'There's this deep mystery of what, actually, is this thing?': the philosopher inside Google DeepMind AI Since 2017, Iason Gabriel has worked at the tech giant, trying to anticipate - and think through - the impact of AI. But as commercial and geopolitical pressures escalate, can ethicists make any difference? In 2017, a 33-year-old political philosopher named Iason Gabriel was told by a friend that he ought to apply for a job at DeepMind, the London-based subsidiary of Google where much of its AI research was concentrated. The suggestion was not an obvious one. Gabriel was a cheerful but intense junior academic with a passion for Vipassana meditation and what his brother calls "enthusiastic" rock climbing. At the University of Oxford, where he was a fellow at St John's College, Gabriel taught courses on political theory and wrote papers on the moral contortions of "yuppie ethics" and the ethical blind spots of effective altruism. When he wasn't there, he did crisis work for the United Nations Development Programme in Sudan and Lebanon. DeepMind, meanwhile, was the world's leading AI research lab. In part, this was because it had the financial and computational backing of Google, which had bought the company in 2014 for $650m. In part, it was because DeepMind had recently shown it could put those resources to stunning use. In Seoul, in 2016, a DeepMind system called AlphaGo defeated Lee Sedol, a South Korean Go champion, in a five-game match. The victory was significant not least because of Go's legendary complexity; the game has more possible configurations than there are atoms in the universe. Thanks to the fuss around AlphaGo, Gabriel was aware of DeepMind.

With the increasing sophistication of Large Language Models (LLMs), it is crucial to develop reliable methods to accurately identify whether an interlocutor in real-time dialogue is human or chatbot. However, existing detection methods are primarily designed for analyzing full documents, not the unique dynamics and characteristics of dialogue. These approaches frequently overlook the nuances of interaction that are essential in conversational contexts. This work identifies two key patterns in dialogues: (1) Human-Human (H-H) interactions exhibit significant bidirectional sentiment influence, while (2) Human-Chatbot (H-C) interactions display a clear asymmetric pattern. We propose an innovative approach named ChatbotID, which applies the Granger Causality Test (GCT) to extract a novel set of interactional features that capture the evolving, predictive relationships between conversational attributes.

An essential catalyst for the remarkable breakthroughs in AI that led to the modern large language models (LLMs) such as ChatGPT and Gemini has been the algorithms used to train these models on massive datasets. While the LLM architectures have gotten progressively more complex, the training algorithms have stayed relatively simple, and in fact, they have all been based on the decades-old paradigm of stochastic gradient descent (SGD). The key idea behind SGD is that in order to minimize a certain objective function (such as an LLM's error on the training data), it suffices to access only a noisy estimate of that objective at any given time (e.g., based on a small sample of the data) while making incremental progress towards the solution. This is essential for LLM training, as the datasets have become so massive one could not hope to perform computations on everything all at once. Commonly attributed to a 1951 paper by Robbins and Monro [34], SGD has seen a resurgence of interest over the last 20 years by AI researchers and computer scientists striving to understand its effectiveness, leading to numerous variants and extensions used in modern LLMs [12, 9], most notably the Adam algorithm [25]. As a result, we have gained a robust mathematical understanding of the computational complexity of SGD algorithms in a wide range of settings (e.g., see [11, 15, 5, 17]). Yet, despite this progress there is a surprising gap in the understanding of SGD: The complexity of an algorithm proposed by Stefan Kaczmarz in 1937 [24] for solving a system of linear equations - the oldest published example of an SGD algorithm, which predates Robbins and Monro's paper by over a decade - has not been settled.

People overthink; language models over-sample, and the extra effort can talk both into a worse answer. Reasoning systems answer a hard question by sampling it many times (test-time scaling), and the more they draw, the more often a correct answer turns up somewhere, so coverage, the fraction of problems with at least one correct try, climbs and appears to be progress. But a deployed system must return one answer, and choosing it, not knowing which try is right, is selection; selection is capped, and past a point extra samples only make the model surer of a confident mistake, even as every draw adds cost. The gap between climbing coverage and stalled selection, the identifiability gap, is the answer a model can produce but not pick. So the real question is not whether to sample but how far, and the answer is: not far. For picking an answer, the vote has already settled within a few dozen draws, the modal ceiling; for scoring a benchmark, sooner still, the correlation ceiling. Beyond that, extra draws cost compute and add nothing, and can even make the answer worse. This paper turns the cutoff into a single number, the effective number of samples, that any sampling run already reveals. The bottleneck is recognizing a right answer, not generating one.

Contrastive embedding models trained with scale-invariant losses are typically paired with distance metrics like cosine similarity, effectively ignoring embedding magnitudes. However, surprisingly, empirical studies reveal that despite this, these "discarded" norms seem to correlate with semantic properties such as concept specificity, token frequency, and human uncertainty. In this work, we provide a formal theoretical framework explaining this phenomenon. By analyzing the optimization dynamics, we derive an analytic formula demonstrating that embedding length naturally encodes this information as a byproduct of the training process. We also show how this gives rise to signals that can serve as "free" calibration tools in specific models and retrieval tasks, providing a grounded explanation for a previously heuristic observation.

