[AI Hanip News] After the Coup... The AGI Express Train Departs

OpenAI Incident Resolved in 5 Days
Focus on AGI Catalyst 'Q*' Discovery
Google, Meta, Anthropic Accelerate AI Development

Editor's NoteSince the emergence of ChatGPT, the world has been buzzing with artificial intelligence (AI). Technology is changing daily, and news is pouring in nonstop. Even top industry experts are struggling to keep up. It's overwhelming to consume everything, but you can't afford to miss out either. We bring you the core highlights from the weekly AI issues, carefully selected for easy consumption.

The Identity of Q* That Turned OpenAI Upside Down

OpenAI has had a dramatic week. CEO Sam Altman was abruptly dismissed and then reinstated after five days, with twists and turns along the way. As the situation settled, a noteworthy letter surfaced. It claimed the discovery of an AI named Q* (Q-star). This powerful AI, Q*, which could pose a threat to humanity, was discovered and a warning message about it was sent to the board?this reportedly played a decisive role in Altman’s dismissal. Q* is known to be capable of solving elementary school-level math problems. It may seem trivial, but experts use math problem-solving ability as a benchmark for AI capability, so this is significant. When asked logical questions, AI often provides multiple answers because it responds based on vast amounts of human-generated text, ranking answers by likelihood of correctness. In contrast, solving math problems with a single correct answer means the AI has developed reasoning skills comparable to humans. This is why Q* is considered a potential breakthrough in research on Artificial General Intelligence (AGI).

Bard Finds Information Inside YouTube Videos

[Image source=Yonhap News]

Google’s AI chatbot Bard has become smarter. It can now acquire information from YouTube videos. Beyond understanding text and images, it can analyze videos. For example, if you ask it to "summarize the recipe" from a video about making carrot cake, it will respond. It can also provide detailed information like "how many eggs were needed to make the cake." If you’re watching travel content and ask, "Where is this place shown now?" it can answer that too.

Anthropic, OpenAI’s Rival, Accelerates AI Upgrades

Anthropic CI [Image=Anthropic]

Anthropic, a competitor to OpenAI, is speeding up AI chatbot upgrades. They announced that the paid service "Pro" of their latest version Claude 2.1 can process up to 200,000 tokens at once. This means it can understand more than 500 pages of material in a single input. It can digest the length of a thick novel all at once. Compared to previous versions of Claude, the amount it can handle has nearly doubled. Compared to OpenAI’s GPT-4, it is six times larger. While enhancing information processing capacity, its functionality has been made closer to ChatGPT. Users can connect API tools to Claude depending on the task or situation to utilize it effectively.

Meta Disbands Responsible AI Team

Meta has reportedly disbanded its Responsible AI team. This organization was created in 2019 to understand AI technologies under development at Meta and to prevent related harms. Most members of the disbanded team have moved to the Generative AI team or AI infrastructure organizations. As the generative AI boom started with ChatGPT, Meta formed a Generative AI team in February and hired many personnel. Meanwhile, rumors have circulated that the Responsible AI team had been effectively a "shell" since early this year. The team reportedly had very little autonomy and its opinions were rarely reflected.