arXiv subjects

Hans Gundlach

Publications and source records attributed to Hans Gundlach.

10 recordsLinked to original sources

arXiv2026

Two AI Metrics Diverged: Will it Make All the Difference?

As exponential compute scaling continues, will the capabilities of frontier AI models outstrip what is accessible to developers on a small fixed budget? Or will capabilities converge, with "meek models inheriting the earth"? Building on Gundlach et al. (2025b), we show that the answer depends on how we value and measure AI capabilities. We discuss conventional performance measures and show that, while validation loss shows a shrinking gap, on other metrics frontier models grow their lead forever. Classifying performance metrics by their functional forms in relation to training (and inference) compute, we provide tight mathematical conditions for determining which metrics favor meek models, and show that bounded performance metrics always do. But careful interpretation of performance metrics is essential: we show that many common bounded metrics have closely-related counterpart metrics that are unbounded (and vice versa). Determining the apt metric in a domain is a prerequisite for policy, since bounded and unbounded metrics may suggest opposing policy responses. If a particular capability -- like software engineering, synthetic biology, or rhetorical persuasiveness -- is unbounded when measured in the terms we care about, frontier-level capability will likely be concentrated in the hands of a few wealthy actors. Conversely, if that capability is instead bounded, frontier-level capabilities proliferate through meek models into the hands of the many.

cs.AI↗

arXiv2026

Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts

Artificial intelligence poses many risks, ranging from familiar present-day harms to unprecedented and potentially catastrophic ones. Effective risk management requires prioritization: we must understand which risks are most severe, who is most vulnerable, and who is most responsible for addressing them. We report results from a three-round Delphi study conducted late 2025 with 272 international AI experts. Experts rated 24 AI risks on harm probability and severity, sector and actor vulnerability, actor responsibility, and overall concern. Experts estimated the five most severe harms in the next 5 years were likely to come from dangerous capabilities, competitive dynamics, weapons & cyberattacks (including CBRNE), power centralization, and false information. In a business-as-usual scenario, experts judged 18 of 24 risks as having a more than 10% probability of catastrophic outcomes (e.g., more than 1 million deaths or more than USD 100B in financial loss) in the next 5 years (2025-2030). In a scenario where pragmatic mitigations are implemented, experts still judged five risks as having a more than 10% probability of catastrophic outcomes: dangerous capabilities, weapons & cyberattacks, environmental harm, inequality & unemployment, and power centralization. All 24 risks were judged as being more than 5% likely to cause catastrophic outcomes. AI users and the general public were judged the most vulnerable to these risks, but experts assigned the highest responsibility for addressing them to general-purpose AI developers and governance actors (including governments, regulators, and standards bodies). Across most risks, experts identified information, finance, and national security as the most vulnerable sectors. These findings can guide AI risk prioritization and clarify expert expectations about who should bear responsibility for mitigation.

cs.CY↗

arXiv2026

The Price of Progress: Price Performance and the Future of AI

Language models have seen enormous progress on advanced benchmarks in recent years, but much of this progress has only been possible by using more costly models. Benchmarks may therefore present a warped picture of progress in practical capabilities *per dollar*. To remedy this, we use data from Artificial Analysis and Epoch AI to form the largest dataset of current and historical prices to run benchmarks to date. We find that the price for a given level of benchmark performance has decreased remarkably fast, around $5\times$ to $10\times$ per year, for frontier models on knowledge, reasoning, math, and software engineering benchmarks. These reductions in the cost of AI inference are due to economic forces, hardware efficiency improvements, and algorithmic efficiency improvements. Isolating out open models to control for competition effects and dividing by hardware price declines, we estimate that algorithmic efficiency progress is around $3\times$ per year. However, at the same time, the price of running frontier models is rising between $3\times$ to $18\times$ per year due to bigger models and larger reasoning demands. Finally, we recommend that evaluators both publicize and take into account the price of benchmarking as an essential part of measuring the real-world impact of AI.

