Ryan Hsiang, William Adkisson, Robert Joseph George, Anima Anandkumar
原文摘要
Building on recent progress in machine learning for formal theorem proving in Lean, we introduce LeanDojo-v2: an end-to-end framework that unifies repository data extraction, LLM fine-tuning, proof search, AI autocompletion, and LLM-assisted theorem proving in a single software library. In addition, we provide Lean4Code, a lightweight IDE that integrates the various features of LeanDojo-v2 through a simple, mathematician-friendly UI. These tools will enable mathematicians and researchers to easily fine-tune language models on domain-specific Lean codebases and experiment with LLM-assisted theorem proving without a complex setup.
Ryan Hsiang, Robert Joseph George, Anima Anandkumar
原文摘要
Certifying the robustness and physical consistency of scientific machine-learning models ensures that their predictions remain reliable beyond the test set. For neural operators, which learn PDE solution operators by mapping infinite-dimensional function spaces, certification means establishing rigorous, domain-wide guarantees for their output functions. However, standard neural network (NN) verification methods are fundamentally designed for vector-to-vector mappings, and they can only certify neural operators at fixed grid resolutions and provide no guarantees over the entire domain or across another resolution. We close this gap by introducing a function-space certification framework that verifies neural operators across the entire continuous domain. We formulate interval bound propagation (IBP) and CROWN in function space, propagating upper and lower bound functions through each layer using rigorous numerical integration. In addition, we extend bound propagation to higher-order derivatives to certify PDE residuals and solution error bounds. Experiments across several neural operator architectures and scientific problems, including the 572.5-million-parameter spherical Fourier neural operator for global weather forecasting, show that our method produces tight, domain-wide guarantees efficiently, providing a foundation for trustworthy operator learning.
Vincent Guan, Joseph Janssen, Hossein Rahmani, Andrew Warren, Stephen Zhang, Elina Robeva, Geoffrey Schiebinger
原文摘要
Stochastic differential equations (SDEs) are a fundamental tool for modelling dynamic processes, including gene regulatory networks (GRNs), contaminant transport, financial markets, and image generation. However, learning the underlying SDE from data is a challenging task, especially if individual trajectories are not observable. Motivated by burgeoning research in single-cell datasets, we present the first comprehensive approach for jointly identifying the drift and diffusion of an SDE from its temporal marginals. Assuming linear drift and additive diffusion, we show that non-identifiability can only arise if the initial distribution possesses generalized rotational symmetries. We further prove that even if this condition holds, the drift and diffusion can almost always be recovered from the marginals. Additionally, we show that the causal graph of any SDE with additive diffusion can be recovered from the identified SDE parameters. To complement this theory, we adapt entropy-regularized optimal transport to handle anisotropic diffusion and introduce APPEX (Alternating Projection Parameter Estimation from $X_0$), an iterative algorithm designed to estimate the drift, diffusion, and causal graph of an additive noise SDE, solely from temporal marginals, and demonstrate its effectiveness on simulated data from linear additive noise SDEs.
Generative creative interfaces routinely report the user's position within a space of possibilities. Such displays imply that the quantities shown were measured from the artefact; in a growing class of systems they are instead predicted by a language model describing its own output, and the two are indistinguishable on screen. Most creative domains lack an agreed measurement against which such a self-report can be checked; audio admits one. In a measurement study of a working LLM-driven timbre interface, what a user can cause proves separable from what the interface reports. The first is exactly characterisable: the reachable set is a zonotope in parameter space whose dimension increases by one per sound placed, so the space the model may act within is demonstrably the user's own. The second is inaccurate. The displayed position differs from measurement by approximately three times the width of that region, and the error has two independent sources: model-predicted coordinates, whose accuracy varies twofold between the two models evaluated, and the interpolation step, which arises from combining synthesiser parameters and is unaffected by model quality. End to end, 27 of 36 refinement instructions produced no perceptible change, and under the weaker model the facility performed no action while reporting convergence. Effective agency (what a user can cause) and epistemic agency (whether they can know it) are therefore separable and independently engineered: structurally enforced constraints survive model failure, whereas guarantees resting on the model's own report do not. The failure mode that remains is silence presented as success. We close with a short protocol for building and auditing such displays.
