Search: “SumUp Plus”

Smart search ranks by meaning as well as keywords (one row per story, last 45 days).

ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents

ScienceBuddy released: interactive scientific agent workspace coupling harness evolution with model reinforcement learning for continual self-improvement across four scientific task families.

ScienceBuddy is an interactive scientific research workspace that turns researcher requests, feedback, and execution evidence into tasks and evaluation rubrics for continual learning. Its recursive-in-recursive self-improvement paradigm couples harness evolution with the model fixed (inner recursion) and model reinforcement learning under the improved harness (outer recursion). Case studies span four scientific task families covering researcher interaction, harness refinement, and model learning. The system is released as a research product at science-buddy.io.

Hugging Face daily papers·updated Sep 15, 17:55 UTC · 1d ago·first Sep 14, 20:00 UTC · 2d agoAI research 2 sources

40

Zombie Card Attack Can Revive Expired Visa Cards for Contactless Payments

UMass Amherst researchers demonstrate Zombie Card, an NFC relay attack that revives expired Visa contactless cards for in-store purchases without breaking cryptography.

Researchers at the University of Massachusetts Amherst presented the Zombie Card attack at USENIX Security 2026, showing that Visa's Kernel 3 does not cryptographically bind the Application Expiration Date (tag 5F24) the terminal reads with the Track 2 expiry seen by the issuer. By positioning an NFC man-in-the-middle relay, an attacker can rewrite the terminal-facing expiration date of an expired card and complete contactless purchases, provided the account remains open under the same PAN and the bank does not independently re-check expiry. Testing across five major US banks found three distinct policies; Visa Kernel 3 accepted the modified date, while Mastercard, American Express, and Discover kernels declined modified transactions. Findings were disclosed to Visa and affected banks in May 2025, no CVE has been assigned, and no exploitation has been reported.

The Hacker News·Aug 20, 18:30 UTC · 27d agoResearch

58

Objective vs. Search: Decomposing What Makes a Good Tokenisernew

New tokeniser study shows search procedure, not optimisation objective, drives bits-per-byte performance across model sizes, vocabulary sizes, and multilingual settings.

The paper disentangles BPE and UnigramLM along two axes: optimisation objective (compression vs log-likelihood) and search procedure (bottom-up merging vs top-down pruning). Two new algorithms, BottomUpLL and TopDownComp, complete the 2x2 design space, and trained language models are evaluated on bits-per-byte and BLiMP across model sizes, vocabulary sizes, and English-only vs multilingual domains. Bottom-up tokenisers consistently achieve lower bits-per-byte in most settings, while BLiMP shows no consistent relationship with design choice.

arXiv cs.AI / cs.LG / cs.CL·Sep 16, 17:59 UTC · 10h agoAI research

28

Playing log(N)-Questions over Wikipedia Abstracts: Communication Efficiency Between Paired Frontier Modelsnew

Six frontier models play a two-agent log(N)-Questions game; Claude Opus 5 lags with 28/68 wins while the top five are near-tied.

The study evaluates six frontier models on a two-agent game where a questioner must identify one of N Wikipedia lead paragraphs in exactly log2 N yes/no questions, run over 408 games at $363 total API cost. Claude Opus 5 wins 28 of 68 games versus 45-56 for GLM-5.3, GPT-5.6 Sol, Grok 4.6, Gemini 3.8 Flash, and Kimi K3. Pooled top-five win rates decline with set size (r=-0.973) and fit win = p^(log2 N) with per-round reliability p=0.928, and information per question correlates with win rate at r=+0.88.

arXiv cs.AI / cs.LG / cs.CL·Sep 16, 17:41 UTC · 10h agoAI research 19 sources

35

Distill Globally, Adapt Locally: Reasoning Distillation and Product-Type Test-Time Training for Scalable Trade-Up Recommendation

A distillation framework compresses LLM reasoning into a 15.5M-parameter trade-up recommendation model reaching AUC 0.941 with product-type test-time training.

