From Stanford TRANSIC: Sim-to-Real Policy Transfer by Learning from Online Correction.
From Stanford.
TRANSIC: Sim-to-Real Policy Transfer by Learning from Online Correction.
Continue reading “TRANSIC: Sim-to-Real Policy Transfer by Learning from Online Correction” »
NASA’s Office of Technology, Policy, and Strategy (OTPS) is asking members of the lunar community to respond to a new Lunar Non-Interference Questionnaire that will inform the development of a framework for further work on non-interference of lunar activities. There is no funding or solicitation expected to follow.
OTPS was created in November 2021 within the Office of the NASA Administrator to work transparently in collaboration across NASA and with the broader space community to provide NASA leadership with a trade space of data-and evidence-driven options to develop and shape NASA policy, strategy, and technology.
As dozens of countries and private sector companies have expressed interest in establishing lunar operations by the end of the decade, including many in the South Pole region, it will be critical to determine how to minimize interference and contamination in lunar activities. Deconfliction has been identified as an area of further work in Section 11 of the Artemis Accords and will be an area of increasing importance as the number of commercial and international actors operating on the lunar surface grows.
Working To Reduce Global Catastrophic Biological Risks — Dr. Jaime Yassif, Ph.D. — VP, Global Biological Policy and Programs, Nuclear Threat Initiative.
Dr. Jaime Yassif, Ph.D. serves as Vice President of Global Biological Policy and Programs, at the Nuclear Threat Initiative (https://www.nti.org/about/people/jaim…) where she oversees work to reduce global catastrophic biological risks, strengthen biosecurity and pandemic preparedness, and drives progress in advancing global health security.
Jason Matheny is a delight to speak with, provided you’re up for a lengthy conversation about potential technological and biomedical catastrophe.
Now CEO and president of Rand Corporation, Matheny has built a career out of thinking about such gloomy scenarios. An economist by training with a focus on public health, he dived into the worlds of pharmaceutical development and cultivated meat before turning his attention to national security.
As director of Intelligence Advanced Research Projects Activity, the US intelligence community’s research agency, he pushed for more attention to the dangers of biological weapons and badly designed artificial intelligence. In 2021, Matheny was tapped to be President Biden’s senior adviser on technology and national security issues. And then, in July of last year, he became CEO and president of Rand, the oldest nonprofit think tank in the US, which has shaped government policy on nuclear strategy, the Vietnam War, and the development of the internet.
From Cornell, Princeton, & Microsoft.
Dataset Reset Policy Optimization for RLHF https://huggingface.co/papers/2404.
Reinforcement Learning (RL) from Human Preference-based feedback is a popular paradigm for fine-tuning generative models, which has produced impressive models such as GPT-4 and…
Continue reading “Dataset Reset Policy Optimization for RLHF” »
Fascinating vision/plan by the one and only Sam Altman of how to update our economic systems to benefit everyone in the context of rapidly accelerating technological change.
My work at OpenAI reminds me every day about the magnitude of the socioeconomic change that is coming sooner than most people believe. Software that can think and learn will do more and more of the work that people now do. Even more power will shift from labor to capital. If public policy doesn’t adapt accordingly, most people will end up worse off than they are today.
We need to design a system that embraces this technological future and taxes the assets that will make up most of the value in that world–companies and land–in order to fairly distribute some of the coming wealth. Doing so can make the society of the future much less divisive and enable everyone to participate in its gains.
Language models often need more exposure to fruitful mistakes during training, hindering their ability to anticipate consequences beyond the next token. LMs must improve their capacity for complex decision-making, planning, and reasoning. Transformer-based models struggle with planning due to error snowballing and difficulty in lookahead tasks. While some efforts have integrated symbolic search algorithms to address these issues, they merely supplement language models during inference. Yet, enabling language models to search for training could facilitate self-improvement, fostering more adaptable strategies to tackle challenges like error compounding and look-ahead tasks.
Researchers from Stanford University, MIT, and Harvey Mudd have devised a method to teach language models how to search and backtrack by representing the search process as a serialized string, Stream of Search (SoS). They proposed a unified language for search, demonstrated through the game of Countdown. Pretraining a transformer-based language model on streams of search increased accuracy by 25%, while further finetuning with policy improvement methods led to solving 36% of previously unsolved problems. This showcases that language models can learn to solve problems via search, self-improve, and discover new strategies autonomously.
Recent studies integrate language models into search and planning systems, employing them to generate and assess potential actions or states. These methods utilize symbolic search algorithms like BFS or DFS for exploration strategy. However, LMs primarily serve for inference, needing improved reasoning ability. Conversely, in-context demonstrations illustrate search procedures using language, enabling the LM to conduct tree searches accordingly. Yet, these methods are limited by the demonstrated procedures. Process supervision involves training an external verifier model to provide detailed feedback for LM training, outperforming outcome supervision but requiring extensive labeled data.
The use of Artificial Intelligence (AI) in education has seen an increase in recent years. The rapid development of this new technology is having a major impact on education. In this edition of Edtech Mondays we will talk about the applications and benefits of AI in the education sector, the necessary implementation frameworks, the policy support and also seek to hear from the end users on the impact AI has or would have.
Tesla executive Rohan Patel clarified some facts about Supercharger NACS access for non-Tesla vehicles like Rivian and Ford.
Patel—Tesla’s Vice President of Public Policy and Business Development—recently replied to a question from Teslavangelist, who questioned the number of Supercharger stalls non-Tesla owners actually had access to with NACS connectors.
Tesla recently opened the Supercharger Network to Ford and Rivian electric vehicles (EVs) through its NACS connecter. Both automakers claim that NACS connectors provide Ford and Rivian owners access to over 15,000 Tesla Supercharger locations. Teslavangelist pointed out that non-Tesla EV owners only have access to V3 and V4 Superchargers, doubting they have access to 15,000 Supercharger stalls.
