OpenAI

ElevenLabs unveils fast and accurate speech-to-text solution

ElevenLabs, a Palo Alto-based AI startup valued at $3.3 billion, has introduced its first stand-alone speech-to-text model, Scribe. The company, which is better known for its audio-generation capabilities, now aims to disrupt the speech detection market by providing a faster and more accurate alternative to existing models like Whisper and Deepgram. Scribe supports over 99 languages, with top accuracy in more than 25, including English, French, and Spanish.

The new AI model has already outperformed competitors like Google’s Gemini 2.0 Flash and OpenAI’s Whisper Large V3 in benchmark tests. It also includes features like speaker diarisation, accurate subtitles, and sound event tagging, which could appeal to customers in media and content creation. While Scribe currently only works with pre-recorded audio, ElevenLabs plans to release a real-time version soon.

Priced at $0.40 per hour of transcribed audio, Scribe offers a competitive rate, though some rivals currently offer lower prices. With this move into speech-to-text, ElevenLabs is positioning itself to expand its AI offerings and challenge established players in the field.

For more information on these topics, visit diplomacy.edu.

Baidu unveils upgraded Ernie 4.5 model in March

Baidu is set to launch the next generation of its Ernie AI model, the Ernie 4.5, in mid-March. This upgraded version will feature improved reasoning capabilities and enhanced multimodal functions, allowing it to process and integrate a variety of data formats, including text, images, audio, and video. Baidu also plans to make the Ernie 4.5 series open source from June 30, marking a significant shift in its approach to AI development.

The Chinese tech giant has faced fierce competition in the AI race, particularly from the rising startup DeepSeek, whose models are seen as rivals to leading US systems at a much lower cost. Despite claiming that Ernie’s performance is comparable to OpenAI’s GPT-4, Baidu has struggled to gain widespread adoption of its AI model. The emergence of DeepSeek has further complicated Baidu’s aspirations in the sector, leading the company to reassess its strategy.

Baidu’s CEO, Robin Li, who once advocated for keeping AI models closed-source, has acknowledged the success of DeepSeek and other competitors. He now views the open-source approach as essential for the future of AI development. This shift comes as Baidu continues to compete with other tech giants, including Alibaba, which recently announced its plans to make its video and image-generating AI model, Wan 2.1, open source.

For more information on these topics, visit diplomacy.edu.

Estonia introduces AI Leap programme for education

Estonia has launched a new initiative aimed at preparing students and teachers for the age of AI. The ‘AI Leap’ programme will provide access to popular AI chatbots, including an educational version of ChatGPT, to help build digital skills. Starting in September 2025, the programme will involve 20,000 high school students and 3,000 teachers, with plans to expand to vocational schools and an additional 38,000 students and 3,000 teachers in 2026.

Education Minister Kristina Kallas emphasised that Estonia’s economic competitiveness depends on how well the country adapts to AI, ensuring young people are equipped for the future. As part of the initiative, Estonia will also invest in teacher training to support the integration of AI in classrooms.

The programme is a public-private partnership, with negotiations underway with major AI companies, including OpenAI and Anthropic. OpenAI has expressed its pride in collaborating with Estonia to bring ChatGPT Edu to the education system, aiming to better prepare students for the workforce. Estonia’s use of AI in education is seen as a model that other countries may follow as the EU pushes to increase digital skills across Europe by 2030.

For more information on these topics, visit diplomacy.edu.

ChatGPT takes a Rorschach test and reveals its findings

For over a century, the Rorschach inkblot test has been used to explore human psychology by revealing the hidden facets of the mind through personal interpretations of ambiguous shapes. The test leverages a phenomenon known as pareidolia, where individuals perceive patterns, such as animals or faces, in random inkblots. Now, thanks to the advances in artificial intelligence, this test has been used to explore how AI interprets these same images.

In an intriguing experiment, ChatGPT was shown five common inkblots to see how it would respond. Unlike humans, who often project their emotions or personal experiences onto the images, the AI offered more literal interpretations, identifying symmetrical shapes or common visual features. However, these responses were based purely on patterns it has been trained to recognise, rather than any true emotional connection to the inkblots.

The AI’s responses were consistent with what it had learned from vast datasets of human interpretations. But while humans might see a butterfly or a skull, the AI merely recognised a shape, demonstrating a key difference between human cognition and machine processing. This experiment highlights the unique human ability to attach emotional or symbolic meaning to abstract visuals, something AI is not equipped to replicate.

For more information on these topics, visit diplomacy.edu.

Google prices Veo 2 as competition with OpenAI heats up

Google has quietly disclosed the pricing for its Veo 2 video-generating AI model, setting the cost at 50 cents per second of video. This translates to $30 per minute or $1,800 per hour, positioning Veo 2 as a premium tool for AI-generated video content.

While unlikely to produce big-budget epics, the model can create clips of two minutes or longer, as highlighted in Google’s initial announcement.

Jon Barron, a Google DeepMind researcher, compared Veo 2’s cost to Hollywood productions, noting that Avengers: Endgame had a production cost of around $32,000 per second.

Though Veo 2 operates at a fraction of that price, its output serves different purposes, targeting creators seeking efficient video generation without traditional production expenses.

