This Week in Artificial Intelligence (November 18, 2025)

What happened in the AI world this week? What new innovations were introduced? In our “AI Diaries” series, we continue to provide a weekly record of the developments in this rapidly growing field.

Artificial intelligence technologies, which have recently become much more common in daily life, continue to evolve at a breathtaking pace. With remarkable breakthroughs occurring almost daily, significant milestones are being crossed one by one. In our “AI Diaries” series, we will continue to regularly document the progress of this technological revolution, which holds the potential to significantly change life on Earth in the coming era.

What Happened in the AI World This Week?

The standout news in the AI world this week was the introduction of GPT-5.1, while supply bottlenecks, which have recently become a critical problem, became more pronounced. Parallel to all this, new AI tools also entered our lives, as they do every week. Here are the top AI developments of the week:

OpenAI Introduces GPT-5.1

OpenAI has introduced GPT-5.1, an updated version of its flagship model GPT-5, which was released in August. The company describes this version as a “refresh” that makes ChatGPT smarter and more enjoyable to converse with.

The new version comes with two models: GPT-5.1 Instant and GPT-5.1 Thinking. The model now adapts its thinking time more precisely to the query, spending more time on complex problems while responding faster to simpler ones. OpenAI has also expanded the model’s conversational style options with the new update. The options now include Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy, and Cynical tones.

OpenAI not only introduced GPT-5.1 this week but also began testing group chats in ChatGPT. The feature, currently available in only a few countries, is expected to be activated in other countries in the future.

Following GPUs, Memory and HDD Supplies Also Tighten

The increasing demand for AI hardware continues to create imbalances in the semiconductor market. According to recent reports, as delivery times for enterprise-grade HDDs exceed two years, companies have begun turning to SSDs. Large AI centers, in particular, are purchasing high-capacity QLC NAND drives to mitigate the impact of supply chain issues. It is reported that some manufacturers’ production capacity is fully booked until 2026, and AI companies are already stockpiling NAND. This means we could see an SSD shortage in both the enterprise and consumer segments in the near future.

On the other hand, a similar supply shortage has emerged for DRAM. As manufacturers struggle to keep up with the increasing demand from AI companies, prices in this area have also begun to rise. In fact, this week, Samsung increased RAM prices by 60% due to the memory shortage.

Google Introduces SIMA 2: A Self-Learning AI System in Virtual Worlds

Google DeepMind has introduced SIMA 2, representing the next major step in AI research. The new-generation general-purpose agent, combined with Gemini’s advanced language and reasoning capabilities, is no longer just a system that follows commands. Instead, it gains the ability to understand and interact with the virtual world it inhabits.

When placed in new environments, SIMA 2 generates tasks from another Gemini model, and an independent reward model scores the agent’s performance. In this loop, SIMA 2 learns from its own mistakes, developing new behaviors through feedback generated entirely by AI. DeepMind sees SIMA 2 as a gateway to more comprehensive robotics platforms in the future. Researchers state that a robot operating in the real world needs high-level comprehension and logic skills, and SIMA 2 operates precisely at this upper layer.

Broadcom Develops AI Chip for Real-Time, Internet-Free Translation

Broadcom has announced the development of a new AI chip that performs real-time audio translation directly on the device. This new chipset, developed in partnership with CAMB.AI, can perform audio translation, dubbing, and screen content description without a cloud connection. This feature aims to increase accessibility while offering significant advantages in privacy and performance.

The new AI chip promises ultra-low latency and enhanced privacy by performing all operations locally on the device. Since data is not sent to the cloud, wireless bandwidth usage is also greatly reduced. This allows for much faster translation and description without relying on an internet connection.

A Black Mirror Episode Becomes Reality: “Living” AI Avatars of the Deceased Introduced

The Los Angeles-based startup 2Wai has introduced its new application that creates digital avatars of deceased individuals, allowing users to interact with them. This ethically controversial application, as you might guess, has gained widespread attention on social media. This technology, reminiscent of the “Be Right Back” episode of Black Mirror from years ago, allows for the creation of interactive digital characters called HoloAvatars. 2Wai states that these avatars can have the same facial features as the desired individuals, speak with the same voice, and even possess a personality fed by the person’s memories.

Anthropic’s AI Becomes a Cyberattack Tool for Chinese Hackers

It was announced that a state-sponsored Chinese hacker group attempted a comprehensive cyberattack against 30 corporate and political targets worldwide using Claude, and was partially successful. Described as the first documented large-scale cyberattack conducted without human intervention in history, the attackers split the attack code into pieces to hide their malicious intent and tricked the AI as if they were a cybersecurity firm. Later, within the developed automation framework, the model captured usernames and passwords, collecting a large amount of private data.

Baidu Claims Its New AI Surpasses GPT-5

Baidu introduced its new open-source multimodal AI model named ERNIE-4.5-VL-28B-A3B-Thinking. The Chinese tech giant claims its new model surpasses OpenAI’s GPT-5 and Google’s Gemini 2.5 Pro on many benchmarks.

Released under the flexible Apache 2.0 license, the model uses a “Mixture of Experts” (MoE) architecture with a total of 28 billion parameters. Despite having 28 billion total parameters, only 3 billion parameters are activated in each run. This significantly reduces computational costs while maintaining high performance.

Microsoft to Acquire Not Only OpenAI’s AI, But Its AI Chips Too

Microsoft’s investment in OpenAI years ago continues to provide the company with significant advantages. Microsoft, which has risen to a prominent position in the AI field by using software developed by OpenAI, is now preparing to do the same on the hardware side. Microsoft CEO Satya Nadella announced that the custom AI chips OpenAI is working on will also be integrated into Microsoft’s hardware development processes.

AI Tools Introduced This Week

• Marble: Allows you to create interactive 3D worlds by giving AI just a written prompt or a single image. You can further detail this 3D world by providing multiple images or prompts.
• Time-to-Move: Enables users to determine the direction and manner of movement for characters or objects within a video.
• Step-Audio-Editx: This open-source model has been released as one of the most remarkable models on the market for text-to-speech. Step-Audio-Editx can vocalize texts in different voices, reflecting different emotions.
• Evtar: Superimposes the clothing from a visual you share onto a photo of a person you provide, showing how it would look.

AI News in Brief

• Analysis shared by Forbes revealed that Sora‘s high usage costs are causing OpenAI to lose $15 million per day.
• ChatGPT is in trouble in Germany. A German court ruled that it committed copyright infringement by using German artists’ song lyrics without permission.
• Firefox announced it is working on a new AI browsing feature called AI Window, which includes an AI assistant and a chatbot.
• AMD has acquired MK1, a company specializing in high-speed inference technologies, to strengthen its AI strategy. This move is set to bring new levels of performance and efficiency to Instinct GPUs.
• Anthropic signed a $50 billion data center deal with Fluidstack. The facilities, to be built in Texas and New York, will become operational in 2026, strengthening the AI infrastructure.

You Might Also Like;