Reddit is being touted as an antidote to AI-generated content. Reddit is being touted as an antidote to AI-generated content. Platform is now Britain's fourth most visited social media site as users seek out human-generated content Reddit, the online discussion platform, has overtaken TikTok as Britain's fourth most visited social media service, as search algorithms and gen Z have dramatically transformed its prominence. The platform has undergone huge growth over the last two years, with an 88% increase in the proportion of UK internet users it reaches. Three in five Brits online now encounter the site, up from a third in 2023, according to Ofcom .

Continual learning (CL) empowers pre-trained vision-language (VL) models to efficiently adapt to a sequence of downstream tasks. However, these models often encounter challenges in retaining previously acquired skills due to parameter shifts and limited access to historical data. In response, recent efforts focus on devising specific frameworks and various replay strategies, striving for a typical learning-forgetting trade-off. Surprisingly, both our empirical research and theoretical analysis demonstrate that the stability of the model in consecutive zero-shot predictions serves as a reliable indicator of its anti-forgetting capabilities for previously learned tasks. Motivated by these insights, we develop a novel replay-free CL method named ZAF (Zero-shot Antidote to Forgetting), which preserves acquired knowledge through a zero-shot stability regularization applied to wild data in a plug-and-play manner. To enhance efficiency in adapting to new tasks and seamlessly access historical models, we introduce a parameter-efficient EMA-LoRA neural architecture based on the Exponential Moving Average (EMA).

The release of open-weight large language models (LLMs) creates a tension between advancing accessible research and preventing misuse, such as malicious fine-tuning to elicit harmful content. Current safety measures struggle to preserve the general capabilities of the LLM while resisting a determined adversary with full access to the model's weights and architecture, who can use full-parameter fine-tuning to erase existing safeguards. To address this, we introduce AntiDote, a bi-level optimization procedure for training LLMs to be resistant to such tampering. AntiDote involves an auxiliary adversary hypernetwork that learns to generate malicious Low-Rank Adaptation (LoRA) weights conditioned on the defender model's internal activations. The defender LLM is then trained with an objective to nullify the effect of these adversarial weight additions, forcing it to maintain its safety alignment. We validate this approach against a diverse suite of 52 red-teaming attacks, including jailbreak prompting, latent space manipulation, and direct weight-space attacks. AntiDote is upto 27.4\% more robust against adversarial attacks compared to both tamper-resistance and unlearning baselines. Crucially, this robustness is achieved with a minimal trade-off in utility, incurring a performance degradation of upto less than 0.5\% across capability benchmarks including MMLU, HellaSwag, and GSM8K. Our work offers a practical and compute efficient methodology for building open-weight models where safety is a more integral and resilient property.

We introduce ANTIDOTE, a new class of objectives for learning under noisy labels which are defined in terms of a relaxation over an information-divergence neighborhood. Using convex duality, we provide a reformulation as an adversarial training method that has similar computational cost to training with standard cross-entropy loss. We show that our approach adaptively reduces the influence of the samples with noisy labels during learning, exhibiting a behavior that is analogous to forgetting those samples. ANTIDOTE is effective in practical environments where label noise is inherent in the training data or where an adversary can alter the training labels. Extensive empirical evaluations on different levels of symmetric, asymmetric, human annotation, and real-world label noise show that ANTIDOTE outperforms leading comparable losses in the field and enjoys a time complexity that is very close to that of the standard cross entropy loss.

Trauma prevalence is vast globally. Evidence-based digital treatments can help, but most require human guidance. Human guides provide tailored instructions and responsiveness to internal cognitive states, but limit scalability. Can generative AI and neurotechnology provide a scalable alternative? Here we test ANTIDOTE, combining AI guidance and pupillometry to automatically deliver and monitor an evidence-based digital treatment, specifically the Imagery Competing Task Intervention (ICTI), to reduce intrusive memories after psychological trauma. One hundred healthy volunteers were exposed to videos of traumatic events and randomly assigned to an intervention or active control condition. As predicted, intervention participants reported significantly fewer intrusive memories over the following week. Post-hoc assessment against clinical rubrics confirmed the AI guide delivered the intervention successfully. Additionally, pupil size tracked intervention engagement and predicted symptom reduction, providing a candidate biomarker of intervention effectiveness. These findings open a path toward rigorous AI-guided digital interventions that can scale to trauma prevalence.

