Etiquetas: Reasoning

OpenAI has launched two new lightweight models, GPT-5.4 mini and GPT-5.4 nano, described as its most capable small models yet. The models are accessible to free and Go-tier ChatGPT users through a new “Thinking” option rather than a direct model selector. In testing, the Thinking models produced more detailed, multi‑step answers than the standard ChatGPT response, offering clearer reasoning for travel planning and online‑income strategies. While slightly less deep than the full‑size GPT-5.4 Thinking model available to Plus users, the mini and nano versions are faster and provide a notable upgrade for free users. Leer más

Google’s latest Gemini model, Gemini 3.1 Pro, shifts focus from raw speed to more thoughtful problem solving. While the earlier Gemini 3 Pro delivered fast, surface‑level answers, the 3.1 update introduces a “deep think” mode that deliberately slows responses to improve logical depth and handle complex tasks such as abstract reasoning, SVG generation, and intricate logistical planning. Early testing shows the new model excelling in nuanced scenarios where multi‑layered constraints and precise code output are required, positioning it as the preferred choice for developers and power users seeking higher‑quality AI output. Leer más

Researchers have found that artificial intelligence models keep memorized facts and reasoning abilities in separate parts of their neural networks. By analyzing the loss landscape, they discovered that memorized items create sharp spikes while reasoning produces smoother curves. The study also explored early techniques for removing specific data from models, noting that complete elimination cannot yet be guaranteed. These insights could guide future efforts to manage and protect sensitive information in AI systems. Leer más