OpenAI has introduced a new phone feature for ChatGPT, initially available only in the USA. Users can call the free number 1-800-ChatGPT to speak with the chatbot for up to 15 minutes per month. German users will soon be able to reach ChatGPT via WhatsApp. This service is text-based and does not support image processing. Communication occurs through WhatsApp Business, and users must agree to terms allowing conversations to be monitored for security reasons.
OpenAI is negotiating with its non-profit parent organization regarding its control rights, which could cost billions. CEO Sam Altman and colleagues need to agree on a fair price for relinquishing control, as reported by the New York Times. Investors expect a restructuring within two years, tied to recent funding. One option is converting to a “Public Benefit Corporation,” where the non-profit retains part of the company. Negotiations also involve a clause with Microsoft about when OpenAI reaches artificial general intelligence, which could affect their partnership.
North America faces potential energy shortages due to the construction of energy-intensive AI and crypto-mining data centers. The North American Electric Reliability Corporation warns of the need to expand energy generation and transmission. By 2025, the Midwest might experience power shortages, with demand rising significantly. The increase in electric vehicles and heat pump usage also contributes to this demand. Tech companies like Microsoft, Amazon, and Meta plan to generate energy for their data centers, with nuclear power playing a significant role. Plans include refurbishing old nuclear plants and building new ones.
Meta and Stanford University researchers are developing Apollo models for better video understanding by AI. Despite advances in language and image processing, video models lag due to higher computational demands. The research found that design choices effective for smaller models work for larger ones, allowing efficient experimentation. Key strategies include integrating timestamps and balanced training data. The Apollo models outperform similar-sized models, with Apollo-3B and Apollo-7B leading. Meta has made the models open source.
Google has rolled out the Gemini 2.0 model in its chatbot. Users with access to Gemini Advanced can use the experimental model “1206” on desktop and mobile. It may be the next major Gemini “Pro” model. The model assists with complex tasks like coding and problem-solving but is in an early preview phase and lacks real-time information access.
The AI start-up Perplexity has acquired Carbon, a company developing connectors for external data sources. This acquisition allows users to connect apps like Notion and Google Docs directly with Perplexity. The entire Carbon team has joined Perplexity to accelerate development. This move highlights a trend where major AI platforms and software solutions integrate into a unified chat interface with internet access.
Microsoft has purchased 485,000 Nvidia Hopper GPUs, becoming Nvidia’s largest customer for AI accelerators in 2024. This investment of $15 billion surpasses Chinese companies Bytedance and Tencent. Tech giants increasingly rely on their AI accelerators, with Google and Meta operating 1.5 million custom chips each, Amazon 1.3 million, and Microsoft 200,000 of its “Maia” accelerators.
The European Data Protection Board has set guidelines for AI model development and use under the General Data Protection Regulation (GDPR). Companies like Meta, Google, and OpenAI can use “legitimate interest” as a legal basis for processing personal data with AI models, subject to conditions. National data protection authorities will assess this interest through a three-step test. If personal data is processed unlawfully, the AI model’s use could be banned unless properly anonymized. The aim is to ensure consistent legal enforcement across the EU.