In a world and industry which is becoming more commercialised, over sanitised, robotic, structured and predictable, Tyler's greatest strength is the opposite to all of that." That's quite the sell for Southampton's 19-year-old midfield star Tyler Dibling, especially given his basic Premier League career numbers amount to 25 appearances, 1540 minutes played, two goals and zero assists. But that gushing description from one senior source at the club, speaking to BBC Sport anonymously, hints at an emerging talent interesting a host of top clubs and why there are some unsubstantiated reports of a 100m price tag on his head. With the Saints facing an immediate relegation back to the Championship, Dibling's future is likely to be one of the summer's more interesting sagas, with Manchester United, Arsenal, Tottenham and Bayern Munich all reportedly chasing his signature. Another source close to the club suggested Southampton turned down previously unreported bids of 35m from Tottenham and 30m from RB Leipzig in January, with the club valuing Dibling at 55m at the start of the winter window. Southampton have not commented on those rumours, but what is known is that Dibling is one of the lowest paid players in Southampton's squad and has a deal that expires in 2027, after Southampton triggered a 12-month extension option. He signed his last contract in December 2023, when he had played just five minutes of senior football. The England Under-21 international has so far resisted the club's offers of a new deal in what has been a breakthrough season for him, despite a wretched campaign which could still see Southampton relegated with the Premier League's lowest ever points total. His dribbles completed per game (2.34) and fouls won per game (2.57) place him in the top 10. "He's the most fearless player I've ever worked with," former Saints Under-21 head coach Adam Asghar tells BBC Sport. "He's totally unique to anything I've seen before.

Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks \cite{qi2023fine}-- a few harmful data mixed in the fine-tuning dataset can break the LLMs's safety alignment. Existing mitigation strategies include alignment stage solutions \cite{huang2024vaccine, rosati2024representation} and fine-tuning stage solutions \cite{huang2024lazy,mukhoti2023fine}. However, our evaluation shows that both categories of defenses fail \textit{when some specific training hyper-parameters are chosen} -- a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense, which however, is necessary to guarantee finetune performance. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains \textbf{\textit{agnostic to the training hyper-parameters in the fine-tuning stage}}. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks.Our project page is at \url{https://huangtiansheng.github.io/Antidote_gh_page/}

The Biden campaign is facing its first major cheapfake scandal this week. Doctored clips of Biden at the G7 Summit and a Hollywood fundraiser have spread across platforms like X, claiming to show Biden wandering off, mumbling unintelligibly, or, uh, even pooping his pants. It's exactly the type of content the right-wing media apparatus drools over to play up Biden's age, despite the clips being edited in a manner reminiscent of the drunk Nancy Pelosi video from last cycle. And while we're all starting to get stressed over simple editing and cropping techniques again, Big Tech is training political campaigns on their generative AI tools. Could a little direction help mitigate the issue?

A large body of work in psycholinguistics has focused on the idea that online language comprehension can be shallow or `good enough': given constraints on time or available computation, comprehenders may form interpretations of their input that are plausible but inaccurate. However, this idea has not yet been linked with formal theories of computation under resource constraints. Here we use information theory to formulate a model of language comprehension as an optimal trade-off between accuracy and processing depth, formalized as bits of information extracted from the input, which increases with processing time. The model provides a measure of processing effort as the change in processing depth, which we link to EEG signals and reading times. We validate our theory against a large-scale dataset of garden path sentence reading times, and EEG experiments featuring N400, P600 and biphasic ERP effects. By quantifying the timecourse of language processing as it proceeds from shallow to deep, our model provides a unified framework to explain behavioral and neural signatures of language comprehension.

In energy-efficient schemes, finding the optimal size of deep learning models is very important and has a broad impact. Meanwhile, recent studies have reported an unexpected phenomenon, the sparse double descent: as the model's sparsity increases, the performance first worsens, then improves, and finally deteriorates. Such a non-monotonic behavior raises serious questions about the optimal model's size to maintain high performance: the model needs to be sufficiently over-parametrized, but having too many parameters wastes training resources. In this paper, we aim to find the best trade-off efficiently. More precisely, we tackle the occurrence of the sparse double descent and present some solutions to avoid it. Firstly, we show that a simple $\ell_2$ regularization method can help to mitigate this phenomenon but sacrifices the performance/sparsity compromise. To overcome this problem, we then introduce a learning scheme in which distilling knowledge regularizes the student model. Supported by experimental results achieved using typical image classification setups, we show that this approach leads to the avoidance of such a phenomenon.