To ensure that large language model (LLM) responses are helpful and non-toxic, a reward model trained on human preference data is usually used. LLM responses with high rewards are then selected through best-of-$n$ (BoN) sampling or the LLM is further optimized to produce responses with high rewards through reinforcement learning from human feedback (RLHF). However, these processes are susceptible to reward overoptimization or `hacking', where responses receive high rewards due to imperfections in the reward model rather than true preference, particularly as prompts or responses deviate from the training data. To address these challenges, we propose to train a Bayesian reward model, which signals higher uncertainty further from the training data distribution. We trained Bayesian reward models using Laplace approximation on LoRA weights, and found that the resulting uncertainty estimates can effectively mitigate reward overoptimization in BoN sampling.

Sam Altman, the CEO of artificial intelligence lab OpenAI, told a Senate panel he welcomes federal regulation on the technology "to mitigate" its risks. The CEO of the company that delivered ChatGPT to the world said Tuesday he was not interested in becoming the federal government's top regulator of artificial intelligence technology. CEO Sam Altman and other witnesses at a Senate Judiciary subcommittee were asked what they would do to ensure the government has a firm grip on how AI is developed and deployed, and Altman said his first step would be to create a new federal agency. "I would form a new agency that licenses any effort above a certain scale of capabilities and can take that license away and ensure compliance with safety standards," he said in response to a question from Sen. John Kennedy, R-La. Sam Altman, CEO and co-founder of OpenAI, speaks during a Senate Judiciary Subcommittee hearing in Washington, D.C., Tuesday, May 16, 2023.

Key trend #1 -- Migration to the cloud will accelerate, although the rising scale of cloud costs will become increasingly scrutinized by finance types as the economy tightens. Organizations will struggle with hybrid and multi-cloud architectures. Watch for new and innovative pricing schema to help combat the rising costs (e.g., fee caps and fixed fee). The flexibility and scalability the cloud offers to organizations. Cloud computing allows companies to easily add or remove resources as needed, without having to invest in expensive hardware and infrastructure.

Join us on November 9 to learn how to successfully innovate and achieve efficiency by upskilling and scaling citizen developers at the Low-Code/No-Code Summit. Traditional contract lifecycle management (CLM) tools focus on improving document workflows. However, Icertis seeks to take the field to the next level with contract intelligence that uses artificial intelligence (AI) and machine learning (ML) to automatically extract contract data at scale. These tools are designed to structure contracts' commercial, legal and operational data and connect that data to procurement, ERP and human capital management apps to help companies accelerate revenue, reduce costs, improve risk management and ensure compliance. When VentureBeat previously covered Icertis in 2019, the CLM market was expected to be worth $3.16 billion by 2023.

It doesn't generate revenue, but it is an essential part of operating effectively as a business today. Whether it's industry specific regulations, or the standout regulation of our time--GDPR--we are all acutely aware of the damage, both reputational and financial, that non-compliance can cause. GDPR has equipped employees across industries with an appreciation of the context, usage, and security of data, but there is another factor that is essential for establishing an effective data strategy, which is data discoverability. To ensure regulatory compliance, data must not only be secure, it must also be discoverable so that compliance personnel can locate all information needed to prove compliance. Increasingly, AI tools are being harnessed to automate workflows and governance, but such capabilities can only be delivered when a strong data foundation is in place.

It's time for employers to start preparing for legislation recently signed into law in Illinois, the Artificial Intelligence Video Interview Act. The new law, which takes effect on January 1, 2020, regulates Illinois employers' use of artificial intelligence (AI) in the interview and hiring process. Under the AI Video Interview Act, employers that record video interviews and use AI technology to analyze applicants' suitability for employment must: Employers that conduct such interviews may not distribute videos to other parties, except as necessary to obtain expert assistance in evaluating a candidate's fitness for a particular position. In addition, an employer has only 30 days to destroy all video copies of the interview if an applicant seeks such destruction. This law highlights a myriad of privacy concerns for employers evaluating the costs and benefits of incorporating AI technology into their hiring practices.

Most people have heard of the headline-making achievements in artificial intelligence (AI); systems winning quiz shows and beating world champions in chess. These are the poster children of the discipline but there is a quieter revolution taking in shape in other areas, including regulatory compliance in financial services. Writing for Banking Technology, Mike MacDonagh, London-based director of enterprise risk management at Wolters Kluwer, examines how AI technologies are promising to transform the way that firms ensure they can comply with a global explosion of new regulation. The problem with regulation Looked at in isolation, a piece of regulation is a relatively simple affair – a legal document containing text that describes what needs to be done, by whom, when, and (sometimes) how. With some understanding of the underlying topic, a compliance officer can read the document; understand what is mandated and where it will affect his or her part of the organisation.