The AI video generation space has grown increasingly competitive, with OpenAI recently releasing its Sora model to ChatGPT Pro subscribers for $200 a month. Google’s move to price Veo 2 publicly reflects the broader push to commercialise AI video tools as demand surges among content creators and businesses.

For more information on these topics, visit diplomacy.edu. 

China and North Korea-linked accounts shut down by OpenAI

OpenAI has removed accounts linked to users in China and North Korea over concerns they were using ChatGPT for malicious activities.

The company cited cases of AI-generated content being used for surveillance, influence campaigns, and fraudulent schemes. AI tools were employed to detect the operations.

Some accounts produced news articles in Spanish that criticised the US and were later published under a Chinese company’s byline. Others, potentially connected to North Korea, created fake resumes and online profiles in an attempt to secure jobs at Western firms.

A separate operation, believed to be tied to financial fraud in Cambodia, used ChatGPT to generate and translate comments on social media.

The US government has raised concerns over China’s use of AI to spread misinformation and suppress its population. Security risks associated with AI-driven disinformation and fraudulent activities have led to increased scrutiny of how such tools are being used globally.

OpenAI’s ChatGPT remains the most widely used AI chatbot, with over 400 million weekly active users. The company is also in discussions to secure up to $40 billion in funding, which could set a record for a private firm.

For more information on these topics, visit diplomacy.edu.

OpenAI’s Operator reaches more global markets amid growing competition

OpenAI has expanded its AI agent, Operator, to ChatGPT Pro subscribers in several new countries, including Australia, Brazil, Canada, India, Japan, Singapore, South Korea, and the UK.

Initially launched in the United States in January, Operator allows users to delegate tasks like booking tickets, making restaurant reservations, filing expense reports, and shopping online.

The tool remains unavailable in the EU, Switzerland, Norway, Liechtenstein, and Iceland.

Operator is currently exclusive to the $200-per-month ChatGPT Pro plan and can only be accessed through a dedicated webpage.

The AI agent operates within a separate browser window, which users can control at any time. OpenAI has announced plans to eventually integrate Operator into all ChatGPT clients, expanding its accessibility.

The AI agent market is becoming increasingly competitive, with Google, Anthropic, and Rabbit developing similar tools. While Google’s project is still waitlisted, Anthropic offers its agentic interface via API, and Rabbit restricts its action model to users of its proprietary device.

OpenAI’s broader rollout of Operator aims to solidify its position in this fast-evolving space.

For more information on these topics, visit diplomacy.edu.

AI adoption surges as OpenAI passes 400 million users

OpenAI’s weekly active users exceeded 400 million in February, marking significant growth in AI adoption. The company had 300 million weekly users in December, and its paid business subscribers have more than doubled since September, surpassing 2 million.

Competition in the AI space remains intense, with China’s DeepSeek claiming its new model can rival or outperform Western alternatives at a lower cost.

Surging demand has led to service outages for the startup, while questions persist about its access to Nvidia’s H800 chips despite US export restrictions.

OpenAI also reported a sharp increase in developer traffic for its AI models, with usage of its latest o3 model rising fivefold since its January launch.

For more information on these topics, visit diplomacy.edu.

Mira Murati launches AI startup Thinking Machines Lab

Former OpenAI chief technology officer Mira Murati has launched a new AI startup called Thinking Machines Lab, backed by a team of around 30 researchers and engineers from companies such as OpenAI, Meta, and Mistral. The startup aims to create AI systems that encode human values and address a wider range of applications than existing rivals, according to a blog post from the company.

Murati’s new venture demonstrates her ability to attract top talent, with two-thirds of the team made up of former OpenAI employees. Among them are Barret Zoph, a well-known researcher who joined Murati in leaving OpenAI in September, and John Schulman, OpenAI’s co-founder and the startup’s chief scientist. Schulman previously left OpenAI for Anthropic to focus on AI alignment, a key goal of Thinking Machines Lab.

The company’s approach differentiates itself by combining research and product teams in the design process. Thinking Machines Lab plans to contribute to AI alignment research by sharing code, datasets, and model specifications. Murati, now CEO of the startup, has previously played a major role in developing ChatGPT, and her exit from OpenAI reflects a broader trend of high-profile departures amid changes at the company.

For more information on these topics, visit diplomacy.edu.

AI tools approved for New York Times staff

The New York Times has officially approved the use of AI tools for its editorial and product teams, allowing AI to assist with tasks such as generating social media copy, writing SEO headlines, and coding. As part of this shift, the publication has introduced an internal AI tool called Echo and outlined strict guidelines for AI use in news production.

While AI can support research, suggest edits, and help brainstorm interview questions, staff have been warned not to use it to draft or substantially revise articles, nor to input confidential source information. The paper is also exploring AI-generated voice articles and translations into other languages.

This move comes as The New York Times remains engaged in a lawsuit against OpenAI and Microsoft, accusing them of copyright violations for training generative AI on its content. Despite these legal disputes, the paper has approved AI programs such as GitHub Copilot, Google’s Vertex AI, and OpenAI’s API for select business applications.

For more information on these topics, visit diplomacy.edu.