On-policy Reinforcement Learning (RL) with PPO-like clip objectives has become the standard choice for reward-based fine-tuning of large language models (LLMs). Although recent work has explored improved estimators of advantages and normalization, the clipping mechanism itself has remained untouched. Originally introduced as a proxy for principled KL-based trust regions, clipping is a crude approximation that often causes unstable updates and suboptimal performance. We replace the clip objective with a novel discrete differentiable trust region projection, which provides principled token-level KL constraints. The projection operates on a sparse subset of the model's most important token logits to balance computational cost and projection effectiveness. Our approach, Trust Region Optimization for Large Language Models (TROLL), serves as a direct replacement for PPO-like clipping during training and does not alter the model's inference behavior. Across datasets, model families, and advantage-estimation methods, TROLL consistently outperforms PPO-like clipping in terms of training speed, stability, and final success rates.

Is a Hiking Game That Trolls'Slightly Problematic' Men The walking simulator, launching September 23 on PlayStation and Steam, stars a jobless 35-year-old "privileged, white male" whose pride stops him from getting help. Game developer Bennett Foddy was watching a Greek myth unfold in front of him. A playtester for his latest project,, was struggling to navigate the game's lead--Nate, a 35-year-old "failson" in a stained onesie--up a slippery hill. Each time, the terrain proved to be too much, and Nate skidded uselessly down it. Foddy has a reputation for making onerous games that take a little bit of masochism to master.

Trolling in online communities typically involves disruptive behaviors such as provoking anger and manipulating discussions, leading to a polarized atmosphere and emotional distress. Robust moderation is essential for mitigating these negative impacts and maintaining a healthy and constructive community atmosphere. However, effectively addressing trolls is difficult because their behaviors vary widely and require different response strategies (RSs) to counter them. This diversity makes it challenging to choose an appropriate RS for each specific situation. To address this challenge, our research investigates whether humans have preferred strategies tailored to different types of trolling behaviors. Our findings reveal a correlation between the types of trolling encountered and the preferred RS. In this paper, we introduce a methodology for generating counter-responses to trolls by recommending appropriate RSs, supported by a dataset aligning these strategies with human preferences across various troll contexts. The experimental results demonstrate that our proposed approach guides constructive discussion and reduces the negative effects of trolls, thereby enhancing the online community environment.

The first time I used SocialAI, I was sure the app was performance art. That was the only logical explanation for why I would willingly sign up to have AI bots named Blaze Fury and Trollington Nefarious, well, troll me. Even the app's creator, Michael Sayman, admits that the premise of SocialAI may confuse people. His announcement this week of the app read a little like a generative AI joke: "A private social network where you receive millions of AI-generated comments offering feedback, advice, and reflections." But, no, SocialAI is real, if "real" applies to an online universe in which every single person you interact with is a bot.

Russian Internet Trolls use fake personas to spread disinformation through multiple social media streams. Given the increased frequency of this threat across social media platforms, understanding those operations is paramount in combating their influence. Using Twitter content identified as part of the Russian influence network, we created a predictive model to map the network operations. We classify accounts type based on their authenticity function for a sub-sample of accounts by introducing logical categories and training a predictive model to identify similar behavior patterns across the network. Our model attains 88% prediction accuracy for the test set. Validation is done by comparing the similarities with the 3 million Russian troll tweets dataset. The result indicates a 90.7% similarity between the two datasets. Furthermore, we compare our model predictions on a Russian tweets dataset, and the results state that there is 90.5% correspondence between the predictions and the actual categories. The prediction and validation results suggest that our predictive model can assist with mapping the actors in such networks.

A topical challenge for algorithms in general and for automatic image categorization and generation in particular is presented in the form of a drawing for AI to understand. In a second vein, AI is challenged to produce something similar from verbal description. The aim of the paper is to highlight strengths and deficiencies of current Artificial Intelligence approaches while coarsely sketching a way forward. A general lack of encompassing symbol-embedding and (not only) -grounding in some bodily basis is made responsible for current deficiencies. A concomitant dearth of hierarchical organization of concepts follows suite. As a remedy for these shortcomings, it is proposed to take a wide step back and to newly incorporate aspects of cybernetics and analog control processes. It is claimed that a promising overarching perspective is provided by the Ouroboros Model with a valid and versatile algorithmic backbone for general cognition at all accessible levels of abstraction and capabilities. Reality, rules, truth, and Free Will are all useful abstractions according to the Ouroboros Model. Logic deduction as well as intuitive guesses are claimed as produced on the basis of one compartmentalized memory for schemata and a pattern-matching, i.e., monitoring process termed consumption analysis. The latter directs attention on short (attention proper) and also on long times scales (emotional biases). In this cybernetic approach, discrepancies between expectations and actual activations (e.g., sensory precepts) drive the general process of cognition and at the same time steer the storage of new and adapted memory entries. Dedicated structures in the human brain work in concert according to this scheme.

