AI Eng

Large AI models can lie to their users under pressure, according to a study called Model Alignment between Statements and Knowledge (MASK). While various tests and tools test AI for accuracy, the MASK benchmark was designed to determine whether an AI believes what it tells users — and under what circumstances it might provide incorrect information. The study tested 27 models from the GPT, Llama, Qwen, Claude, and DeepSeek families.

My name is Diana, I am a software developer at the product company BOTMAKERS LLC. For some time, I have been actively working on Instafill.ai technology, the main task of which is to relieve lawyers and ordinary users from the routine of filling out PDF forms. In this article, I want to reveal what solution became the foundation for our project and show how we improved it in practice. My task is not just to tell about my own experience, but also to show the evolution of the Instafill.ai idea: from the initial concept in the form of a simple Python script to a modern tool. This material will be useful for developers who want to optimize work with documentation and create intelligent systems for automation. Who knows — maybe our experience in developing Instafill.ai will become a starting point for your experiments!

In the online gambling industry, artificial intelligence technologies are opening up not only new horizons for entertainment, but also powerful tools for player protection. This includes personalizing limits and restrictions, as well as early detection and prevention of gambling addiction - problems that require special attention.

OpenAI has trained its GPT-4o model to generate images and has made it available to paid subscribers. The model has been shown to be adept at imitating the distinctive style of famous cartoons and anime, which has led social media users to take a liking to the way it converts famous images (including the White House brawl) into different settings. Here’s what intellectual property experts have to say about it.

A powerful historical update of ChatGPT has taken place. Now the chatbot can be used to generate images and process existing visuals and photos. Previously, I always said that ChatGPT or Gemini are, to put it mildly, not the best AI services for generating images and working with this content format. This is not their main profile. But the situation is changing.

OpenAI CEO Sam Altman has unveiled a major update to ChatGPT’s image generation capabilities. Now, the AI ​​chatbot can use OpenAI’s GPT-4o model to create and modify images and photos. What are the advantages and disadvantages of image generation in GPT-4o?

Cybercriminals are actively implementing artificial intelligence to scale attacks and generate profit. Malware generation, automated phishing, and self-learning AI worms are just some of the tools used in modern cybercrime. dev.ua tells how malicious AI models are changing the cyberthreat landscape.

Google and the Computer History Museum (CHM) have released the source code for AlexNet, a convolutional neural network (CNN) that many credit with transforming the AI ​​field in 2012. The Python code, now available on CHM’s GitHub page as open source software, offers AI enthusiasts and researchers a glimpse into one of the pivotal moments in IT history. We tell the story of the legendary neural network, which was trained on a computer in the bedroom of one of its creators' parents. It was created by Ukrainian Alex Kryzhevsky, OpenAI co-founder Ilya Sutzkever, and computer scientist Jeffrey Hinton.

This is a new episode of our show, Sho on AI, in which we analyze the latest news in the field of artificial intelligence, as well as learn a little about how to work with new AI tools, joke around and invite interesting guests. We often show scary things. In this episode, we show how the Chinese AI agent Manus simultaneously manages several dozen X-accounts and more.