Twitch will use streamers’ channel content to train Amazon’s generative AI models unless users opt out in settings. The opt-out toggle appears in the bottom of users’ security settings and defaults to being allowed. Users who disable it will prevent their streams, VODs, clips, stream chats, and channel text/images from being used in future Amazon model training, though some AI features like AutoMod and captioning may still use data without training.
AllenAI’s tutorial shows how to run an end-to-end post-training pipeline for a compact instruction-tuned language model on the Open Instruct framework, covering SFT, DPO, and GRPO with verifier-based evaluation for math answers. It adapts the original multi-GPU Tulu 3 setup to fit within a 16 GB runtime. As a result, the workflow keeps Open Instruct’s optimization logic while swapping out distributed training pieces (vLLM/Ray/DeepSpeed/async queues) for lightweight Hugging Face and PyTorch code suited to Colab.
Booksellers suspect AI firms are buying rare books and then destroying the physical copies after removing pages to scan them for training AI models. The practice is described as involving wood chippers and cropping, scanning, and trashing torn-out pages. As a result, some physical copies may be lost permanently beyond what has been reported so far.
Wiley Science and Engineering Content Hub·2 months ago·
7
A white paper from IEEE Spectrum and Wiley reports how more than 700 practitioners build visual and physical AI and why models fail. It says 78% of teams already see measurable value from visual and physical AI while 74% view the field as underinvested. The report argues that successful teams spend nearly 3x more time on data work and that data curation—not larger model architectures or collecting more data—drives production outcomes.
Meta released Muse Glimmer, a 30-billion-parameter open-weight model distilled from Muse Spark and shipped as a complete distillation pipeline with licensing and deployment artifacts. It is designed to fit on a 24GB machine. This makes enterprise model management shift toward tracking both the upstream “teacher” and downstream “student” refresh cadence, quantization, paired speculative decoding artifacts, and licenses rather than treating distillation as a single generic technique.
Docling researchers released DocLang, an XML sub-language intended for LLMs to interpret document structure directly. The new design explicitly maps each semantic token to a single LLM token, with the DocLang syntax vocabulary capped at about 1,000 tokens. This shifts work from later, error-prone structure recovery to a canonical, machine-native representation that can reduce token usage and lower latency/inference costs.
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.
