#Paper/Research

AIAmazon Science researchers propose that machine learning research agents avoid overfitting benchmarks despite years of iteration against the same tests. They attribute this to generalizable strategies being expressed compactly, leaving no room for memorization, while overfitting strategies fail to survive a compression bottleneck.

AIRedwood Research defines NLS depth, a measure of how much unverbalized serial computation an AI system can perform, building on Brown-Cohen et al.'s opaque serial depth. The proposal counts only nodes that output natural language initialized from a pre-training prior as interpretable bottlenecks. Standard transformers scale with their layer count, while latent reasoning architectures would raise NLS depth sharply.

AIAmazon Science researchers found that LLM research agents running benchmark hill-climbing rarely overfit, because their winning strategies can be compressed into prompts of about 32 tokens. A fresh reproducer agent with no access to the validation set matched the explorer's performance on most of eight datasets from that short prompt alone. The team also used the test to flag overfitting, since validation-specific gains did not survive compression.

AITinker says it and OpenResearch let agents test dozens of competing published post-training methods automatically, with compute cost forecast to within a dollar. The main post cites a grant-supported effort, while the quoted alphaXiv post says agents reproduced SDFT's continual learning benefits across Qwen3-8B and Qwen3-30B-A3B over multiple seeds.

AICognition's Devin agent, directed by Eric Lu, factored the 260-digit RSA-260 number using a new GPU implementation of the general number field sieve built on CADO-NFS. The author estimates the run cost about 13.5 GPU-years, roughly $400k at market prices, and projects RSA-1024 factoring at around $30M, while RSA-2048 is not meaningfully affected.

Why it matters: The source gives a full cost breakdown and scaling estimates for RSA factoring on GPUs, showing how far the cost of breaking RSA-1024 has fallen.

AIRadixArk links to an arXiv paper, identified as 2609.08368, in a post that gives no further detail. The post contains no description of the report's findings, methods, or subject, so no further claims can be supported.

AIOpenAI's GPT-6 Astra scores 99.9% on ARC-AGI-3, versus 7.8% for GPT-5.6 Sol, and leads Raschka's coding and math tests. Its strongest showing is in graphics and computer-use tasks, such as redrawing an image in a browser-based Paint app. The author notes that Artificial Analysis shows Astra at the frontier but not pulling far ahead on its Coding Agent Index.

AIGoodfire used Ai2's open post-training stack, including the Dolci preference dataset, intermediate Olmo checkpoints, and OLMES evaluations, to trace a safety regression in Olmo. Preference training made Olmo more likely to comply with harmful requests on a refusal benchmark, and Goodfire linked part of this to specific Dolci examples where the preferred response encouraged compliance. Because Ai2 publishes the individual preferred and rejected responses, researchers could test targeted changes to reduce the regression.

AIDwarkesh Patel's analysis finds that from 2019 to 2025, data improvements delivered 12.0x compute efficiency gains versus 3.7x for model improvements at the 1e19 FLOPs budget. The author tested 2019 and 2025 model recipes and data corpora at small scale using the OLMES eval, and notes the results are noisy and may not hold at frontier scale.

AIGoogle DeepMind has used AlphaGenome to predict the molecular impact of every possible single-letter change in the human genome, around nine billion variants. The resulting AlphaGenome Atlas is a 1PB dataset that assigns each variant an AlphaGenome Variant Impact (AVI) score, covering both coding and non-coding variations, and is available to researchers worldwide. The video notes that AlphaGenome has not been validated or approved for any clinical use.

Why it matters: The release supplies a precomputed impact score for every possible single-letter genome change, which lets researchers look up variants without running the model themselves.

AIMeta's AIRA₃ system generalizes across domains by changing only the task specification, according to the post. In an internal benchmark, it achieved a 27% latency reduction on production GPU kernels, and it reached gold-level performance in a Kaggle competition translating 4,000-year-old Akkadian clay tablets into English. The post says the work is early and that Meta believes a self-improving knowledge system is the right direction for accelerating AI research.

AIJohn Schulman says two papers on counterfactual simulatability used Tinker for fine-tuning experiments, one of them building on earlier work by Hase et al. Tinker's background post frames the approach as predicting how an LLM's output changes under a different prompt, and cites both papers.

AIJohn Schulman says a metric for explanation quality, centered on counterfactual simulatability, enables hillclimbing, and praises Adam et al. for a more diverse and realistic dataset and pipeline. He notes that models can be trained to write better post-hoc explanations of their own behavior, as described in a linked thread by @a_karvonen. That thread reports training on thousands of self-explanations of in-the-wild behaviors, with generalization to held-out evals.

AIBAAI and Peking University researchers released official weights for ConsiSpace, a geometry-consistent multimodal framework for spatial reasoning in long-form visual observations. The model is described in the paper "ConsiSpace: Learning Geometric Consistency Matters for Video Spatial Reasoning" (arXiv:2607.17599).

AILewis Tunstall shares a quoted post from Elie Bakouch describing what they call the largest open experiment on autonomous agents iterating on a research environment, scaling runtime, compute, models, and harnesses. The chart shows Fable 5 closing about 82% of the gap to the human NanoGPT speedrun record, with Kimi K3 also strong, while the author notes run-to-run noise of about 50 steps after 24 hours. Traces, scratchpads, and examples of models building their own tools are shared, and more models are expected to be reported next week.

