digitado

Shock! Shock! I learned yesterday that an open problem I’d been working on for several weeks had just been solved by Claude Opus 4.6 – Anthropic’s hybrid reasoning model that had been released three weeks earlier! It seems that I’ll have to revise my opinions about “generative AI” one of these days. What a joy it is to learn not only that my conjecture has a nice solution but also to celebrate this dramatic advance in automatic deduction […]

Ver mais

Alibaba’s Qwen team has released the Qwen3.5 Small Model Series, a collection of Large Language Models (LLMs) ranging from 0.8B to 9B parameters. While the industry trend has historically favored increasing parameter counts to achieve ‘frontier’ performance, this release focuses on ‘More Intelligence, Less Compute.‘ These models represent a shift toward deploying capable AI on consumer hardware and edge devices without the traditional trade-offs in reasoning or multimodality. The series is currently available on Hugging Face and ModelScope, […]

Ver mais

In this tutorial, we design an end-to-end, production-style analytics and modeling pipeline using Vaex to operate efficiently on millions of rows without materializing data in memory. We generate a realistic, large-scale dataset, engineer rich behavioral and city-level features using lazy expressions and approximate statistics, and aggregate insights at scale. We then integrate Vaex with scikit-learn to train and evaluate a predictive model, demonstrating how Vaex can act as the backbone for high-performance exploratory analysis and machine-learning workflows. Copy […]

Ver mais

Alibaba has released OpenSandbox, an open-source tool designed to provide AI agents with secure, isolated environments for code execution, web browsing, and model training. Released under the Apache 2.0 license, the proposed system targets to standardize the ‘execution layer’ of the AI agent stack, offering a unified API that functions across various programming languages and infrastructure providers. The tool is built on the same internal infrastructure Alibaba utilizes for large-scale AI workloads. The Technical Gap in Agentic Workflows […]

Ver mais

Google has released Gemini 3.1 Flash-Lite, the most cost-efficient entry in the Gemini 3 model series. Designed for ‘intelligence at scale,’ this model is optimized for high-volume tasks where low latency and cost-per-token are the primary engineering constraints. It is currently available in Public Preview via the Gemini API (Google AI Studio) and Vertex AI. https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-lite/? Core Feature: Variable ‘Thinking Levels’ A significant architectural update in the 3.1 series is the introduction of Thinking Levels. This feature allows […]

Ver mais

In this tutorial, we demonstrate how to efficiently fine-tune a large language model using Unsloth and QLoRA. We focus on building a stable, end-to-end supervised fine-tuning pipeline that handles common Colab issues such as GPU detection failures, runtime crashes, and library incompatibilities. By carefully controlling the environment, model configuration, and training loop, we show how to reliably train an instruction-tuned model with limited resources while maintaining strong performance and rapid iteration speed. Copy CodeCopiedUse a different Browser import […]

Ver mais

Everyone believes that the digits of constants such as π or √2 cannot be distinguished from a sequence of random bits. The first few trillion successfully pass all tests of randomness. However, proving that they indeed behave perfectly randomly is arguably one of the oldest and most difficult unsolved math conjectures. So far, nobody succeeded in proving even the most basic facts for any of these constants, for instance: In the binary digits of π, is the proportion […]

Ver mais

Gemini 3.1 Flash-Lite is our fastest and most cost-efficient Gemini 3 series model yet.

Ver mais

Learn more about Google DeepMind’s Project Genie and how to write prompts to create your own worlds.

Ver mais

Reactions rippled through Alibaba’s Qwen team after tech lead Junyang Lin stepped down following a major model launch.

Ver mais