GlucoFM was developed as a self-supervised foundation model for continuous glucose monitoring that separates slow glycemic trends from short-term deviations while accounting for time of day and missing data. It was pre-trained on 109,066 hours of unlabeled CGM data and, across 14 cohort–task evaluations, achieved an average PR-AUC of 58.8% versus 54.7% for the best GluFormer baseline variant, and 21.88 mg/dL MAE for two-hour post-meal glucose-change forecasting. The model improved diabetes-risk, beta-cell-dysfunction, and most insulin-resistance predictions, transferred well across cohorts, and adapted with very limited labeled data.
The article lays out a process for preparing supervised fine-tuning datasets by auditing raw data quality and then formatting examples for training. It highlights that 1,000 carefully curated examples can match models trained on far more data, and notes that filtering to the cleanest 20 percent can improve training speed and scores. As a result, teams should verify correctness, diversity, deduplication, and safety, then structure each example as valid conversational JSONL (including system prompts and, when enabled, reasoning traces and tool-call blocks) before training.
An Amazon employee interviewed about the VGT3 warehouse described scanning and destroying books there by unboxing shipments, cutting spines, and feeding pages into shuttle-style containers for AI training data. The employee said the site uses at least 20 to 25 page-scanning machines that flip through pages very fast. The described workflow changes booksellers’ ability to keep books in circulation, as books that are scanned and processed are mixed in bulk and cannot be put back together for reuse.
Generalist, a robotics startup founded by former DeepMind and Boston Dynamics researchers, raised about $200M in an extension to its Series B, taking its valuation from $2B to $3B in roughly six weeks. The financing pushed the total Series B to $600M after a $400M initial raise in June. The new capital is set to expand compute, hiring, and support for more robot hardware, with the goal of improving model performance such as Gen 1.5 learning from short video demonstrations.
Researchers adapted the Dolma open data-curation toolkit to build the Mangosteen Thai pretraining corpus for Thai language model training. They created a 47-billion-token corpus and removed more than 80% of Common Crawl data and nearly half of FineWeb2 while maintaining or improving Thai LLM performance. The pipeline was redesigned to handle Thai text boundaries and quality issues, replacing filters and tools so training data better reflects Thai sources and supports stronger evaluations on Thai cultural knowledge.
Generalist, a robotics startup, was valued at $3 billion after raising additional funding led by 8VC. The new round adds nearly $200 million, bringing the Series B extension’s total to $600 million. The company’s AI foundation model for multi-robot learning is set to expand use with customers as feedback is used to tailor it to specific tasks.
Tool-calling distillation pipelines that repeatedly re-run supervised fine-tuning on teacher-generated trajectories keep paying the frontier-teacher cost while generating-and-filtering only passing cases. On τ 2-bench, 57% of teacher trials fail, with two-thirds of those failures being near-misses, so the current process leaves the hardest scenarios without useful training signal. PROOF-Gen changes the post-training approach by turning those optimized data signals (including near-misses) into better distillation for deployable models.
Sentence Transformers released v6.0 with a new MultiVectorEncoder model type and a complete finetuning training approach for ColBERT-style late interaction retrieval. In a medical evaluation, the author’s finetuned mLateOn-medical model was trained in 14.5 hours on a single RTX 3090 and outperformed general-purpose retrieval models on that benchmark. The update enables multi-vector (late-interaction) embedding models to be fine-tuned or trained from scratch within the Sentence Transformers training pipeline.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
