Large language models have a range of beneficial uses: they can assist in prose, poetry, and programming; analyze dataset biases; more. However, their flexibility generative capabilities also raise misuse concerns. This report discusses OpenAI's work related to the release its GPT-2 model. It staged release, which allows time between model releases conduct risk benefit analyses as sizes increased. ongoing partnership-based research provides recommendations for better coordination responsible...

Evaluating the quality of a dialogue interaction between two agents is difficult task, especially in open-domain chit-chat style dialogue. There have been recent efforts to develop automatic evaluation metrics, but most them do not generalize unseen datasets and/or need human-generated reference response during inference, making it infeasible for online evaluation. Here, we propose an unreferenced automated metric that uses large pre-trained language models extract latent representations...

ABSTRACT We study environmental and social (E&S) disclosures in annual reports. Using the word embedding model to examine over 210,000 reports from 24,271 public firms 30 international countries/regions between 2001 2020, we create an E&S dictionary that allows us document trends report disclosure. Specifically, find: (1) increases length boilerplate language (2) decreases specificity. Our results also suggest disclosure quality improves after adoption of voluntary ESG reporting...

ABSTRACT We study firms’ decisions to provide non-GAAP income statements and the information environment consequences of Securities Exchange Commission (SEC) comment letters directing them stop disclosing such statements. find firms voluntarily disclose when firm disclosure complexity, analyst following, institutional ownership are higher. Using a difference-in-differences design, we that, after full at direction SEC, informativeness earnings overall announcements decreases, asymmetry...

ABSTRACT We examine how managers’ incentives to minimize litigation risk interact with the unique features of environmental information shape disclosure decisions. rely on peer firms’ lawsuits generate variation in risk, consistent prior research and anecdotal evidence suggesting that firms perceive an increase after a firm is sued for related disclosures. Although we provide mixed around changes total response lawsuits, offer robust more forward-looking (and less historical) disclosures...

We study environmental and social (E&S) disclosures in annual reports. Using the word embedding model to examine more than 210,000 reports from 24,271 public firms 30 international countries/regions between 2001 2020, we create an E&S dictionary that allows us document trends report disclosure. Specifically, find: 1) increases length, boilerplate language, stickiness disclosure, use of infographics, 2) decreases specificity. Our results also suggest disclosure quality improves after adoption...

We study firms’ decisions to provide non-GAAP income statements and the information environment consequences of SEC comment letters directing them stop disclosing such statements. find firms voluntarily disclose when analyst following institutional ownership are higher firm disclosure complexity is higher. Using a difference-in-differences design, we that, after full at direction SEC, informativeness earnings overall announcements decreases, asymmetry increases, forecasts become less...

This study examines the role of Securities and Exchange Commission (SEC) in mergers acquisitions (M&As) involving publicly traded target firms. We find that deals receiving comment letters have an increased likelihood deal completion price revision, consistent with SEC review process reducing information asymmetry, albeit at cost delaying M&A process. Further analyses suggest generates new value-relevant via firms’ disclosure amendments response to letters. address endogeneity concerns using...

We investigate the role of litigation risk in environmental disclosure decisions. find that after a peer firm is sued for its disclosures, firms provide more forward-looking (and less historical) disclosures their conference calls. Our evidence consistent with managers seeking to minimize being misrepresenting response perceived increase risk. The main result persists both broader sample using Kim and Skinner (2012) measure proxy risk, emissions targets historical do not any respond peers'...

We examine whether auditor style is associated with non-GAAP disclosures. Specifically, we find that clients audited by the same are more likely to disclose earnings in a similar manner. assess disclosure similarity using (1) decision earnings, (2) prominence of and (3) discussion management analysis annual report. association between determined Big 4 accounting firms audit office within an firm. also some evidence clients' choices exclude recurring items quality. Finally, level results...

Abstract Regulators and practitioners have concerns that the lack of standardization in non‐GAAP disclosure can make it challenging for users to process earnings use decision‐making. We examine whether auditor style extends beyond mandatory disclosures induce similarity disclosures. Specifically, we find clients audited by same are more likely disclose a similar manner. assess using (1) decision earnings, (2) prominence press release, (3) discussion management analysis annual report, (4)...

Evaluating the quality of a dialogue interaction between two agents is difficult task, especially in open-domain chit-chat style dialogue. There have been recent efforts to develop automatic evaluation metrics, but most them do not generalize unseen datasets and/or need human-generated reference response during inference, making it infeasible for online evaluation. Here, we propose an unreferenced automated metric that uses large pre-trained language models extract latent representations...

We study the winner’s curse in mergers and acquisitions, which winning bidders fail to account for uncertainty about target value thus overpay. Using a unique setting where firms hire multiple investment banks as advisors, we construct novel measure of valuation based on banks’ disagreement valuation. find that when is higher, pay significantly higher acquisition premiums. Moreover, who high premiums have lower announcement returns long-term post-merger high. These also create merger...