The paper targets trade-up recommendation, which identifies higher-quality alternatives that preserve customer purchase intent. A retrieval-augmented few-shot LLM teacher generates labels and rationales that supervise a compact embedding-pair classifier; at inference the 15.5M-parameter student uses only two precomputed 768-dimensional embeddings with no LLM calls. On 8,352 annotated pairs, label-only training scored AUC 0.912, reasoning distillation reached 0.924, and product-type test-time training lifted it to 0.941 with average precision 0.940. The distilled student is roughly 5,000x faster and 10,000x cheaper than direct LLM inference on a 100K-pair proxy catalog.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 17:08 UTC · 12d agoAI research

25

Competence-Gated Pooling of Language Models and Priors for Event Forecasting

Paper proposes a competence gate pooling language model forecasts with external priors, improving Brier score from 0.0771 to 0.0732 across 2,357 binary questions.

The paper defines a language model's relative competence as its marginal value beyond an available external forecast, and derives conditions under Brier loss where model disagreement improves that forecast. A competence gate estimates domain-level source weights from resolved outcomes, shrinks uncertain estimates toward a global weight, and recalibrates the pooled forecast. Across 2,357 resolved binary questions and five language models, it improves the external baseline from 0.0771 to 0.0732 Brier and beats global forecast combinations, though it defers to the market on ForecastBench. Across four Qwen models, verbal confidence failed to identify when the model outperformed the external forecast, while outcome-estimated competence supported better abstention.

Hugging Face daily papers·Sep 9, 20:00 UTC · 7d agoAI research

38

Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluationsnew

Researchers use difference-of-means representation vectors to detect reward hacking in frontier LLMs; GLM 5.2 hacks 73% of SWE-bench rollouts.

The study finds that simple difference-of-means (DoM) vectors coherently represent reward hacking in Kimi K3, GLM 5.2, and Qwen 3.8 Max across common evaluations. GLM 5.2 reward-hacks in 57.2% of rollouts on DeepSWE and 73% on SWE-bench. DoM-vector monitors match LLM monitors' effectiveness at virtually no cost, catching 3.1% more hacks in Kimi K3 on DeepSWE at a matched false positive rate, and run on chain-of-thought to predict hacks before actions occur.

arXiv cs.AI / cs.LG / cs.CL·Sep 16, 17:31 UTC · 10h agoAI safety & security

55

[AINews] Jev: a “System One Model” that only decides/classifies/routes/scores — >100x faster, >200x cheaper than small frontier LLMs

TypeSafe launches Jev, an RLCD-trained decision model claiming 20-200x faster, 40-400x cheaper classification than frontier LLMs, alongside Gemini 3.8 Live and Neon.

TypeSafe's Jev is a 'System One' decision model trained with RLCD, claiming 20-200x faster and 40-400x cheaper classification and routing than frontier LLMs with free output tokens and no hallucinated text. Google launched Gemini 3.8 Live and 3.8 Live Extended Thinking, supporting 97 languages and async tool calls, debuting #1 on Artificial Analysis' speech-to-speech index at 82.6. Periodic Labs' Neon is a ~1T-parameter XRD analysis model trained with RL on proprietary lab data using 1,300 H200s, lifting FrontierXRD success from 2.7% to 55.3% and beating GPT-6 Astra at lower inference cost.

Latent Space·Sep 16, 11:09 UTC · 17h agoModel release1

75

E2A-Bench: Benchmarking Evidence-to-Action Reliability in Financial Chart Reasoning

E2A-Bench, a 969-query financial chart reasoning benchmark, finds VLMs fail evidence-to-action consistency, with fine-tuning amplifying BUY:SELL bias 4-6x.

E2A-Bench is a 969-query benchmark built from 323 HS300 constituents across three input modalities with deterministic OHLCV-derived evidence anchors, evaluating grounding, reasoning-action consistency, evidence-confidence calibration, and directional coverage via UCR, RCI, ECI, and NDR metrics. Testing 20 VLMs showed the lowest-hallucination model ranked near the bottom on coverage with only 6.4% directional coverage, and oracle-aided verification reduced unsupported claims but could collapse coverage. Financial fine-tuning amplified the BUY:SELL ratio by factors of 4.21 to 4.68 across base-fine-tuned pairs.

Hugging Face daily papers·Sep 12, 20:00 UTC · 4d agoAI research

34

Generative Marketing Mix Modeling: A Causal Inference Framework Linking GEO and GEM to Business Impact

Paper proposes Generative Marketing Mix Modeling to causally estimate Generative Engine Optimization and Marketing effects on business outcomes.

