• AI KATANA
  • Posts
  • OpenAI's Ambitious Mission to Align Superintelligence with Human Intent

OpenAI's Ambitious Mission to Align Superintelligence with Human Intent

Also: AI tests into top 1% for original creative thinking

Author

AI KATANA
July 06, 2023

Welcome!

In today's AI update, we spotlight OpenAI's launch of a new team, Superalignment, aiming to ensure that superintelligent AI aligns with human intent, a mission led by Ilya Sutskever and Jan Leike. On a similar note, a study led by Dr. Erik Guzik has found that our GPT-4 powered application, ChatGPT, tested into the top 1% for original creative thinking. Meanwhile, Eric Schmidt, Google's former CEO, anticipates AI's transformative role in science, with companies such as Nvidia and DeepMind leading the change. In economics, generative AI's substantial potential to turbocharge productivity in Asia is making headlines. In our tool and dataset updates, learn about KokoMind, a dataset to evaluate LLMs' social understanding abilities. Finally, despite a general downturn in VC funding globally, AI ventures are still attracting considerable investment. Dive in for more details.

Slicing away:

  • 🤖 OpenAI's Ambitious Mission to Align Superintelligence with Human Intent

  • 1️⃣ AI tests into top 1% for original creative thinking

  • 🧪 Eric Schmidt: This is how AI will transform the way science gets done

  • 🐯 Turbo-charging productivity in Asia: the economic benefits of generative AI

OpenAI is launching a new team, Superalignment, led by Ilya Sutskever and Jan Leike, with the aim of addressing the urgent challenge of superintelligence alignment – ensuring AI systems smarter than humans follow human intent. With 20% of their secured compute resources devoted to this cause over the next four years, they plan to tackle the inadequacies of current alignment techniques which may fail to control AI beyond human-level intelligence. Their approach includes building an automated alignment researcher, scaling efforts using vast computational power, and running tests to detect and rectify misalignments. While this is a highly ambitious goal, the team is optimistic that focused research can yield solutions. OpenAI will continue to share its progress and findings, welcoming contributions from non-OpenAI models, while tackling other AI risks, such as misuse and bias. OpenAI encourages leading machine learning experts to join this effort, even if they haven't worked on alignment before, emphasizing the significance of superintelligence alignment as one of the most crucial technical problems of our time.

In a groundbreaking study led by Dr. Erik Guzik from the University of Montana, the GPT-4 powered application, ChatGPT, demonstrated a level of creativity equivalent to the top 1% of human thinkers in the Torrance Tests of Creative Thinking (TTCT). The AI outperformed a control group of students, scoring highly for fluency and originality in generating ideas, and a bit lower (97th percentile) for flexibility. Despite the machine's achievement, Guzik highlighted the need for better tools to differentiate between human and AI creativity, cautioning against drawing hasty conclusions. The results of the study, suggesting AI's rapidly growing capacity for creativity, potentially indicates a significant shift in the landscape of business and innovation.

AI, spearheaded by advancements from organizations such as Nvidia, Google's DeepMind, and McMaster and MIT, is set to revolutionize the way science is conducted, predicts Eric Schmidt, Google's former CEO. For instance, Nvidia's AI-powered digital twin of the planet, Earth-2, which uses the FourCastNet AI model, can predict weather tens of thousands of times faster and more accurately than current methods. AI tools are also transforming literature reviews, hypothesis formulation, experimental processes, and result interpretation, streamlining the entire scientific process. Examples include identifying antibiotics to combat dangerous pathogens, controlling plasma in nuclear fusion reactions, and improving radiology. Self-driving labs, automated with AI, are emerging to further increase efficiency and accuracy. AI is also expected to democratize science, with tools that reduce barriers to entry, such as LLMs that assist with coding. However, human skills are still essential, and careful regulation is required due to potential risks associated with AI, especially in the wrong hands.

Generative AI, with its capacity to create, organise, and analyse content, is set to make significant social and economic transformations in Asia. By democratising the field, it enables anyone to become a creator while also leading to changes in job processes and needed skills. Research commissioned by Microsoft suggests a boost of productive capacity by US$621 billion in India, US$1.1 trillion in Japan, and US$79.3 billion in the Philippines, with the technology poised to impact nearly all forms of work. The economic growth will primarily stem from generative AI's potential to stimulate creativity, expedite discovery, and amplify efficiency. The technology is projected to be used in varying capacities in different roles, transforming tasks rather than eliminating jobs. However, careful management of the transition is required, with the need for policies, regulations, and an environment that promotes innovation and responsible use. Moreover, efforts to improve digital literacy across the entire population are crucial for individuals to fully leverage the benefits of generative AI.

🛠️ AI tools updates

KokoMind is a dataset of social interactions that was created to evaluate the ability of large language models (LLMs) to understand and navigate social interactions. The dataset contains 150 complex multi-party social interactions, each with free-text questions and answers. The goal of KokoMind is to help researchers develop LLMs that can better understand and respond to social cues, such as emotions, relationships, and intentions.

💵 Venture Capital updates

Global VC funding plunged by almost half in the first six months of 2023, according to data from research firm PitchBook. The decline was driven by a slowdown in the United States, where VC funding fell by 23% year-over-year. However, AI-related deals bucked the trend, with funding for AI startups rising by 15% in the first half of the year. This suggests that AI is still a hot area for VC investment, even as the overall market cools.

🫡 Meme of the day

⭐️ Generative AI image of the day