Author HubMatthias Bastian

Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.
AI research

Apple AI researchers question OpenAI's claims about o1's reasoning capabilities

AI research

Study reveals major reasoning flaws in smaller AI language models

AI and society

OpenAI claims New York Times' prompting strategy violates its terms of service

Ad
Ad
Ad
AI in practice
Short

O2? Can't do! OpenAI has confirmed it will bypass the name "o2" for its next reasoning model to avoid trademark conflicts with British telecommunications giant O2. Instead, the company will jump straight to "o3" for its next release. The company is currently investing heavily in developing the o-series, its first "reasoning" models. To support this work, OpenAI is developing a new large language model called "Orion" that will generate synthetic training data, according to The Information. Microsoft's Phi-4 model recently demonstrated the effectiveness of incorporating synthetic data in AI training, at least in achieving impressive benchmark results. Interestingly, Sébastien Bubeck, who helped create the Phi series at Microsoft, has since joined OpenAI's team.

Ad
Ad
AI in practice
Short

AI answer company Perplexity has acquired Carbon, a startup specializing in connectors that link external data sources to LLMs. The acquisition aims to let Perplexity users directly connect services like Notion and Google Docs to their AI search interface. As part of the deal, Carbon's entire team will join Perplexity to accelerate the development of these integration features. The acquisition highlights a growing pattern in the AI industry, where major platforms like ChatGPT and Perplexity are converging to offer similar solutions: chat interfaces that combine Internet access with connections to users' personal data sources. Beyond the big players, there are a myriad of B2B AI solutions doing similar things.

Ad
Ad
AI in practice
Short

OpenAI is making ChatGPT more accessible by letting users interact with the AI assistant through phone calls and WhatsApp messages. U.S. users can now call ChatGPT for free at 1-800-242-8478, with a monthly limit of 15 minutes of conversation time. The phone service works with any device, including landlines and basic mobile phones, eliminating the need for a smartphone or internet connection to access ChatGPT. For users worldwide, OpenAI has added WhatsApp support, allowing anyone to chat with ChatGPT by saving the service's number as a contact. Currently, the WhatsApp integration only supports text messages, though OpenAI plans to add authentication features that will enable advanced capabilities like image recognition. The company says it developed these new features during an internal hackathon and implemented them in just a few weeks.

Google News

Bank details

IBAN: DE87 1203 0000 1086 0070 75
Account holder: DEEP CONTENT GbR
Purpose: Support THE DECODER
AI in practice

OpenAI unveils o3, its most advanced reasoning model yet

AI research

Study shows: 'Test-time compute scaling' is a path to better AI systems

AI in practice

Google launches Gemini 2.0, focusing on AI agents and multimodal capabilities

Google News