The manifold hypothesis suggests that data lies on manifolds with smaller intrinsic dimension (ID) than their ambient dimension. However there is no empirical agreement on the estimates for ID from different estimators for realistic datasets. Thus it is important to test ID estimators (IDEs) with targeted stressors. In this work, we consider the role of anisotropy. To this end, we propose HomID, a collection of homogeneous spaces with anisotropic embedding, for benchmarking ID estimators. We observe that methods that perform well on standard benchmarks systematically degrade on HomID under identical resource allocation. We further observe that anisotropic distortion of such benchmarks also results in performance degradation. Finally, we demonstrate that controlled anisotropic distortions induce systematic shifts in the distributions on which these methods rely, providing a concrete mechanism for the resulting estimation errors in two particular IDEs. The code for the project is available at https://github.com/a-writes-code/homid
Joseph H. Rudoler, Kevin Tan, Giles Hooker, Konrad P. Kording
原文摘要
Deep learning systems are known to exhibit implicit regularization (alt. implicit bias), favoring simple solutions instead of merely minimizing the loss function. In some cases, we can analytically derive the implicit regularization -- connecting it to an equivalent penalty that augments the learning objective. However, modern deep learning systems are complex, carrying modifications to the training procedure and architecture (e.g. early stopping, minibatching, dropout) whose effects are not always directly interpretable. Although estimating the resulting implicit regularization could aid theorists in algorithm design and practitioners in interpreting their hyperparameter choices, this problem has received little direct attention. It is also tractable: regularization makes weight updates deviate from loss gradients, promising a signal for identifying implicit bias. Here we provide gradient matching methods that can be used to empirically estimate the implicit regularization. Our method works on networks with known regularization, recovering popular explicit penalties like $\ell_1$ and $\ell_2$. It also replicates known implicit effects, like the quadratic weight penalty induced by early stopping in gradient descent, demonstrating that it can be used to test theories of implicit regularization. Crucially, because our method is empirical, it can handle implicit regularization in arbitrary networks. We demonstrate this use by characterizing implicit $\ell_2$ effects of dropout in deep networks. Our work shows that practitioners can use gradient matching to understand regularization in networks with implicit biases that are too complicated to derive analytically.
Li Guan, Rachael Granberry, Sahar Aseeri, Sahar Rostami, Taylor Niehues, Sai Dhanoosh Marri, Mia Huang, Dingtian Zhang 等 104 位作者
原文摘要
Collecting naturalistic human manipulation demonstrations with accurate contact force and hand pose is essential for robot dexterous manipulation policy and foundational world model learning - yet no existing wearable platform provides both calibrated fingertip force sensing and occlusion-resilient hand tracking in a form factor that preserves natural manipulation behavior. We present Tactus, a textile-integrated sensing glove that co-optimizes three design pillars: (1) a thin, conformal 3D-knit form factor (1.5 mm fingertip stackup) designed to minimize interference with natural manipulation, (2) strategically placed capacitive tactile sensors at manipulation-relevant regions informed by empirical contact studies, and (3) IMU-fused pose tracking that maintains accuracy through the hand-object occlusions where manipulation data is most valuable. Tactus includes a documented calibration pipeline for cross-glove consistency and per-session calibration for subject-dependent correction. To enable objective comparison, we introduce standardized benchmarking protocols for tactile force regression and hand pose estimation, and apply them to Tactus and two commercially available tactile gloves - to our knowledge the first quantitative cross-platform evaluation of this kind. Tactus resolves a minimum detectable force of 0.59 N at 95% confidence, 1.6 - 2.7x lower than the commercial baselines, and with 0.87 cross-participant consistency, while our pose tracking sustains 11.0 mm fingertip keypoint relative error during manipulation, when occlusion is heaviest. We show that these combined signals support object property estimation, slip detection, and representative recording of hand-object interaction - yielding demonstration data with the fidelity required for contact-aware manipulation policies and world models.
