🤖LLMs Transform Robot Movement!

MIT researchers use large language models to optimize robot navigation

Key Points 

• Instead of relying on intricate visual models, their method converts visual information into textual descriptions from the robot's perspective

• This innovative approach simplifies the training process and enhances the robot's ability to understand and execute complex tasks more efficiently.

🦾Context of the news - MIT researchers have developed a new method for robots to navigate complex tasks based on user instructions. Traditionally, achieving tasks like having a robot transport laundry to a specific spot in a home involves multiple machine-learning models and vast amounts of visual data. This approach is not only labor-intensive but also computationally demanding.

🤖News - To address these challenges, MIT and MIT-IBM Watson AI Lab researchers devised a novel navigation strategy. Instead of relying on intricate visual models, their method converts visual information into textual descriptions from the robot's perspective. These descriptions are then processed by a large language model to determine the sequence of actions required to fulfill user commands. 

This innovative approach simplifies the training process and enhances the robot's ability to understand and execute complex tasks more efficiently.

😀Key advantages - Their approach, although not surpassing vision-based techniques in performance, provides several advantages: 

1. Textual data requires fewer computational resources compared to complex visual data, enabling rapid generation of synthetic training data. For instance, they generated 10,000 synthetic trajectories from just 10 real-world visual trajectories.

2. It helps bridge the gap between simulated and real-world environments, where computer-generated images may differ significantly due to factors like lighting and color. Language descriptions are more consistent across these environments.

3. The model's representations in natural language are more intuitive for humans to understand compared to visual representations.

4. It offers flexibility across different tasks and environments with minimal adaptation, as it relies on a single type of input—language data—regardless of the task or setting.

• However, one drawback is that the method inherently lacks some detailed information captured by vision-based models, such as depth perception.

OpenAI ramps up lobbying efforts to influence AI regulations

Key Points 

• The San Francisco-based start-up has grown its global affairs team from three members at the start of 2023 to 35, with plans to increase to 50 by the end of 2024. 

• This move comes as governments consider AI safety laws that could limit the company's growth and the development of its innovative models, such as ChatGPT.

☕News - OpenAI is assembling an international team of lobbyists to influence politicians and regulators who are increasingly scrutinizing advanced AI. The San Francisco-based start-up has grown its global affairs team from three members at the start of 2023 to 35, with plans to increase to 50 by the end of 2024. 

This move comes as governments consider AI safety laws that could limit the company's growth and the development of its innovative models, such as ChatGPT.

🥸OpenAI's objective - Anna Makanju, OpenAI's vice-president of government affairs, explained that the company's approach to lobbying isn't about stopping regulations to maximize profits. Instead, their goal is to ensure that artificial general intelligence (AGI) benefits all of humanity.

According to Makanju, OpenAI is addressing lingering issues from the social media era, which have fostered a general distrust of Silicon Valley companies.  She noted that this skepticism often spills over into perceptions of AI, adding that it is therefore important to clarify that AI technology is fundamentally different and requires regulatory approaches that recognize its unique characteristics.

🫠In this context - OpenAI participated in discussions regarding the EU's AI Act. The company, along with others, argued that some of its models should not be classified as "high risk" in early drafts to avoid stricter regulations. However, its most advanced models will still be subject to the act's rules.

Chinese universities crack down on AI use in student assignments as graduation season nears

Key Points 

• According to a November 2023 survey conducted by China Youth Daily, nearly 85% of over 7,000 respondents reported using AI tools.

👨🏻‍🎓News - Several universities in China are implementing guidelines regarding the use of artificial intelligence in student assignments, particularly as graduation season approaches. 

This is in response to worries about maintaining academic honesty, fueled by a survey revealing widespread adoption of AI tools by college students.

• According to a November 2023 survey conducted by China Youth Daily, nearly 85% of over 7,000 respondents reported using AI tools.

• Additionally, the lifestyle app Xiaohongshu has seen over 10,000 posts sharing strategies to evade AI detection.

🏫Universities take action -

• In April, Fuzhou University announced that all undergraduate theses submitted by graduating students this year will undergo checks for AI-generated content both before the final draft and after the paper is reviewed. The results of these checks will be considered during thesis evaluation and in the selection of outstanding undergraduate theses.

• Similarly, Tianjin University of Science and Technology introduced measures to check undergraduate theses for AI-generated content. According to China Daily, if AI-generated content exceeds 40%, students are required to revise their theses.

• Hubei University has also implemented comparable measures. If a student's thesis is deemed to pose a "high risk of AI-generated content," teachers will advise revisions.

📖What's more? In August of the previous year, state broadcaster CCTV revealed that a proposed law on academic degrees submitted to the Standing Committee of the National People's Congress included a prohibition on using AI to write dissertations. However, when the law was adopted on April 26, this specific ban on AI was not included. Instead, the law specifies that academic degrees may be withheld or rescinded if the dissertation involves academic misconduct such as ghostwriting, plagiarism, or falsification. According to the state news agency Xinhua, the law is scheduled to come into effect on January 1, 2025.