The authors develop GMMM, a causal inference framework for measuring how often users see and notice a firm's name in generated answers, which standard marketing data ignore. For GEO it combines repeated generated answers with question counts, shares of generative-system usage and notice probabilities; for GEM it uses sponsored placement records with notice probabilities. The framework compares expected business responses under alternative treatment sequences, establishes identification conditions, and is evaluated on simulated product-recommendation answers in English and Japanese.

arXiv cs.AI / cs.LG / cs.CL·Sep 10, 17:57 UTC · 6d agoAI research

20

Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents

Duplex Cue evaluation shows PersonaPlex full-duplex agents adapt in-turn to listener contributions in only 34.8% of collaborative cases versus 68.2% for humans.

The paper introduces Duplex Cue, an evaluation of in-turn adaptation in full-duplex voice agents that separates listener intent (backchannel, collaboration, interruption) from speaker behavior (continue, adapt, yield). Using 208 scorable pairs from 300 human-confirmed cues in unscripted English conversations, it compares recorded human responses with PersonaPlex continuations generated while replaying listener audio. Humans adapt within the turn in 68.2% of collaborative pairs versus 34.8% for PersonaPlex, which otherwise continues unchanged (42.4%) or yields (22.7%).

arXiv cs.AI / cs.LG / cs.CL·Sep 11, 17:46 UTC · 5d agoAI research

15

Attention Quantization for Tabular Foundation Models

FP8 quantization of attention queries, keys, and values speeds tabular foundation model inference up to 1.7x with no accuracy loss.

The paper develops an FP8 quantization strategy targeting attention calculations (queries, keys, values) in tabular foundation models, arguing attention matters more than weight or KV cache quantization given their differing size and serving patterns versus LLMs. Aligning quantization error between test rows and training rows proves crucial, since misalignment causes drastic accuracy drops. A Triton kernel using explicit FP8 matrix multiplication achieves up to 1.7x speedup over regular 16-bit kernels, with no relevant accuracy loss on TabPFN-v3 and TabICLv2 across TabArena and BeyondArena benchmarks.

arXiv cs.AI / cs.LG / cs.CL·Sep 11, 16:31 UTC · 5d agoAI research1

28

Verifiable by Construction: Claim-Level Evaluation of Verbatim Citation in Clinical Question Answering

Evaluation of twelve LLMs on 222 clinical questions shows verbatim quotes rarely substantiate claims; claude-opus-5 fully substantiates only 37.1%.

The authors build a standardized harness over four clinical practice guidelines and evaluate twelve LLMs on 222 synthetic clinical questions, measuring citation attachment, verbatim quote production, and claim substantiation. Most models attach verbatim quotes to over 90% of claims from prompting alone, though lightweight models like claude-haiku-4.5 struggle. Quotes frequently fail to substantiate claims: claude-opus-5 quotes 98.0% of claims but fully substantiates only 37.1%, exposing a capability gap for verifiable clinical QA.

arXiv cs.AI / cs.LG / cs.CL·Sep 14, 17:53 UTC · 2d agoAI research

38

Quantum Feature Engineering for Credit Default Prediction: When and Why IQP Circuits Help Linear Classifiers

Quantum IQP circuit features lift logistic-regression credit-default F1 from 0.462 to 0.517, beating Kernel PCA at an equal feature budget.

Using the UCI Default of Credit Card Clients dataset and five-fold cross-validation, an 8-qubit IQP circuit adds 16 features that raise Logistic Regression F1 from 0.462 to 0.517 (+0.055, p < 0.0001). Kernel PCA, the best classical non-linear alternative, reaches only 0.493 at the same feature count, with the gap surviving Benjamini-Hochberg correction across 12 tests (p = 0.00007). Only the linear classifier benefits, pointing to a linear-expressivity mechanism. Feature selection matters: Random Forest importance-guided selection reaches F1 = 0.523 while maximally uncorrelated features drop to 0.496.

arXiv cs.AI / cs.LG / cs.CL·Sep 9, 17:41 UTC · 7d agoAI research

30

Show HN: LLM Attention Visualization

A developer released a browser-based tool that visualizes which past tokens influence each LLM output token using aggregated, value-weighted attention scores.

A Show HN project presents a React application built on Transformers.js that renders per-token attention influence by aggregating attention weights scaled by value-vector magnitudes across all attention heads and layers. To expose internal tensors, the author instrumented the ONNX computation graph, hosted a modified model on Hugging Face, and pre-generated prompts to avoid long model downloads in the browser. Demos with a 600-million-parameter model show how verbatim copying draws heavily on source tokens and how single outputs blend information from multiple phrases.

