#Data/Training

AIEpoch AI gave Fable 5 and GPT-5.6 Sol 3000 GPU-hours each to develop a new post-training technique intended to outperform the standard GRPO baseline. The post reports the task setup but does not state the results.

AILucas Beyer says robotics is accelerating in the physical world, not just in AI research. He attributes this partly, though not only, to progress in coding models over the past year. The post cites a related thread on scaling a UMI data collection operation from 5 to 90 operators and reaching over 1M unique tasks.

AIGitHub reports that one in three pull requests now involves an AI agent, and that public secret exposures rise with the volume of pushes rather than from declining developer care. It introduces a ModernBERT-based classifier with Microsoft Applied Sciences that evaluates candidate secrets in under two milliseconds and could more than double the secrets prevented at push time. The feature is in private preview, with availability for GitHub Secret Protection customers later this month.

AIDongxi NLP recommends Sherpa, a framework for training LLMs to teach adaptively, tailoring instruction to individual learners rather than simply solving problems. The post presents this as a way for AI to support human learning instead of replacing human teachers, citing the principle of teaching according to each student's aptitude.

AIThe post argues that companies are increasingly recognizing business opportunities from reinforcement learning, since many real-world tasks need specialized models, harnesses, and data flywheels rather than AGI. It predicts a new post-training era led by full-stack AI companies, though it provides no concrete figures or raw data to support the claim.

AIUnsloth says users can train their own Decision Model on as little as 2.5GB of VRAM using small models such as Laya. Training is done through a UI, with a video tutorial and guide linked from the post and the project's GitHub repository.

AIA new model extracts values absent from a document by computing them from figures that are present, such as deriving a yearly product cost from a monthly price. Santiago says the video shows examples of inferring complex formulas. The background post describes this as Higher-Order Extraction, which deterministically computes needed numbers from raw page values.

AIPerplexity has published both pplx-embed-v2 models on Hugging Face, with the collection available at the linked URL. The post provides no further details on model sizes, benchmarks, or capabilities.

AIMicrosoft Research Asia has open-sourced Agent Lightning v1.0, a roughly 3,500-line agentic RL framework that trains the same agent harness used in deployment. In an end-to-end coding agent pipeline, Qwen3.5-9B rose from 41.8% to 56.4% Pass@1 on SWE-bench Verified using about 6,000 training samples. The framework runs agents as standard Kubernetes jobs without paid commercial sandbox services.

Why it matters: The source shows how training with the deployed agent harness avoids rebuilding agents, and reports concrete SWE-bench Verified gains from about 6,000 samples.

AIAi2 is making new Stage 1 checkpoints available, with the byte-level components already trained and the original model's weights left unchanged. Researchers can use them to test new architectures and train the full system without repeating that initial stage.

AIAi2 introduces Bolmo, a model that operates on bytes rather than tokens, which the team hopes will lead to larger models adapting how they represent information across languages and domains. Because bytes can also represent images and audio, the approach could eventually extend beyond text.

AIAi2 describes a recipe that adapts existing models to process raw bytes through a brief additional training run. The approach keeps the model's core intact and adds components that group bytes into variable-length patches for processing, then expand outputs back to byte-level predictions of the next byte.

AIMost language models split text into subwords from a fixed vocabulary, which can obscure spelling details and fragment meaningful units in code or math. Byte-level models instead work directly with the bytes computers use to represent text.

AINVIDIA's cuOpt decision optimization software now targets problems with 100 million variables using mPDLP. The source says cuOpt already delivers more than 10x speedups over CPU solvers, while detailing little else about mPDLP's design or availability.

AIGoogle's Gemma account promotes research showing how AI can help scientists secure the global food supply, linking to a Gemma 4 and BOTANIC-1 demo on Hugging Face. The post also offers the BOTANIC-1 download and the accompanying paper.

AINVIDIA's cuPhoton targets the computational bottleneck in scientific image pipelines, where data from observatories, telescopes, lasers, and X-ray light sources arrives faster than CPU-bound processing can handle. The source says the bottleneck is usually the whole path from raw sensor data to decision, not one slow kernel. The available text does not give benchmark figures, pricing, or availability details.

AIGoogle says its SynthID detector now supports watermarks from all of its partners and is available to anyone on a new SynthID.com website. Previously, outside users could only check for Google's own watermarks by asking Gemini, and the detector did not flag images from other providers such as ChatGPT.

AIHiPHI is a 617.5-hour whole-body human motion dataset captured with optical motion capture at sub-millimeter accuracy, including 245.7 hours of human-object interaction with synchronized object trajectories and meshes. The dataset organizes coverage using FrameNet, a linguistic framework for human action. The white paper also reports results from policies trained on HiPHI and deployed on a physical Unitree G1 humanoid robot.

AINVIDIA reports that fine-tuned Nemotron models reached gold-medal level at both IOI 2026, scoring 535.4 out of 600, and IMO 2026, scoring 30 out of 42. The IOI run was a live, unofficial, unsupervised benchmark, while IMO proofs were graded by official IMO graders. The post also releases checkpoints, datasets, a new 200-problem benchmark, and inference pipelines on Hugging Face and NeMo-Skills.

Why it matters: The post traces how SFT, RL, and a generate-verify-refine loop turned Nemotron into gold-level specialists for IOI and IMO, with the training and inference details shared.

AIOrange AI released "podcast-to-article," a free MIT-licensed open-source skill that reorganizes podcast transcripts by topic rather than speaker timeline. The author says it keeps every number and original quote while cutting off-topic tangents, and that early readership and engagement were good.

AIThe final post in O'Reilly Radar's four-part post-training series walks readers through implementing the classic ChatGPT pipeline on Qwen2.5-1.5B, covering SFT, reward model training, and PPO. The walkthrough uses torchtune for SFT and verl, a Ray-based RL framework from ByteDance's team, for reinforcement learning. The author says the goal is hands-on understanding rather than reproducing InstructGPT, which took a large team and thousands of GPU-hours.

AIAi2 has published its Bolmo byte-level language model research in Nature and released new checkpoints on Hugging Face. The byteifying process converts an existing subword model into a byte-level one with a relatively short additional training run, and the paper reports that it also works for Qwen 3 8B and Llama 3 8B, producing Bwen 8B and Blama 8B. Ai2 also released Stage 1 checkpoints for researchers extending the architecture.

AIFrançois Chollet asks whether the jagged frontier of AI capability is mainly math and code, which can be pushed far with RLVR. He questions whether steady gains in non-verifiable areas come from higher generalization driven by RLVR or only from continued injection of new human data.