cs.LG↗

arXiv2025

On the Origin of Algorithmic Progress in AI

Algorithms have been estimated to increase AI training FLOP efficiency by a factor of 22,000 between 2012 and 2023 [Ho et al., 2024]. Running small-scale ablation experiments on key innovations from this time period, we are able to account for less than 10x of these gains. Surveying the broader literature, we estimate that additional innovations not included in our ablations account for less than 10x, yielding a total under 100x. This leads us to conduct scaling experiments, which reveal that much of this efficiency gap can be explained by algorithms with scale-dependent efficiency improvements. In particular, we conduct scaling experiments between LSTMs and Transformers, finding exponent differences in their compute-optimal scaling law while finding little scaling difference for many other innovations. These experiments demonstrate that - contrary to standard assumptions - an algorithm's efficiency gains are tied to compute scale. Using experimental extrapolation and literature estimates, we account for 6,930x efficiency gains over the same time period, with the scale-dependent LSTM-to-Transformer transition accounting for the majority of gains. Our results indicate that algorithmic progress for small models has been far slower than previously assumed, and that measures of algorithmic efficiency are strongly reference-dependent.

cs.LG↗

arXiv2025

Quantum Deep Learning Still Needs a Quantum Leap

Quantum computing technology is advancing rapidly. Yet, even accounting for these trends, a quantum leap would be needed for quantum computers to meaningfully impact deep learning over the coming decade or two. We arrive at this conclusion based on a first-of-its-kind survey of quantum algorithms and how they match potential deep learning applications. This survey reveals three important areas where quantum computing could potentially accelerate deep learning, each of which faces a challenging roadblock to realizing its potential. First, quantum algorithms for matrix multiplication and other algorithms central to deep learning offer small theoretical improvements in the number of operations needed, but this advantage is overwhelmed on practical problem sizes by how slowly quantum computers do each operation. Second, some promising quantum algorithms depend on practical Quantum Random Access Memory (QRAM), which is underdeveloped. Finally, there are quantum algorithms that offer large theoretical advantages, but which are only applicable to special cases, limiting their practical benefits. In each of these areas, we support our arguments using quantitative forecasts of quantum advantage that build on the work by Choi et al. [2023] as well as new research on limitations and quantum hardware trends. Our analysis outlines the current scope of quantum deep learning and points to research directions that could lead to greater practical advances in the field.

quant-ph↗

arXiv2025

Quantum Advantage in Computational Chemistry?

For decades, computational chemistry has been posited as one of the areas in which quantum computing would revolutionize. However, the algorithmic advantages that fault-tolerant quantum computers have for chemistry can be overwhelmed by other disadvantages, such as error correction, processor speed, etc. To assess when quantum computing will be disruptive to computational chemistry, we compare a wide range of classical methods to quantum computational methods by extending the framework proposed by Choi, Moses, and Thompson. Our approach accounts for the characteristics of classical and quantum algorithms, and hardware, both today and as they improve. We find that in many cases, classical computational chemistry methods will likely remain superior to quantum algorithms for at least the next couple of decades. Nevertheless, quantum computers are likely to make important contributions in two important areas. First, for simulations with tens or hundreds of atoms, highly accurate methods such as Full Configuration Interaction are likely to be surpassed by quantum phase estimation in the coming decade. Secondly, in cases where quantum phase estimation is most efficient less accurate methods like Couple Cluster and Moller-Plesset, could be surpassed in fifteen to twenty years if the technical advancements for quantum computers are favorable. Overall, we find that in the next decade or so, quantum computing will be most impactful for highly accurate computations with small to medium-sized molecules, whereas classical computers will likely remain the typical choice for calculations of larger molecules.