Hacker News · AI·Sep 8, 16:59 UTC · 8d agoAI tools & infraHN 41↑ · 9 comments

28

FlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience

FlowBalance is a verifier-grounded self-improvement method that beats FlowRL on Qwen3-4B and Qwen3-8B math reasoning while improving training stability.

FlowBalance calibrates dense self-guidance scores with verifier-derived group advantages: guidance is retained on positive-advantage trajectories, reversed on negative-advantage trajectories, and disabled when rollout groups show no outcome preference. The method exponentially reweights a reference policy via trajectory balance, with guarantees including within-group contrast preservation and a minimum-change reverse-KL characterization. On mathematical reasoning it outperforms FlowRL on Qwen3-4B and Qwen3-8B, trains faster and more stably, avoids direct OPSD's response-length collapse, and shows higher correct-strategy diversity on AIME24.

Hugging Face daily papers·Sep 2, 20:00 UTC · 14d agoAI research

32

ActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation

ActReview post-trains Qwen3-8B-Base on OpenReview rebuttals to generate actionable peer-review feedback with grounded revision suggestions, benchmarked on 1,000 curated instances.

The paper defines Actionable Peer-review Generation as diagnostic claim generation plus revision suggestion generation and introduces ActReview, a rebuttal-guided post-training framework. From OpenReview review-rebuttal threads the authors build ActReview-40K, aligning reviewer weaknesses with author responses grounded in localized paper evidence, and post-train Qwen3-8B-Base with multi-task SFT followed by GRPO using weakness-specific rubric rewards. They also release ActReview-Bench, a human-curated 1,000-instance benchmark, on which ActReview outperforms prior specialized review-generation models on actionability and grounding while remaining competitive with strong prompt-based LLMs. Human evaluation confirms improved revision usefulness but identifies a remaining gap in technical accuracy.

Hugging Face daily papers·Sep 7, 20:00 UTC · 9d agoAI research

30

CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs

CoVeR, a training-free coverage-based token pruner, preserves 93.5% of VLM 3D-reasoning performance using only about 8% of visual tokens.

Researchers introduce CoVeR, a deterministic, training-free selector that chooses visual tokens to cover every region of a multi-view 3D scene using only token coordinates. Unlike learned-importance and voxelization pruners, it enforces an exact per-scene token budget, avoids saturation plateaus, and prevents near-duplicate selections. Experiments across four vision-language models show it surpasses prior state of the art by 3.9 percentage points on average across three 3D reasoning benchmarks.

Hugging Face daily papers·Sep 7, 20:00 UTC · 9d agoAI research

45

Technical Manual for a Toolkit for Measuring Contextual Individuation in Transformer Language Models

An open methodology toolkit measures whether transformer language models contextualize fixed word forms across domains using bridge forms and layer-wise silhouette analysis.

The manual documents an open toolkit built around 'bridge forms' - identical written words recurring across two or more subject domains with a different sense in each - to test whether transformer language models individuate word occurrences by context beyond the embedding layer. It covers declarative specification of bridge forms, Wikipedia corpus acquisition, occurrence localization, layer-wise representation extraction, domain-pairwise silhouette measurement, and visualization, justifying each choice against failure modes such as sense contamination and subword-tokenization misalignment. It is a methodological and implementation reference and reports no empirical results.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 16:38 UTC · 12d agoAI research

14

Locus: A Framework for Exploring and Optimizing Point Addition Hardware for Zero-Knowledge Proofsnew

Locus framework automates ASIC and FPGA point-addition designs for elliptic curves, achieving 2.71x speedups and 3.11x area reductions for ZKPs.

Locus is a framework that automatically generates ASIC and FPGA implementations of elliptic curve point addition (PADD) for supported equation forms, enabling exploration of over 1,000 design points. On a 12nm technology node, its designs achieve a 2.71x geomean speedup and 3.11x geomean area reduction versus prior ASICs, plus 34.67x geomean speedup over CPU. Integrated into a prior ZKP accelerator at iso-area, it yields a 3.15x geomean speedup on end-to-end proof generation. The framework is open source on GitHub.

arXiv cs.CR·Sep 16, 15:53 UTC · 12h agoResearch

20

Learning to Solve Hard Problems in RL for LLMs by Never Giving Up

Paper introduces Never Give Up adaptive sampling, fixing RL's 'Matthew Effect' where compute is wasted on easy problems and hard problems see little improvement.

