AI Diaries: This week in the world of artificial intelligence (January 27, 2026)
AI technologies, which have become increasingly integrated into our daily lives, continue to evolve at breakneck speed. While we witness remarkable developments almost every day, significant milestones are being crossed one by one. In our “AI Diaries” series, we continue to record the evolution of this technological revolution that holds the potential to dramatically reshape life on Earth.
What Happened in the World of AI This Week?
The events of the past week indicate that we are on the verge of a major shift in the artificial intelligence landscape. On one hand, efforts to reach AGI (Artificial General Intelligence) have accelerated; on the other, timelines have been set for the release of dedicated AI hardware. Parallel to these strategic moves, new AI tools have entered our lives, as they do every week.
Here are the standout developments from this week in the world of AI:
AGI Predictions: The Threshold is Near
AGI—artificial general intelligence that possesses cognitive abilities indistinguishable from or surpassing humans—has been a frequent topic of discussion lately. Judging by the successive statements made this week, industry leaders expect this threshold to be crossed in the very near future.
Speaking at the World Economic Forum in Davos this week, Elon Musk predicted that AI will surpass individual human intelligence by the end of 2026, and exceed the collective intelligence of all humanity within five years.
A similar sentiment came from the Google camp. Shane Legg from Google DeepMind stated that AGI is now on the horizon and predicted radical economic transformations as a result. While there seems to be a consensus on the timing, companies differ on how to get there. Google DeepMind CEO Demis Hassabis noted this week that Large Language Models (LLMs like ChatGPT) alone are not enough to reach AGI. According to Hassabis, AI requires a true “world model” to cross this threshold. DeepMind is taking significant steps to create this next generation of AI, as reflected in Legg’s comments.
Nvidia Ushers in a New Era with PersonaPlex-7B-v1
Nvidia researchers have announced PersonaPlex-7B-v1, a new speech model targeting natural, seamless, and real-time voice interactions. Developed with a full-duplex structure, this model distinguishes itself from classic voice assistant architectures by its ability to listen and speak simultaneously.
Traditional voice assistants use a multi-stage structure (ASR for speech-to-text, an LLM for response, and TTS for audio generation). This approach causes latency at every stage and fails to support fundamental human speech features like interruption or natural back-and-forth flow. PersonaPlex eliminates this multi-layer chain, performing both understanding and speech generation simultaneously within a single Transformer model. It works on a continuous audio stream encoded by a neural codec, making it ideal for scenarios requiring natural dialogue flow and low latency.
AI Hardware is Coming: Apple and OpenAI Set Their Calendars
The AI race, which has so far progressed through software and voice assistants, appears set to jump to hardware by the end of 2026. OpenAI, having recruited Apple’s legendary designer Jony Ive, announced that it will unveil its first AI hardware led by Ive later this year.
Simultaneously, reports have surfaced that Apple has begun developing an AI-powered wearable device. Like OpenAI, Apple is preparing to launch AI-focused hardware. However, Apple seems likely to follow the industry from behind in this regard, with a targeted release date of 2027.
South Korea Enacts the World’s First Comprehensive AI Law
South Korea has become one of the first countries to implement a comprehensive legal framework for AI, with the Framework Act on Artificial Intelligence coming into force on January 22. The regulation aims to clarify where and how AI is used and assigns specific responsibilities.
The new law focuses on “high-impact” areas such as nuclear safety, water production, transport infrastructure, healthcare, and credit scoring. In these areas, fully autonomous AI systems are prohibited, and human oversight is mandatory. The law also enforces strict transparency, requiring companies to inform users when they are interacting with generative AI or high-impact systems.
xAI Launches the Gigawatt-Scale Colossus 2
Elon Musk’s AI company, xAI, has activated Colossus 2, the world’s first AI training cluster operating at a gigawatt scale. Like its predecessor, Colossus 2 will be used to train Grok. While Colossus 1 took 122 days to become fully operational, the new system surpassed the 1 gigawatt (GW) threshold the moment it went online. Musk stated that the system will be upgraded to 1.5 GW in April, with an ultimate goal of approximately 2 GW total capacity.
Apple Prepares a Massive Overhaul for Siri
Apple is preparing to make a significant move in the AI space, where it has long faced criticism. According to Mark Gurman from Bloomberg, the company will transform Siri from a classic voice assistant into an AI chatbot similar to ChatGPT.
Reportedly, Apple will shelve the current Siri interface entirely. In its place, it will offer an experience where users can interact via both text and voice, akin to modern AI chatbots. The new Siri is expected to launch on iPhone, iPad, and Mac devices before the end of this year.
OpenAI Quietly Trains Robots
OpenAI is making silent progress in robotics. According to Business Insider, the company established a robotics lab in San Francisco in February 2025 and has since quadrupled the unit’s size. Approximately 100 data collectors and at least a dozen robotics engineers are producing training data by remotely controlling robotic arms.
The work focuses on low-cost robotic arms rather than full humanoids. Employees use a 3D-printed control device called “GELLO” to map human hand movements directly to the robot arms. These robots are being trained on everyday tasks like putting bread in a toaster or folding laundry.
New AI Tools Released This Week
- FlowAct-R1 (ByteDance): Allows users to livestream using digital avatars. It stands out with its reaction speed and avatars that change their attitude based on the conversation content.
- Waypoint-1: Creates interactive worlds that can be edited in real-time, reflecting user commands into video almost instantly.
- OmniTransfer: Allows you to easily transfer camera movements, actions, or effects from one video to another.
- Qwen3-TTS (Alibaba): An open-source model delivering highly successful text-to-speech results, including voice cloning capabilities.
- VibeVoice-ASR (Microsoft): A new tool capable of deciphering voice recordings and producing written transcripts.
AI Shorts: News in Brief
- ChatGPT has activated age estimation features to protect young users, automatically applying content filters to minors.
- Elon Musk announced that Tesla’s Dojo3 supercomputer project will restart as the AI5 chip design nears completion.
- OpenAI is working on commerce features, including shopping carts and seller upload pages, aiming to turn ChatGPT into an e-commerce platform.
- ChatGPT Atlas, OpenAI’s AI-powered browser, received a major update including automatic switching between search tools and tab groups.
- Non-game mobile app spending surpassed game revenues for the first time in 2025, driven by Generative AI apps. AI in-app purchase revenue tripled to over $5 billion.
- OpenAI has turned its route to the Middle East for a new investment round seeking approximately $50 billion.
You Might Also Like;