We introduce PLANTEUR, a benchmark for crop-type mapping from satellite images in the French overseas territories. Building on the PASTIS design, PLANTEUR provides georeferenced and aligned Sentinel-2, Sentinel-1, and Landsat-8 sequences with parcel-derived labels spanning six tropical regions and an 8-class label space. We define both a standard test split and a leave-one-island-out protocol to measure cross-territory generalization, and evaluate fully supervised models, linear probes, and fine-tuned foundation models. The benchmark code and data are open source and available on GitHub and Hugging Face.
Gabriel Diaz-Aylwin, Joseph Neighbour, Abiel Malkani Talwar, Rui-Yang Zhang, Henry B. Moss
原文摘要
Bayesian optimisation is the natural tool for shape design when objectives are expensive and non-differentiable, but it needs a compact yet expressive parameterisation of the search space. Hand-crafting one is a complex endeavour requiring domain expertise, and often yields implicit infeasible regions, artificial bounds, and coupled, unordered coordinates. We instead learn the parameterisation from a collection of existing designs, applying principal component analysis to the deformations between shapes. The result is a linear, interpretable search space in which the number of components explicitly trades expressivity against dimensionality. Across aerofoils, wings, and radio-frequency cavities, spanning 2D geometry to 3D aerodynamics and electromagnetics, we show improved sample efficiency and the ability to explore beyond the confines of hand-crafted baselines.
Huanzhi Mao, Yichuan Wang, Michael Koo, Wenyuan Huang, Huy Doan, Jinghan Xia, Ziyao Wang, Joseph E. Gonzalez
原文摘要
Vision-language model (VLM) agents that act in the open world need two abilities: world knowledge to interpret what they see, and exploration to find what they cannot yet see. We study how agents combine the two through interactive visual geolocation. GeoGuesser Bench contains 282 tasks in an offline Street View environment where agents rotate, zoom, and move between real panoramas. Its three task types shift the balance from knowledge to exploration: agents (1) report the coordinates of their starting location, (2) name the enclosing street, city, state, or country, or (3) find, read, and count nearby objects. The coordinate tasks in (1) also include 42 real GeoGuessr World Championship rounds, whose expert guesses provide a human reference. We evaluate 23 VLMs: the best accuracy is 62.5% on place naming and 35.0% on object finding, and the best agent averages 4352.4 GeoGuessr points on the championship rounds, against 4770.3 for the better expert in each round. Beyond these headline numbers, world knowledge and exploration are only weakly correlated across models: agents that localize well do not necessarily explore well. To understand where agents fail, we categorize their errors into vision, navigation, and tool-use failures. Further ablations show that the strongest agents, such as Claude Opus 4.8, still perform well with limited views, while weaker agents degrade; customizing the environment thus adjusts task difficulty. The environment's verifiable rewards and restorable state also support training, and our Planner-Worker-Compressor harness cuts planner tokens by 57-61% by compacting the interaction history. We hope GeoGuesser Bench and its environment serve as a high-quality foundation for simulating real-world exploration at scale.
