arXiv subjects

Jinsuk Park

Publications and source records attributed to Jinsuk Park.

3 recordsLinked to original sources

arXiv2025

Impact of Volatility on Time-Based Transaction Ordering Policies

We study Arbitrum's Express Lane Auction (ELA), an ahead-of-time second-price auction that grants the winner an exclusive latency advantage for one minute. Building on a single-round model with risk-averse bidders, we propose a hypothesis that the value of priority access is discounted relative to risk-neutral valuation due to the difficulty of forecasting short-horizon volatility and bidders' risk aversion. We test these predictions using ELA bid records matched to high-frequency ETH prices and find that the result is consistent with the model.

cs.GT↗

arXiv2025

TimeBoost: Do Ahead-of-Time Auctions Work?

We study the performance of the TimeBoost auction, by comparing cumulative fixed time markout of fast lane trades over the TimeBoost interval to bids for the fast lane. Such comparison allows us to assess how well bids predict future extracted value from the time advantage. The correlation between winning bids and markouts is weak across bidders, suggesting that bids are a noisy predictor of extracted value. The correlation slightly improves when comparing paid bids (the second highest bid) instead of winning bids to markouts, which we attribute to the fact that the auction is more of a common value type. In all settings, the relative order of the most frequent bidder performance remains the same, together with their absolute profits. Bids and markouts aggregated over long time intervals exhibit much higher correlation, indicating that bidders detect trends much better than identify when the high arbitrage value is exactly available. One possible explanation for this is the fact that the correlation between previous minute markouts and current minute bids is significant, suggesting that the previous minute markouts is used to predict the next minute value when bidding.

cs.GT↗

arXiv2025

DarkBench: Benchmarking Dark Patterns in Large Language Models

We introduce DarkBench, a comprehensive benchmark for detecting dark design patterns--manipulative techniques that influence user behavior--in interactions with large language models (LLMs). Our benchmark comprises 660 prompts across six categories: brand bias, user retention, sycophancy, anthropomorphism, harmful generation, and sneaking. We evaluate models from five leading companies (OpenAI, Anthropic, Meta, Mistral, Google) and find that some LLMs are explicitly designed to favor their developers' products and exhibit untruthful communication, among other manipulative behaviors. Companies developing LLMs should recognize and mitigate the impact of dark design patterns to promote more ethical AI.

cs.CL↗