07 de fevereiro de 2026

Revisão Semanal de IA 26.02.07 - Substack

AI Week in Review 26.02.07

Top AI news of the week and recommended reads on AI.

AI News of the Week
Recommended AI reads of the week

AI News of the Week

1. Google DeepMind’s AlphaGeometry solves hard geometry problems

Google DeepMind’s AlphaGeometry solved 25 out of 30 geometry problems from the International Mathematical Olympiad (IMO), approaching the level of a human gold medalist. This system combines a neural language model with a symbolic deduction engine, allowing it to explore creative solutions and verify their correctness. The paper was published in Nature and the code is available on GitHub.

2. ChatGPT’s memory feature is rolling out

OpenAI is rolling out a new memory feature for ChatGPT, allowing it to remember details from past conversations. This feature is available to a small portion of free and Plus users and will be extended to more users in the coming weeks. The memory feature can be enabled or disabled in settings, and users can instruct ChatGPT to remember specific details or let it pick up information over time. This feature is not available for ChatGPT Team, Enterprise, and ChatGPT for EDU users, and there are separate controls for memory in custom GPTs.

3. Stability AI launches Stable Cascade

Stability AI has released Stable Cascade, a new text-to-image model based on the Würstchen architecture. Stable Cascade is designed to be faster and more efficient than previous models, using a three-stage approach to generate images. The model is available under a non-commercial license, but users can apply for a research and commercial license. Stability AI claims that Stable Cascade outperforms other models in terms of speed and efficiency, making it a significant advancement in the field of text-to-image generation.

4. Google announces Gemini API & AI Studio

Google has announced the Gemini API and AI Studio, providing developers with access to Google’s latest AI models. The Gemini API allows developers to build applications using Google’s powerful language models, while AI Studio offers a web-based IDE for prototyping and developing AI-powered applications. The Gemini API is available in over 180 countries and supports text-based input and output. Google also announced upcoming features, including context caching and function calling, to enhance the capabilities of the API.

5. Microsoft’s AI CEO says AI will be smarter than humans

Mustafa Suleyman, the CEO of Microsoft AI, has stated that AI will be smarter than humans in a few years. In an interview with MIT Technology Review, Suleyman discussed the rapid progress of AI and its potential impact on society. He emphasized the need for responsible development and regulation of AI to ensure its benefits are realized while mitigating potential risks. Suleyman’s comments reflect the growing concern among AI experts about the pace of AI development and its implications for humanity.

6. Mistral releases new AI model

Mistral, a French AI startup, has released a new AI model called Mixtral 8x7B. This model is designed to be more efficient and powerful than its predecessors, with improved performance on various benchmarks. Mixtral 8x7B is available under a non-commercial license, but users can apply for a research and commercial license. Mistral’s new model demonstrates the company’s commitment to advancing AI technology and providing accessible tools for developers and researchers.