Huanzhi Mao, Chengkun Cao, Shuo Yuan, Joseph E. Gonzalez
原文摘要
Large language models are increasingly capable of writing and executing code, suggesting a natural paradigm for tool-augmented agents: expressing tool use as programs rather than isolated calls. Programmatic tool calling (PTC) captures this idea through loops, batching, abstraction, and structured computation, but existing PTC approaches are largely stateless, limiting their ability to reuse intermediate results and adapt over time. This is problematic for real-world tasks, where tool environments and problem-solving processes are inherently stateful. We therefore present I-PTC, an interactive framework that executes stateful PTC: models issue code snippets that call tools, maintain reusable state, and evolve the tool interface through API corrections. To evaluate the effectiveness of I-PTC and the general PTC setting, we introduce PTC-Bench, a benchmark of long-tail, production-style user requests whose difficulty comes from large initial environments, many dependent API interactions, extended planning, and recovery across follow-ups and transient failures. Experiments on PTC-Bench show that I-PTC improves performance on complex multi-step tasks with large tool sets while substantially reducing token consumption over conventional tool-calling baselines and the original PTC baseline, highlighting stateful programmatic tool calling as a promising direction for scalable tool-augmented agents.
Difficulty Calibration Across Time for Membership Inference in Diffusion Models
Michael Joseph, Yedid Hoshen
AI 速读 基于摘要
通过沿训练时间和扩散时间两个轴校准图像难度，提升扩散模型成员推理攻击的准确性。
cs.CV · cs.CR · cs.LG
查看原文摘要
Modern diffusion models are trained on massive datasets that often include copyrighted or private images. Membership inference attacks (MIAs) address this by testing whether a given image was in a model's training set. The standard approach scores an image by its reconstruction loss and flags low-loss images as members. This often fails because some images are simply easier to reconstruct than others, so a simple unseen image can score a lower loss than a complex training image. Difficulty calibration separates memorization from how difficult an image is to begin with. We show how to calibrate along two axes of time: training time and diffusion time. First, across training time, we introduce Reference-Model Calibration (RMC), which compares a fine-tuned model's loss against an earlier checkpoint. This cleanly isolates newly memorized data and gives near-perfect auditing, even without text captions. Since earlier checkpoints are not always available, we next present Temporal Self-Calibration (TSC), which calibrates across diffusion time using only a single model. Our key observation is that an image's difficulty persists across noise levels, whereas memorization is localized in time. By comparing the loss at a target timestep against neighboring timesteps from the same model, TSC attenuates image difficulty using the model's own trajectory. Empirically, our methods beat the state of the art by a wide margin, and unlike most existing attacks, generalize across different model parameterizations such as $x$-, $v$-, and $ε$-prediction models. Code is available at https://github.com/MichaelJoseph777/TSC.
We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.
Pre-training foundational models is computationally expensive as models scale. In this work, we introduce Bi-Phase Training (BPT), a parameter-efficient pre-training method that preserves the high-rank learning dynamics observed in full-parameter optimization with substantially reduced trainable parameters. BPT parameterizes each weight update as a composition of constrained high-rank left/right transformations via trainable diagonal matrices and low-rank adapters whose updates are periodically merged into the frozen base weights, resulting in a compounding expansion of accessible directions during training. Across language-model pre-training from scratch (Qwen2.5 models from 100M to 1.5B parameters and OLMo-2-1B) on FineWeb-Edu (10B and 100B tokens), BPT achieves validation perplexity close to full-parameter baselines while using far fewer trainable parameters; e.g., at the 1.5B scale, BPT uses 66% fewer trainable parameters with only a small validation-perplexity gap. Additionally, a comprehensive evaluation across diverse downstream tasks shows that BPT preserves much of the full-parameterized model's downstream performance.
