, while skills in system design, complex problem solving, and collaboration remain essential. The developers who will succeed will be those who know how to use AI as an efficiency tool, while continuing to learn and refine their analytical and organizational skills. For companies, this means training their teams in generative AI tools and recruiting profiles capable of bringing real added value.

Tokens are the units of text that an artificial intelligence model like GPT-4 uses to understand and generate responses. They can be whole words, parts of words, or even punctuations. Each interaction with the AI consumes tokens, both for input and for output. Understanding this concept is essential to optimize the use of AI models, especially in APIs where each token used has a cost. The article explains the concept using an excerpt from The Little Prince and provides tips for using tokens properly.

According to Gartner, 85% of data projects fail, often due to a lack of a clear objective, poorly exploited data, or a lack of buy-in from business teams. To succeed, it is essential to define a specific business need, to structure and clean the data beforehand, and to involve the teams from the start. The tool must be integrated into existing processes and evolve with the company to ensure a lasting impact.

Mistral AI launches Mistral OCR, a new optical character recognition solution that goes beyond simple text extraction by integrating the management of tables, equations and complex layouts. With an accuracy of 98.96% and a processing speed of up to 2,000 pages per minute, this tool is a reference for automating document processing. It has applications in scientific research, heritage preservation, customer service, and regulatory document management. Available on The Platform, it will soon be accessible via the main cloud providers.

Google has extended its AI Overviews feature to several European countries, including Switzerland, Germany, and Italy. These summaries generated by artificial intelligence offer direct answers to users, raising concerns about the impact on organic traffic and SEO strategies. To adapt, advertisers and content creators will have to adjust their approaches and offer unique added value.

Finding relevant and reliable information on the Internet is a constant challenge. Between the multitude of sources, the sorting of results and the risk of misinformation, the search for information takes time and requires increased expertise. That's where Deep Search, ChatGPT's new feature, comes in. Unlike traditional search engines that list links, Deep Search goes further: it collects, analyzes, and synthesizes information from the web to provide a structured and sourced document. A promising advance for professionals, but also a solution that poses certain limitations. We tested Deep Search to understand how it works and assess its effectiveness.

AI in Business Intelligence should not flood the user with an avalanche of useless indicators, but on the contrary prioritize and contextualize data to make it usable. An effective approach is based on fewer numbers, better explained, better context, and intelligent prioritization of insights. StratBoard™ embodies this philosophy by putting forward relevant and actionable data, rather than an excess of information that is difficult to interpret.

AI is a powerful tool that can either accelerate our skills or make us dependent if we rely too much on it. Rather than seeing it as a substitute, it should be used as a lever by maintaining control and strengthening our own skills. The balance between automation and human expertise is key to getting the most out of it without losing control.

The debate between open-source AI and proprietary models is often presented as a binary choice, when in reality, everything depends on the need. Proprietary models offer performance and simplicity, while open-source models offer flexibility and control. The challenge is not to choose a side, but to adopt the most appropriate solution according to business constraints, costs and the need for data control.