For generative AI to succeed, how engaging a conversationalist must it be? For almost sixty years, some conversational agents have responded to any question or comment to keep a conversation going. In recent years, several utilized machine learning or sophisticated language processing, such as Tay, Xiaoice, Zo, Hugging Face, Kuki, and Replika. Unlike generative AI, they focused on engagement, not expertise. Millions of people were motivated to engage with them. What were the attractions? Will generative AI do better if it is equally engaging, or should it be less engaging? Prior to the emergence of generative AI, we conducted a large-scale quantitative and qualitative analysis to learn what motivated millions of people to engage with one such 'virtual companion,' Microsoft's Zo. We examined the complete chat logs of 2000 anonymized people. We identified over a dozen motivations that people had for interacting with this software. Designers learned different ways to increase engagement. Generative conversational AI does not yet have a clear revenue model to address its high cost. It might benefit from being more engaging, even as it supports productivity and creativity. Our study and analysis point to opportunities and challenges.

In the dynamic and rapidly evolving world of social media, detecting anomalous users has become a crucial task to address malicious activities such as misinformation and cyberbullying. As the increasing number of anomalous users improves the ability to mimic normal users and evade detection, existing methods only focusing on bot detection are ineffective in terms of capturing subtle distinctions between users. To address these challenges, we proposed SeGA, preference-aware self-contrastive learning for anomalous user detection, which leverages heterogeneous entities and their relations in the Twittersphere to detect anomalous users with different malicious strategies. SeGA utilizes the knowledge of large language models to summarize user preferences via posts. In addition, integrating user preferences with prompts as pseudo-labels for preference-aware self-contrastive learning enables the model to learn multifaceted aspects for describing the behaviors of users. Extensive experiments on the proposed TwBNT benchmark demonstrate that SeGA significantly outperforms the state-of-the-art methods (+3.5\% ~ 27.6\%) and empirically validate the effectiveness of the model design and pre-training strategies. Our code and data are publicly available at https://github.com/ying0409/SeGA.

The detection of state-sponsored trolls operating in influence campaigns on social media is a critical and unsolved challenge for the research community, which has significant implications beyond the online realm. To address this challenge, we propose a new AI-based solution that identifies troll accounts solely through behavioral cues associated with their sequences of sharing activity, encompassing both their actions and the feedback they receive from others. Our approach does not incorporate any textual content shared and consists of two steps: First, we leverage an LSTM-based classifier to determine whether account sequences belong to a state-sponsored troll or an organic, legitimate user. Second, we employ the classified sequences to calculate a metric named the "Troll Score", quantifying the degree to which an account exhibits troll-like behavior. To assess the effectiveness of our method, we examine its performance in the context of the 2016 Russian interference campaign during the U.S. Presidential election. Our experiments yield compelling results, demonstrating that our approach can identify account sequences with an AUC close to 99% and accurately differentiate between Russian trolls and organic users with an AUC of 91%. Notably, our behavioral-based approach holds a significant advantage in the ever-evolving landscape, where textual and linguistic properties can be easily mimicked by Large Language Models (LLMs): In contrast to existing language-based techniques, it relies on more challenging-to-replicate behavioral cues, ensuring greater resilience in identifying influence campaigns, especially given the potential increase in the usage of LLMs for generating inauthentic content. Finally, we assessed the generalizability of our solution to various entities driving different information operations and found promising results that will guide future research.

Jacob Browning is a postdoc in NYU's Computer Science Department working on the philosophy of AI. Yann LeCun is a Turing Award-winning machine learning researcher, an NYU professor and the chief AI scientist at Meta. With artificial intelligence now powering Microsoft's Bing and Google's Bard search engines, brilliant and clever conversational AI is at our fingertips. But there have been many uncanny moments -- including casually delivered disturbing comments like calling a reporter ugly, declaring love for strangers or rattling off plans for taking over the world. To make sense of these bizarre moments, it's helpful to start by thinking about the phenomenon of saying the wrong thing.