Daily AI Newsletter: Altman says ChatGPT will have to evolve in “uncomfortable” ways

In today’s email:

🏅 DeepMind approaches gold standard in complex maths in latest AI breakthrough
🇺🇸 Samsung’s latest Galaxy phones offer live translation over phone calls, texts
☠️ AI poisoning could turn open models into destructive “sleeper agents,” says Anthropic
🧰 10 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

Altman says ChatGPT will have to evolve in “uncomfortable” ways

Sam Altman, CEO of OpenAI, discussed the future of AI technology and its implications in an interview with Axios at Davos. He highlighted that AI is evolving rapidly and acknowledged the necessity for "uncomfortable" decisions in its development. Altman believes future AI products should offer extensive individual customization, even though it may lead to discomfort due to varying responses based on users' values and cultural backgrounds.

Altman specifically mentioned that while AI should not support extreme and harmful values, it might cater to cultural differences. He anticipates AI's role in changing knowledge work, accelerating scientific discovery, and impacting global elections. Altman is currently focused on launching what is likely to be called GPT-5 and expressed concern about AI's influence on elections. He mentioned OpenAI's efforts to combat election misinformation.

Regarding internal company dynamics, Altman touched upon the tensions that led to his brief removal as CEO last November and the uncertainty surrounding Ilya Sutskever's return to OpenAI. He emphasized his commitment to OpenAI, dismissing notions that he is distracted by other ventures.

Altman also defended OpenAI's content licensing strategies and its decision to allow military use of its models, emphasizing support for the U.S. government. He advised CEOs to leverage AI for improving internal workflows and shared personal lessons from his experiences in 2023, including the importance of addressing non-urgent problems.

Visual-based newsletter on business and tech

With Bay Area Times, you will get graphics like these, so you quickly understand the news without having to read through long and boring paragraphs.

We explain the latest business, finance, and tech news with visuals and data. 📊

All in one free newsletter that takes < 5 minutes to read. 🗞

Save time and become more informed today.👇

DeepMind approaches gold standard in complex maths in latest AI breakthrough

Google’s AI unit shows the technology’s growing proficiency with AlphaGeometry getting close to matching top students Story illustration

Google DeepMind's recent achievement with its AI system, AlphaGeometry, demonstrated remarkable skills in solving geometry problems. This AI system successfully answered 25 out of 30 questions from the International Mathematical Olympiad, a high school-level competition. This performance is close to the standard set by gold medal-winning human competitors, showcasing AI's growing ability in complex mathematical reasoning.

AlphaGeometry is a neuro-symbolic system, combining language learning with deductive reasoning, inspired by the concept of "Thinking, Fast and Slow" by psychologist Daniel Kahneman. It utilizes both rapid pattern recognition and logical thinking. The system was trained on a dataset of 100 million synthetic geometry examples. Its score of 25 was significantly higher than the previous automated systems, but it still faced challenges with some complex problems, such as one solved by Vietnamese mathematician Lê Bá Khánh Trình in 1979.

The development of AlphaGeometry is seen as a step towards building an Artificial General Intelligence (AGI) that can perform at or above the human level across various fields. While this represents significant progress, the ultimate goal of creating AI systems capable of discovering new mathematical solutions to unsolved problems remains a more challenging and elusive objective. DeepMind, however, has not planned to enter AlphaGeometry in the International Mathematical Olympiad yet, as it continues to advance in the field of mathematics.

Google introduces ‘Circle to Search,’ a new way to search from anywhere on Android using gestures

Google announced a new feature for Android phones called "Circle to Search" alongside Samsung's launch event. This feature allows users to search directly from any place on their phone using gestures like circling, highlighting, scribbling, or tapping, making it easier to use Google Search in various contexts like watching videos or messaging. Although named "Circle to Search," the feature isn't limited to circling gestures; it includes various interactions with text or images on-screen to initiate searches.

For example, while watching a video, users can circle an item to search for it or tap a restaurant's name in a chat to get more details. Scribbling can be used on both images and words for searching. The search results vary depending on the query and whether users have opted into Google Labs products, with AI-powered answers available through Google's Search Generative Experience experiment.

Google aims to make search more accessible without needing to switch apps or take screenshots for later searches. However, this launch also reflects Google's efforts to adapt to challenges like SEO-optimized spam pages and the rise of generative AI chatbots, which are changing how people search online. This feature signifies a deeper integration of Google's Search into the smartphone OS.

"Circle to Search" will debut on January 31 on the new Galaxy S24 Series and premium Android phones like the Pixel 8 and 8 Pro, available in all languages and locations where these phones are sold. Eventually, more Android smartphones will support this feature. The announcement was part of several Google AI updates across various platforms, including a new AI-powered feature for multisearch in Google Lens.