quant-ph↗

arXiv2025

Introducing the Quantum Economic Advantage Online Calculator

Developing a systematic view of where quantum computers will outperform classical ones is important for researchers, policy makers and business leaders. But developing such a view is challenging because quantum advantage analyses depend not only on algorithm properties, but also on a host of technical characteristics (error correction, gate speeds, etc.). Because various analyses make different assumptions about these technical characteristics, it can be challenging to make comparisons across them. In this paper, we introduce an open-access web-tool designed to make such comparisons easy. Built on the framework introduced by Choi, Moses, and Thompson (2023), it calculates when quantum systems will outperform classical computers for a given algorithmic problem. These estimates can be easily updated based on various assumptions for error correction, overhead, and connectivity. Different hardware roadmaps can also be used and algorithm running times can be customized to particular applications. It can currently be accessed at https://futuretech.mit.edu/quantum-economic-advantage-calculator. This integrated prediction tool also allows us to explore which technical factors are most important for quantum ``economic" advantage (outperforming on a cost-equivalent basis). Overall, we find that for some algorithms (e.g. Shor's) the timing of advantage is quite robust, whereas for others (e.g. Grover's) it is contingent, with numerous technical characteristics substantially impacting these dates. In the paper, we discuss both why this occurs and what we can learn from it.

quant-ph↗

arXiv2025

Meek Models Shall Inherit the Earth

The past decade has seen incredible scaling of AI systems by a few companies, leading to inequality in AI model performance. This paper argues that, contrary to prevailing intuition, the diminishing returns to compute scaling will lead to a convergence of AI model capabilities. In other words, meek models (those with limited computation budget) shall inherit the earth, approaching the performance level of the best models overall. We develop a model illustrating that under a fixed-distribution next-token objective, the marginal capability returns to raw compute shrink substantially. Given current scaling practices, we argue that these diminishing returns are strong enough that even companies that can scale their models exponentially faster than other organizations will eventually have little advantage in capabilities. As part of our argument, we give several reasons that proxies like training loss differences capture important capability measures using evidence from benchmark data and theoretical performance models. In addition, we analyze empirical data on the capability difference of AI models over time. Finally, in light of the increasing ability of meek models, we argue that AI strategy and policy require reexamination, and we outline the areas this shift will affect.

cs.AI↗

arXiv2024

REBUS: A Robust Evaluation Benchmark of Understanding Symbols

We propose a new benchmark evaluating the performance of multimodal large language models on rebus puzzles. The dataset covers 333 original examples of image-based wordplay, cluing 13 categories such as movies, composers, major cities, and food. To achieve good performance on the benchmark of identifying the clued word or phrase, models must combine image recognition and string manipulation with hypothesis testing, multi-step reasoning, and an understanding of human cognition, making for a complex, multimodal evaluation of capabilities. We find that GPT-4o significantly outperforms all other models, followed by proprietary models outperforming all other evaluated models. However, even the best model has a final accuracy of only 42\%, which goes down to just 7\% on hard puzzles, highlighting the need for substantial improvements in reasoning. Further, models rarely understand all parts of a puzzle, and are almost always incapable of retroactively explaining the correct answer. Our benchmark can therefore be used to identify major shortcomings in the knowledge and reasoning of multimodal large language models.

cs.CL↗

arXiv2020

Noise-Robust End-to-End Quantum Control using Deep Autoregressive Policy Networks

Variational quantum eigensolvers have recently received increased attention, as they enable the use of quantum computing devices to find solutions to complex problems, such as the ground energy and ground state of strongly-correlated quantum many-body systems. In many applications, it is the optimization of both continuous and discrete parameters that poses a formidable challenge. Using reinforcement learning (RL), we present a hybrid policy gradient algorithm capable of simultaneously optimizing continuous and discrete degrees of freedom in an uncertainty-resilient way. The hybrid policy is modeled by a deep autoregressive neural network to capture causality. We employ the algorithm to prepare the ground state of the nonintegrable quantum Ising model in a unitary process, parametrized by a generalized quantum approximate optimization ansatz: the RL agent solves the discrete combinatorial problem of constructing the optimal sequences of unitaries out of a predefined set and, at the same time, it optimizes the continuous durations for which these unitaries are applied. We demonstrate the noise-robust features of the agent by considering three sources of uncertainty: classical and quantum measurement noise, and errors in the control unitary durations. Our work exhibits the beneficial synergy between reinforcement learning and quantum control.

quant-ph↗