Researchers identify a 'Matthew Effect' in reinforcement learning for LLMs, where RL yields large gains on easy problems but minimal improvement on hard ones because compute is misallocated. They propose Never Give Up (NGU), an adaptive sampling method that keeps generating samples for a problem until one is correct, using asynchronous RL to filter easy problems cheaply and concentrate compute on hard ones. NGU improves performance per compute on the Deepscaler math benchmark and iteratively solves the Manufactoria coding task where standard GRPO with per-test reward fails.

Hugging Face daily papers·Sep 10, 20:00 UTC · 6d agoAI research

45

StepAudio 3 Music Technical Report

StepAudio 3 Music introduces long-form text-controlled music generation using ABC-notation planning and flow-matching diffusion, ranking near the top music arena.

StepAudio 3 Music generates long-form, text-controlled music using a 50-Hz single-codebook tokenizer with 65,536 entries and a flow-matching diffusion Transformer over VAE latents. A Mixture-of-Experts autoregressive model first plans an arrangement in ABC notation (ABC-CoT) before predicting music tokens. With DPO fine-tuning, it tops AudioBox content and production quality scores and reaches Quality Elo 1105 on the Artificial Analysis Music Arena, behind Suno V5.5 and Mureka. Generation covers songs, accompaniment from dry vocals, and cover synthesis up to 5 minutes 30 seconds at 48-kHz output.

Hugging Face daily papers·Sep 10, 20:00 UTC · 6d agoAI research

30

ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement

Researchers propose ModularRSI, a modular benchmark-disjoint recursive self-improvement framework that evolves agent harnesses across five modules, improving TB2.0 and SWE-Bench Verified results.

ModularRSI targets generalizable recursive self-improvement (RSI) for agent harnesses by contrasting successful and failed trajectories for the same task and aggregating evidence across tasks to find recurring behavioral deficiencies. It decomposes the evolvable harness into five modules—Agent Loop, Tool Use, Observation Management, Context Management, and Task Completion Detection—each evolved independently within a restricted scope, then integrated with conflict resolution. Using 2,000 executable evolution tasks disjoint from evaluation benchmarks, it shows consistent gains on TB2.0 and SWE-Bench Verified and transfers across different foundation models.

Hugging Face daily papers·Sep 13, 20:00 UTC · 3d agoAI research

30

What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation

A new benchmark shows LLMs reach 68.3-93% accuracy propagating local revisions across conversationally generated artifacts, with parallel-sample selection most cost-effective.

The paper introduces a benchmark for revision propagation: when users request a local change, LLMs must identify dependencies and update all affected parts of an artifact generated through conversation, where context lives in the chat history. Nine revision methods, including sequential reflection and parallel sampling variants, were evaluated on gpt-oss-20b/120b, gpt-5.4-mini, and qwen3.5-9b/27b/122b. Baselines scored 68.3-93% accuracy, and selecting among three parallel samples via LLM-based or medoid selection improved accuracy by 2.2-9.7% as the most cost-effective test-time compute strategy. Code and dataset are released.

Hugging Face daily papers·Sep 2, 20:00 UTC · 14d agoAI research

30

Unlocking Lossless Speedups in LLMs via Discrete Diffusion

Uno pairs autoregressive LLMs with lightweight diffusion weights to draw multiple tokens in parallel, delivering up to 3x lossless speedup without a draft model.

The paper introduces diffusion-augmented LLMs: autoregressive weights trained with the standard next-token objective plus lightweight diffusion weights trained via a Diffusion Distillation phase to emit multiple tokens in parallel. Psi-Spec samplers enable lossless acceleration without the separate draft model required by speculative decoding. The 8B Uno model outperforms the 26B open DiffusionGemma and proprietary Mercury 2 on agentic tool use, coding, and long-context reasoning benchmarks, with up to 3x throughput gains over the base model at all evaluated batch sizes. Code and checkpoints are released publicly.

Hugging Face daily papers·Sep 2, 20:00 UTC · 14d agoAI research

55

Register Tokens for Bounded-State Reasoning in Diffusion Language Models

Register tokens let diffusion language models like LLaDA and Dream carry reasoning state across cleared chunks, gaining up to 19.5 points on code.

