投稿者: ミハエル・ラングマイエル

As generative AI tools like Microsoft Copilot, ChatGPT or Gemini become more common, their real-world impact on jobs is starting to take shape. In a major 2025 study, Microsoft Research analyzed 200,000 anonymized U.S.-based conversations with Bing Copilot to see what kinds of work tasks people are using AI for—and how well the AI handles them. By matching these tasks to official job categories from the O*NET database, the study maps out which occupations are most exposed to AI capabilities. Instead of speculating how many jobs AI might replace, the researchers introduced an “AI applicability score” for each job, based on task frequency, success rate, and how much of the […]

In 2024, experts warned that the internet is rapidly reaching a tipping point: up to 50% of the images we see online could now be fake. AI tools like Midjourney, DALL·E 3, and Stable Diffusion are generating high-quality photos that look nearly identical to real ones — and they’re spreading across platforms like Twitter, Instagram, and TikTok at scale. Hany Farid, a professor at UC Berkeley and one of the world’s leading digital forensics experts, has spent the last 30 years building tools to detect manipulated media. In his TED Talk, he walks through how his team verifies whether an image is authentic or AI-generated. His work has been used by […]

Moonshot AI, the Alibaba-backed Chinese startup, has released Kimi K2, an open-source language model with 1 trillion total parameters and 32 billion activated parameters. This release signals China’s most significant attempt yet to close the gap between open-source and proprietary models like OpenAI’s GPT-4 and Anthropic’s Claude Opus. The model is available in two variants: Kimi-K2-Base for researchers and Kimi-K2-Instruct for general-purpose and agentic tasks. Built on a Mixture-of-Experts (MoE) architecture, Kimi K2 supports 128K context length and is designed for complex multi-step reasoning and tool usage. Moonshot has priced access to the model competitively. The input token cost is $0.15 per million, and the output token cost is $2.50 per million, making it significantly cheaper than GPT-4.1 or Claude Opus. According to CNBC, analyst […]

On July 17, OpenAI has unveiled ChatGPT Agent—a general-purpose AI assistant that not only talks but acts. More than just a chatbot, this new system can autonomously browse websites, use APIs, manipulate spreadsheets, run code in a terminal, generate editable documents—and crucially, do all of this using its own virtual computer. While Silicon Valley has been buzzing for over a year about so-called AI agents—intelligent software that can perform multi-step tasks on your behalf—OpenAI’s rollout marks one of the first times that such capabilities are available to mainstream users. Starting this week, subscribers to ChatGPT Pro, Plus, and Team plans can activate Agent Mode to let the model plan their week, generate pitch decks, […]

Most of us need a decent headshot at some point — for LinkedIn, a website, a speaking event, or just looking more put-together online. But getting one isn’t always easy. Professional photographers can be expensive, booking a shoot takes time, and not everyone feels comfortable posing in front of a camera. That’s why a lot of people are now using AI tools to create headshots from a regular selfie. And honestly? It works better than you might expect. With the right prompt, you can turn an everyday photo into something that looks clean, professional, and polished — without paying hundreds or leaving your home. Whether you want a serious corporate […]

Choosing the right AI assistant in 2025 goes beyond brand hype — it comes down to what the model can actually do. Whether you’re coding, writing, solving complex problems, or just chatting, four AI models dominate the conversation: Grok 4, ChatGPT-4o, Claude Opus 4, and Gemini 2.5 Pro. Each excels in different areas — one leads in deep reasoning, another in natural conversation, and others in creativity or multimodal capabilities. In the past year, AI development turned into an actual race. Each company released their answer to GPT-4o, which had over a year’s head start. Anthropic launched Claude Opus 4 in May 2025, Google DeepMind followed with Gemini 2.5 Pro in June, and xAI dropped Grok 4 in […]

OpenAI is preparing to launch its own AI-powered web browser, reportedly arriving within weeks. Built on Chromium and tightly integrated with ChatGPT, the browser is designed to compete directly with Google Chrome and reshape how people interact with the web. This move is about more than user experience. It’s a strategic push to take control of the same data stream that fuels Google’s core business. Chrome, used by over 3 billion people, captures vast amounts of behavioral data—searches, clicks, and browsing patterns—all of which feed into Google’s advertising engine. OpenAI wants to reroute that stream. With a growing base of over 500 million weekly ChatGPT users, the company is betting […]

New research reveals that even the most popular AI models can’t be trusted when it comes to basic safety. The most powerful LLMs today — including models from OpenAI, Google, and Cohere — were put through a rigorous safety benchmark. The results? Not great. Out of 20 models tested across 10 real-world risk areas, none passed all the tests, and 76% failed one of the most basic challenges: impersonation and privacy violations. If you’re building with AI or using it in your product, this should make you pause. Because once something goes wrong, you’re the one holding the bag — not the model provider. The Aymara Matrix: Safety Benchmark for LLMs The research comes […]

Grok 4 is the latest AI model from xAI, Elon Musk’s AI company. It launched on July 9 and immediately made headlines for scoring 50.7% on Humanity’s Last Exam—the highest result ever recorded. Unlike previous versions, Grok 4 was trained with tools built in from the start, giving it stronger reasoning and the ability to solve complex tasks like math, code, and simulations. The model is available through xAI and the X platform, but full access normally requires a paid subscription. If you’re on macOS, though, there’s a faster way to start using Grok 4—without needing an X account or using a browser. A Mac app called Fello AI gives […]

On July 9, 2025, Perplexity AI officially launched Comet, a new AI-powered web browser designed to reimagine how people interact with the internet. Marketed as the world’s first “AI-native browser,” Comet introduces features like persistent context memory, seamless AI assistance, and integrated task automation — aiming to replace the traditional tab-cluttered, search-centric web experience. The browser is currently limited to users subscribed to Perplexity Max, the company’s $200-per-month premium tier, with a broader invite-only rollout expected later this summer. According to CEO Aravind Srinivas, support for Android is “coming soon”, while Windows and macOS are supported from day one. Comet is built on Chromium, allowing users to keep their existing Chrome extensions, bookmarks, and settings. […]