In recent years, models based on the Transformer architecture have seen widespread applications and have become one of the core tools in the field of deep learning. Numerous successful techniques, such as parameter-efficient fine-tuning and efficient scaling, have been proposed surrounding their applications to further enhance performance. However, the success of these strategies has always lacked the support of rigorous mathematical theory. To study the underlying mechanisms behind Transformers and related techniques, we first propose a Transformer learning framework motivated by distribution regression, with distributions being inputs, connect a two-stage sampling process with natural language processing, and present a mathematical formulation of the attention mechanism called attention operator. We demonstrate that by the attention operator, Transformers can compress distributions into function representations without loss of information. Moreover, with the advantages of our novel attention operator, Transformers exhibit a stronger capability to learn functionals with more complex structures than convolutional neural networks and fully connected networks. Finally, we obtain a generalization bound within the distribution regression framework. Through the aforementioned theoretical results, we further discuss some successful techniques emerging with large language models (LLMs), such as prompt tuning, parameter-efficient fine-tuning, and efficient scaling. We also provide theoretical insights behind these techniques within our novel analysis framework.

Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these explanations are sufficient, i.e., if they contain enough information to explain the model's output-generating process. We generalize classical sufficiency from feature attributions to arbitrary explanations and prove that explanation sufficiency can change depending on the input distribution, which must be explicitly defined for LLM explanations. We propose using the LLM itself to generate alternative inputs conditioned on an explanation, capturing its beliefs about possible inputs. We formalize self-consistent sufficiency as a goal for free-text explanations and introduce an information-theoretic metric, SCSuff, that enables evaluation of free-text explanations without relying on predefined biases or shortcuts. Our experiments show that SCSuff agrees with targeted perturbation tests where applicable and demonstrate that explanation sufficiency can vary with the input distribution. We find LLM explanations are generally insufficient and weakly correlated with model size, accuracy, or output entropy. Analysis of final-token hidden states shows that top and bottom SCSuff scores can be predicted from internal representations, suggesting that SCSuff can guide detection and improvement of sufficient LLM explanations. The code for this paper is available at https://github.com/rajesh-lab/self-consistent-sufficiency .

We study how training data contributes to the emergence of toxic behaviors in large language models. Most prior work on reducing model toxicity adopts *reactive* approaches, such as fine-tuning pre-trained (and potentially toxic) models to align them with human values. In contrast, we propose a *proactive* approach--IF-Guide--that leverages influence functions to identify and suppress harmful tokens in the training data. To this end, we first show that standard influence functions are ineffective at discovering harmful training records. We then present a novel adaptation that measures token-level attributions from training data to model toxicity, along with techniques for selecting toxic training documents and a learning objective that can be integrated into both pre-training and fine-tuning. Moreover, IF-Guide does not rely on human-preference data, which is typically required by existing alignment methods. In our evaluation, we demonstrate that IF-Guide substantially reduces both explicit and implicit toxicity--by up to 10$\times$ compared to uncensored models, and up to 3$\times$ compared to baseline alignment methods such as DPO and RAD--across both pre-training and fine-tuning scenarios. IF-Guide is computationally efficient: a billion-parameter model is *not necessary* for computing influence scores; a million-parameter model--with 7.5$\times$ fewer parameters--can effectively serve as a proxy for identifying harmful data.

Google reportedly capped Meta's use of Gemini AI for coding and chatbots Google reportedly capped Meta's use of Gemini AI for coding and chatbots Even tech giants with their own LLMs are having trouble finding enough computing power. Google was forced to cap Meta's use of its Gemini AI model after Mark Zuckerberg's company exceeded its computing capacity, sources familiar with the matter told The Financial Times . The incident reveals that even tech giants with their own LLMs are having trouble finding enough computing power for themselves, let alone their customers. Meta does not operate its own cloud business and is trying to rapidly expand its own data center build out, having pledged $600 billion in cloud computing investments over the next two years. Google reportedly warned the social media company about its capacity limits in March, in turn forcing Meta to request that employees use tokens more efficiently, the sources said.

Reinforcement learning (RL) has become popular in enhancing the reasoning capabilities of large language models (LLMs), with Group Relative Policy Optimization (GRPO) emerging as a widely used algorithm in recent systems. Despite GRPO's widespread adoption, we identify a previously unrecognized phenomenon we term Lazy Likelihood Displacement (LLD), wherein the likelihood of correct responses marginally increases or even decreases during training. This behavior mirrors a recently discovered misalignment issue in Direct Preference Optimization (DPO), attributed to the influence of negative gradients. We provide a theoretical analysis of GRPO's learning dynamic, identifying the source of LLD as the naive penalization of all tokens in incorrect responses with the same strength. To address this, we develop a method called NTHR, which downweights penalties on tokens contributing to the LLD. Unlike prior DPO-based approaches, NTHR takes advantage of GRPO's group-based structure, using correct responses as anchors to identify influential tokens. Experiments on math reasoning benchmarks demonstrate that NTHR effectively mitigates LLD, yielding consistent performance gains across models ranging from 0.5B to 3B parameters.