Researchers propose register tokens: dedicated fixed-position tokens whose continuous hidden states are trained to carry reasoning progress across generation chunks in masked diffusion language models. After decoding and clearing a chunk, the model continues from the prompt and the carried register state instead of retaining earlier text. On LLaDA and Dream, registers outperform discrete-text carry on every benchmark, with gains up to 8.5 points on math and 19.5 points on code. Registers are especially effective for bounded code generation and can be further refined with reinforcement learning on long-horizon reasoning tasks.

Hugging Face daily papers·Sep 13, 20:00 UTC · 3d agoAI research

38

Google's new AI model predicts the future from sales data, weather, and discount schedules

Google Research released TimesFM-3, a 330M-parameter multivariate time series forecasting model that tops Gift-Eval, FEV-Bench, and Time benchmarks and is on Hugging Face.

Google Research released TimesFM-3, a 330-million-parameter Transformer-based time series forecasting model trained on more than one trillion real and synthetic data points. It works zero-shot and adds multivariate support, ingesting related series, historical-only covariates, and known future events such as discount schedules and weather forecasts, while filling all future time steps in a single one-shot pass. Google reports first place among pretrained forecasting models on Gift-Eval, FEV-Bench, and Time, ahead of Amazon's Chronos-2, the Toto-2.0 family, and its own TimesFM-2.5. Weights are available on GitHub and Hugging Face, with BigQuery integration planned in the coming weeks.

The Decoder·Sep 12, 09:26 UTC · 4d agoModel release2

48

Google Releases Gemini 3.8 Live and 3.8 Live Extended Thinking for Production Grade Voice Agents

Google launches Gemini 3.8 Live and Extended Thinking speech-to-speech models for production voice agents, topping speech-to-speech benchmarks.

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, native speech-to-speech models for real-time voice agents, available hosted via the Gemini Live API and AI Studio. Extended Thinking ranks #1 on Artificial Analysis' Speech-to-Speech Quality Index with 82.6, scores 68.6% on τ-Voice, 35.1% on Sierra's τ-Voice-banking, and 97.7% on Big Bench Audio. The models support asynchronous function calling, near-real-time visual context, alphanumeric precision, and 97 languages, priced at $0.005/min audio input and $0.018/min audio output. All generated audio carries Google DeepMind's imperceptible SynthID watermark.

MarkTechPost·Sep 15, 20:53 UTC · 1d agoModel release 3 sources

78

Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation

Google Research and partners introduce ToolGrad, a verified tool-chain-first data generation framework reaching 99.8% pass rate and boosting Gemma-3-12B to 83.1 on BFCL.

Researchers from Google, the University of Tokyo, RIKEN AIP, and Tohoku University released ToolGrad, which inverts query-first tool-use data generation by executing and verifying API chains before annotating them with user queries. On the ToolBench database of 16,000+ APIs, ToolGrad raised generation pass rate from 63.8% to 99.8% while increasing tool uses per sample from 2.1 to 3.4 and cutting tool-use steps from 34.3 to 20.0. Fine-tuning Gemma-3 at 1B, 4B, and 12B parameters on the 500-sample ToolGrad-500 dataset lifted ToolGrad-12B to 83.1 on the Berkeley Function Calling Leaderboard, near Gemini 2.5 Pro at 83.2 and ahead of GPT-5 at 74.4. Code is Apache-2.0, with the dataset, PyPI package, and models available on Hugging Face.

MarkTechPost·Sep 11, 06:01 UTC · 5d agoAI research1

58

CanvasAnneal: Curriculum Reinforcement Learning for Diffusion Language Models

CanvasAnneal injects teacher reasoning traces into diffusion canvases during curriculum RL, improving diffusion LLMs on MATH500, Countdown, and Tau2.

CanvasAnneal is a curriculum-guided reinforcement learning framework for diffusion language models that addresses exploration bottlenecks in standard RL. It warm-starts exploration by injecting teacher-generated reasoning traces into the initial diffusion canvas, then gradually removes this guidance so the model generates reasoning trajectories independently. Across mathematical reasoning and tool-use benchmarks, it improves over standard diffu-GRPO on MATH500, Countdown, and Tau2 and accelerates reward improvement, though gains are task-dependent.

arXiv cs.AI / cs.LG / cs.CL·Sep 11, 16:59 UTC · 5d agoAI research1

28