#Data/Training

AIShanghai company Maniformer has opened a crowdsourcing platform that pays ordinary people to record everyday tasks, such as folding clothes or making tea, as training data for robots. The platform lists 22 categories, over 5,000 task types and more than 50,000 real scenes, and nearly 20,000 users signed up during a closed test from August 15 to September 15. Maniformer also announced a RMB 100 million subsidy plan and a goal of 1 million daily active users and 1 million robot trainers within two years.

AIMetabee launched Mifeng Pai, which it calls the world's first full-category crowdsourcing platform for high-quality physical AI data, at a Shanghai event on September 23. Contributors earn from tasks such as folding clothes or making tea, with payouts within two minutes, and the platform covers 22 categories, more than 5,000 task types and 50,000 real-world scenes. Metabee also announced a 100 million yuan subsidy plan and a robot trainer certification program.

AIvLLM now supports NVIDIA Vera Rubin NVL72, with daily container builds and support for models from DeepSeek, Moonshot AI, Z.ai, and MiniMax. In early AgentX benchmarks, vLLM running MiniMax M3 delivered up to 7.84x the throughput per GPU of GB200 NVL72 at matched interactivity. The post is an early look, and the team expects further gains from ongoing optimizations.

AIDatology's team tested its Curation Studio by generating large volumes of synthetic data on Modal, scaling to hundreds of B300 GPUs on demand. The post frames Curation Studio as a customizable data curation tool for teams training their own models.

AISabi has raised $50 million from Khosla Ventures, Accel, Initialized, Kevin Weil and DST Global to build a wearable brain-computer interface cap. The author says the company already has a TSMC-fabricated chip that reads brain signals without touching the scalp, with custom sensors and an in-house AI model that decodes the signals into text. The author notes that the device is a wearable cap, not an implant.

AIA 10-person stealth lab founded by Peking University PhD student Tian Keyu raised $30 million from Wolves VC and IDG at a $200 million valuation, according to Leiphone. Tian, who was sued by ByteDance for 8 million yuan over a 2024 sabotage incident, plans a base model trained on about 100 million hours of video using a 200,000-symbol visual vocabulary, targeting a 2027 release.

AIGeekPark reports that Krinwave, a Shenzhen company, has completed a new 400 million yuan round with XVC and Sequoia China as new investors. The company says its low-frequency ultrasound captures brain structures through the skull, which conventional high-frequency devices cannot, and it aims to supply hardware and its kOS software platform to brain-computer interface and NeuroAI groups.

AISharpa has unveiled the D01 humanoid robot, the W02 dexterous hand and the AE01 exoskeleton data glove at IROS. The article says the D01 has full-body electronic skin, and the W02 has 21 active degrees of freedom and a hand about 30% smaller than the W01.

AIBASAL Intelligence, founded by Tsinghua AIR's first PhD graduate Li Jianxiong, has secured lead funding from Five Source Capital, with Ivy Capital, Linear Capital and WestSummit Capital participating. The company says it is betting on an action-native In-Context embodied foundation model that lets robots learn on site from a few human demonstrations and trial attempts within minutes, without retraining after a scene change.

AIVoice Arena's Monsoon ASR dataset fine-tuned Whisper Medium on Bengali FLEURS, reducing LLM word error rate from 85.27% to 7.65%. The corpus spans 100,000 hours across 50 languages, and Voice Arena says more than 80 organisations have asked to license it since its launch a week ago.

AIAtomic Canyon's Nuclear Industry Virtual Assistant (NIVA), developed with nuclear-industry groups, is now available to the entire U.S. fleet of 94 reactors after pilot testing at Constellation Energy plants. Nuclearn says its products have reached more than 65 U.S. partners, and the article says the industry is turning to AI to help manage regulatory paperwork and a shrinking, aging workforce.

AIAdaption Labs has launched Invent-a-dataset, which creates post-training datasets from a natural language description alone. A technical report compares the Invent API against frontier proprietary and open-weight LLMs used for data generation. Sara Hooker's post links to the launch page and the report's technical details.

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.

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.

AIPenguin Mail 1.0.5 is a free, GPL-3.0-or-later email and calendar app for x86_64 Linux that supports Gmail, Microsoft, IMAP and POP3 accounts. The app includes an optional AI assistant that stays off until a model is chosen and can run locally through LM Studio or Ollama, asking before it sends mail or changes settings.

AIDeveloper Niko1221 open-sourced Strata, an engine that runs quantized Qwen3.8-Flash-Next, a 125B-parameter model, on consumer GPUs with at least 12GB of VRAM. It loads the MoE model into RAM and keeps only frequently used experts in VRAM, and uses a lightweight model for speculative decoding. On an RTX 5070 with 64GB RAM, the Q2_0 quantization generates 94 tokens per second, and an RX 9070 XT reaches 60 tokens per second.

AILiving Models paired Gemma 4 with BOTANIC-1, a model trained on plant DNA across 320 species, to identify which mutations alter crop traits. In a melon yield test, the pipeline ranked the target mutation first out of 2,494 candidates in under four minutes. Google AI says the approach could help researchers engineer climate-resilient crops faster than traditional breeding trials.

AIEra, launched by Ofir Ehrlich's team, generates a complete simulated company spanning Salesforce, Slack, Jira, Zendesk, Gong, and Deel, plus cloud databases and storage. Agents interact with it through live MCP and API interfaces, and because Era generated the company, it knows the exact ground truth for testing and benchmarking. The post says the product is live today and free.

AIEmpero announces Homebrew, an open agent that builds AI models for users. According to the post, users describe what they want, and the agent picks a base model, prepares data, sets parameters, and runs training. The project is available on GitHub at empero-org/homebrew-ai.

AIMastra platform now lets users attach a VPC-isolated Postgres database to any environment, restricting access to resources on the same network. The database cannot be reached from outside the network, so psql connections from external clients return an error. VPC Postgres joins Turso and Neon as managed database options, with MongoDB and Redis coming soon; it requires mastra@1.32.0 or later.

AIFulcrum's Echo was trained using Fireworks' serverless training API on a rank 32 LoRA of Kimi K3. The team's writeup describes a "persona post-training" approach and a total training cost of $4.5k.

AIwh announced joining Proximal in San Francisco, citing open problems in data and post-training for large models. Proximal builds infrastructure that lets AI improve from real-world experience, and the background describes over $200M in annualized revenue from helping frontier labs and enterprises improve coding agents.

AIProximal, a year-old startup supplying coding data to AI labs, raised funding from General Catalyst at a $300m valuation. In the past 10 months, it has surpassed $200m in annualized revenue, reflecting the labs' ongoing demand for data.

AIMastra says it now lets developers add lifecycle and transform hooks to its observational memory system. Lifecycle hooks log cycle starts and ends with token usage, while transform hooks can modify observations and reflections before they are stored. The release also includes skillResultRedactor(), which replaces skill file contents in tool results with a placeholder so the agent still remembers the skill.