Joseph Natter, Long Vu, Yifan Zhang, Bobin Deng, Kun Suo, Rui Wu
原文摘要
Traffic forecasting requires modeling complex temporal dynamics and the spatial dependencies imposed by road topology. Supervised graph-based models achieve strong performance, but are often trained for a particular dataset and sensor graph, making deployment to new or evolving networks costly. Prior-Data Fitted Networks (PFNs) offer a zero-shot alternative by learning in-context inference from synthetic task priors, which provide an explicit mechanism for encoding domain-specific assumptions. Existing forecasting PFNs, however, do not explicitly represent the graph-structured dependencies governing traffic dynamics. To combine graph-aware traffic modeling with zero-shot adaptation, we introduce the Graph-Aware Spatiotemporal Prior-Data Fitted Network (GAS-PFN). GAS-PFN incorporates sampled road topology into its synthetic task prior through random-walk diffusion, teaching the model to learn graph-coupled dynamics during pretraining. Its co-designed inference architecture uses two mask-conditioned passes through a shared Transformer backbone to separately model sensor-specific temporal dynamics and spatiotemporal interactions among neighboring sensors before integrating their representations for prediction. This design enables GAS-PFN to use target-network structure and context without parameter updates. Across six benchmarks at the 90-minute horizon, GAS-PFN reduces MSE by 34.0% on average relative to TimePFN, its closest PFN-based predecessor. It also reduces MSE by 27.7% on average relative to four recent industry-developed general-purpose zero-shot forecasters. In this comparison, all models receive 90 minutes of contiguous recent history, and any additional context is drawn from each dataset's chronological training split. These results indicate that aligning PFN priors and inference architectures with graph structure improves in-context learning for zero-shot traffic forecasting.
Joseph H. Rudoler, Alexander Asemota, Giles Hooker
原文摘要
As algorithmic decision making has become increasingly ubiquitous even in high-stakes settings, significant effort has been invested in defining, detecting, and eliminating unfair model-driven outcomes. Prior work has largely focused on estimating the fairness properties of a given predictive model, and quantifying uncertainty based on variability in a fixed model's performance across evaluation datasets. Alternatively, we can also ask about uncertainty that the model inherits from its training data; here we evaluate data fairness - fairness in the data generating process used for training. We perform inference on data fairness using targeted learning, a flexible framework for nonparametric inference, and derive estimators for demographic parity, equal opportunity, and conditional mutual information. Simulations show that our data fairness estimators improve confidence-interval coverage and reduce false detections of population unfairness relative to fixed-model inference. We also demonstrate their application to real data.
Summing vectors is a basic task in differentially private data analysis. Standard private sum algorithms clip vectors to some bound, sum them, and add noise scaled to the bound. The differential privacy literature has developed a deep library of noise addition mechanisms, but it offers few tools for privately choosing a clipping bound based on the data. We introduce a method that privately chooses a clipping bound by optimizing residual coherence, the directional alignment of vectors exceeding the bound, against the variance cost of additional noise. We prove a close relationship between residual coherence and bias for a general class of heavy-tailed data regimes and show empirically that, across several datasets, our method outperforms existing baselines.
Aditya Chatterjee, Anand Joseph X M, Sanjay Singh Chauhan
原文摘要
We address a fundamental question: How far should kernel fusion extend and which boundaries are worth preserving? Unlike the general convention of using one monolith megakernel at lower batch sizes or supporting only a selective model component as megakernel and rest of the model as conventional graph, we show that the best configuration is a composition of multiple megakernels with no overlapping conventional graph component, or a fully conventional graph. We present a framework that selects megakernel regions in a streaming inference model from conventional-graph profiles and standalone kernel benchmarks, without running any candidate design end to end. It simulates each candidate's tile and event graphs to predict time to first output and selects the measured best design in all 13 combinations of model and batch size we tested. Across NH-TTS, Qwen3-TTS and Llama-1B, the best composition ranges from two to four megakernels to fully conventional execution and none is a single megakernel. We introduce four model categories (megakernel-stable/ receding/ advancing/ averse) that capture how a model's best megakernel composition changes with batch size. Using NH-TTS model on H100 NVL GPU, we demonstrate that preserving three specialized megakernels can improve both latency and throughput despite introducing additional kernel boundaries. At batch size 32, this design delivers 21.4% lower mean batch-p95 time to first audio (TTFA) and 28% higher throughput than the evaluated monolith megakernel. Against conventional TensorRT graph execution, it reduces engine-level mean batch TTFA p95 by 4.05x and increases throughput by 19.7%.