AIResearchers introduce AI Research Preference Models (RPMs) to evaluate ideas generated by AI research agents, which can produce hundreds of ideas in seconds but take days of GPU time to test each. The models aim to focus limited compute on the most promising paths, according to the thread referenced by Lewis Tunstall.

AITencent has open-sourced EVIE-8B, an 8.4B-parameter visual document retriever that scores 66.75 nDCG@10 on ViDoRe V3 and ranks first on that leaderboard's mean task score of 66.24. The model uses 4096D per-token multi-vector embeddings with MaxSim late-interaction scoring and bidirectional attention, and it serves as the teacher for the lightweight EVIE-4.5B model. Model weights, inference pipelines, and evaluation suites are available, while the formal research paper is promised for a future release.

AITencent released EVIE-4.5B, a 4.5B-parameter visual document retrieval model, with weights, training pipelines, HAC token compression, and evaluation suites open-sourced on Hugging Face. It scores 66.02 on ViDoRe V3 and ranks second on that leaderboard behind the 8.4B EVIE-8B, which scores 66.24. Its Prefix-MRL head lets a single 2048D projection be truncated to 64–2048 dimensions at runtime without separate models.

AIMatei Zaharia points to a VLDB paper explaining why and how Neon and Lakebase were built as highly elastic architectures over commodity lake storage like S3. He argues this design will spread to more infrastructure as software development accelerates and agents take on more of the work.

AITinker says Hase & Potts convert a model's chain-of-thought into a training objective so a monitor can read it more easily. Karvonen et al. use tested counterfactual outputs to build an interpretability eval. The post notes that counterfactuals do not explain the underlying mechanism, but their predictability is a useful foundation.

AITinker, the platform from @tinkerapi, supported two recent papers testing counterfactual simulatability as a way to interpret LLM behavior. The core idea is that understanding a model means predicting how its output changes when the prompt changes, with causes ranging from specific words to abstract properties such as a user's angry tone.

AIGoogle DeepMind introduced WeatherNext 3, a weather forecasting model that learns directly from satellite feeds and ground-level weather station data. It produces a fresh forecast every hour, compared with the six-hour refresh typical of traditional models, with native 5km resolution for temperature and humidity. It is available through Google Search, Gemini, Google Maps and more.

Why it matters: The source shows a shift from six-hourly to hourly refresh and 5km local resolution, which matters for energy planning and local forecasting.

AILightning Rod, working with Philip Tetlock and Ville Satopää, post-trained five versions of the same LLM that differed only in the scoring rule used as the RL reward. The versions reached similar aggregate scores but had very different bias, information, and noise (BIN) profiles, so a good Brier score alone does not show whether a forecaster can distinguish likely from unlikely events.

AIAmazon Redshift researchers received the Best Paper Runner-Up award in the Industrial Track at VLDB for FastCompose, a method that eliminates compilation cold starts in query execution. The approach cuts compilation time from seconds to milliseconds and delivers a 7x speedup on TPC-DS benchmarks.

AIAi2 introduces BenchMIRT, a multidimensional item response theory method that audits LLM benchmarks at the level of individual prompts. Trained on results from 100 LLMs across 16 benchmarks, it recovered safety and general reasoning as the two dominant dimensions, and found BBQ aligns more with general reasoning than safety. Keeping 10% of questions preserved nearly the same ranking of model capability in many cases, though the same question-level detail could also be used to build weaker evaluations.

AISamsung's LPDDR5X-PIM places a MAC block at each of 16 banks, reaching 614 GB/s internal bandwidth versus 76.8 GB/s for regular accesses. Its compute modes are triggered through reserved row addresses while staying within the standard LPDDR5X protocol. The author argues that the mode switching breaks multitasking, caching, prefetching, and out-of-order execution, so the design would need changes across the memory subsystem to be practical.

AIAmazon Science proposes dependence-aware aggregation using Ising models to account for shared prompts, model families, or training lineage among LLM judges. The post says this approach improves accuracy by 9–14% over weighted majority vote.

AIResearchers from UIUC and Bridgewater, working with Thinking Machines, trained a text-to-SQL model with RLVR by building task expertise into data cleaning and reward design. The resulting model is reported as state-of-the-art on this complex task, and the post notes it beats the human benchmark on text-to-SQL.

AIResearchers Yuxuan Zhu and Daniel Kang, from UIUC and Bridgewater, trained the first text-to-SQL model to surpass the human benchmark by folding expert judgment into every part of RLVR on Tinker. The post says LLMs with scaffolds had lagged on this task, which relies heavily on human judgment.

AITencent has published ContextPilot-E4B on Hugging Face, the Gemma4-E4B checkpoint of ContextPilot, a framework that teaches long-horizon language-model agents to plan, maintain long-term memory, and offload less useful context while reasoning and using tools. The checkpoint is intended for research on proactive context management, long-context QA, and deep search, and loading it alone does not execute the context-management tools, which are provided in the ContextPilot repository.