Joseph An, Phillip Keung, Jiaqi Wang, Orevaoghene Ahia, Noah A. Smith
原文摘要
Audio language models process input audio into rich frame-level representations, but the standard approach to temporal localization generates timestamps as sequences of text tokens, which discards the frame-level representations in favor of autoregressive decoding. However, generating timestamps as tokens is slow and not parallelizable, and tends to hallucinate when producing timestamps outside the training distribution. We propose internal frame-level reuse, a method that trains audio LMs to reuse their own internal audio representations for temporal localization directly, bypassing token generation altogether. We introduce a lightweight prediction head trained via different frame-level objectives: a binary frame classifier and a novel inhomogeneous Poisson process (IHP) loss that models temporal event intensity. Across word localization, speaker diarization, and event localization tasks, our approach can achieve a >50x inference speedup over token-based generation and demonstrates robust length generalization, maintaining high accuracy on out-of-distribution audio durations where token-based models collapse completely. We find that reusing audio frame-level representations yields localization accuracies comparable to (and often better than) finetuned token-based baselines.
Peter Brodeur, Jacob M. Koshy, Anil Palepu, Khaled Saab, Ava Homiar, Roma Ruparel, Charles Wu, Ryutaro Tanno 等 48 位作者
原文摘要
Large language model (LLM)-based AI systems have shown promise for patient-facing diagnostic and management conversations in simulated settings. Translating these systems into clinical practice requires assessment in real-world workflows with rigorous safety oversight. We report a prospective, single-arm feasibility study of an LLM-based conversational AI, the Articulate Medical Intelligence Explorer (AMIE), conducting clinical history taking and presentation of potential diagnoses for patients to discuss with their provider at urgent care appointments at a leading academic medical center. 100 adult patients completed an AMIE text-chat interaction up to 5 days before their appointment. We sought to assess the conversational safety and quality, patient and clinician experience, and clinical reasoning capabilities compared to primary care providers (PCPs). Human safety supervisors monitored all patient-AMIE interactions in real time and did not need to intervene to stop any consultations based on pre-defined criteria. Patients reported high satisfaction and their attitudes towards AI improved after interacting with AMIE (p < 0.001). PCPs found AMIE's output useful with a positive impact on preparedness. AMIE's differential diagnosis (DDx) included the final diagnosis, per chart review 8 weeks post-encounter, in 90% of cases, with 75% top-3 accuracy. Blinded assessment of AMIE and PCP DDx and management (Mx) plans suggested similar overall DDx and Mx plan quality, without significant differences for DDx (p = 0.6) and appropriateness and safety of Mx (p = 0.1 and 1.0, respectively). PCPs outperformed AMIE in the practicality (p = 0.003) and cost effectiveness (p = 0.004) of Mx. While further research is needed, this study demonstrates the initial feasibility, safety, and user acceptance of conversational AI in a real-world setting, representing crucial steps towards clinical translation.
Shu Liu, Asim Biswal, Amog Kamsetty, Audrey Cheng, Luis Gaspar Schroeder, Liana Patel, Shiyi Cao, Xiangxi Mo 等 11 位作者
原文摘要
Batch data analytics is a growing application for Large Language Models (LLMs). LLMs enable users to perform a wide range of natural language tasks, such as classification, entity extraction, and translation, over large datasets. However, LLM inference is highly costly and slow: for example, an NVIDIA L4 GPU running Llama3-8B can only process 6 KB of text per second, taking about a day to handle 15 GB of data; processing a similar amount of data costs around $10K on OpenAI's GPT-4o. In this paper, we propose novel techniques that can significantly reduce the cost of LLM calls for relational data analytics workloads. Our key contribution is developing efficient algorithms for reordering the rows and the fields within each row of an input table to maximize key-value (KV) cache reuse when performing LLM serving. As such, our approach can be easily applied to existing analytics systems and serving platforms. Our evaluation shows that our solution can yield up to 3.4x improvement in job completion time on a benchmark of diverse LLM-based queries using Llama 3 models. Our solution also achieves a 32% cost savings under OpenAI and Anthropic pricing models.
Dacheng Li, Shiyi Cao, Chengkun Cao, Xiuyu Li, Shangyin Tan, Kurt Keutzer, Jiarong Xing, Joseph E. Gonzalez 等 9 位作者
原文摘要
Increasing test-time compute for LLMs shows promise across domains but remains underexplored in code generation, despite extensive study in math. In this paper, we propose S*, the first hybrid test-time scaling framework that substantially improves the coverage and selection accuracy of generated code. S* extends the existing parallel scaling paradigm with sequential scaling to push performance boundaries. It further leverages a novel selection mechanism that adaptively generates distinguishing inputs for pairwise comparison, combined with execution-grounded information to robustly identify correct solutions. We evaluate across 12 Large Language Models and Large Reasoning Model and show: (1) S* consistently improves performance across model families and sizes, enabling a 3B model to outperform GPT-4o-mini; (2) S* enables non-reasoning models to surpass reasoning models - GPT-4o-mini with S* outperforms o1-preview by 3.7% on LiveCodeBench; (3) S* further boosts state-of-the-art reasoning models - DeepSeek-R1-Distill-Qwen-32B with S* achieves 85.7% on LiveCodeBench, approaching o1 (high) at 88.5%. Code will be available under https://github.com/NovaSky-AI/SkyThought.
Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, James Zou
AI 速读 基于摘要
TextGrad 利用 LLM 提供的文本反馈对复合 AI 系统进行类似自动微分的优化，并在多种任务上展示了效果。
cs.CL · cs.AI · cs.LG
查看原文摘要
AI is undergoing a paradigm shift, with breakthroughs achieved by systems orchestrating multiple large language models (LLMs) and other complex components. As a result, developing principled and automated optimization methods for compound AI systems is one of the most important new challenges. Neural networks faced a similar challenge in its early days until backpropagation and automatic differentiation transformed the field by making optimization turn-key. Inspired by this, we introduce TextGrad, a powerful framework performing automatic ``differentiation'' via text. TextGrad backpropagates textual feedback provided by LLMs to improve individual components of a compound AI system. In our framework, LLMs provide rich, general, natural language suggestions to optimize variables in computation graphs, ranging from code snippets to molecular structures. TextGrad follows PyTorch's syntax and abstraction and is flexible and easy-to-use. It works out-of-the-box for a variety of tasks, where the users only provide the objective function without tuning components or prompts of the framework. We showcase TextGrad's effectiveness and generality across a diverse range of applications, from question answering and molecule optimization to radiotherapy treatment planning. Without modifying the framework, TextGrad improves the zero-shot accuracy of GPT-4o in Google-Proof Question Answering from $51\%$ to $55\%$, yields $20\%$ relative performance gain in optimizing LeetCode-Hard coding problem solutions, improves prompts for reasoning, designs new druglike small molecules with desirable in silico binding, and designs radiation oncology treatment plans with high specificity. TextGrad lays a foundation to accelerate the development of the next-generation of AI systems.
Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda 等 58 位作者
原文摘要
We introduce Codex, a GPT language model fine-tuned on publicly available code from GitHub, and study its Python code-writing capabilities. A distinct production version of Codex powers GitHub Copilot. On HumanEval, a new evaluation set we release to measure functional correctness for synthesizing programs from docstrings, our model solves 28.8% of the problems, while GPT-3 solves 0% and GPT-J solves 11.4%. Furthermore, we find that repeated sampling from the model is a surprisingly effective strategy for producing working solutions to difficult prompts. Using this method, we solve 70.2% of our problems with 100 samples per problem. Careful investigation of our model reveals its limitations, including difficulty with docstrings describing long chains of operations and with binding operations to variables. Finally, we discuss the potential broader impacts of deploying powerful code generation technologies, covering safety, security, and economics.