Topic
Neural networks trained on very large text corpora, used to generate text and to classify or normalise language at scale.
Cryptography specialists say they are overwhelmed by requests to validate amateurs' GPT-6 Astra cipher solutions, Nature reports. A claimed solution can be ready by morning, and in Nature's two examples each went to one named person for confirmation.
Reality
LLMs fail to learn from atomic coordinates at up to 70 billion parameters and 2 million structures, a paper on nature.com reports. Specialized geometric architectures still win when structure dominates the property, the authors report.
Reality
Minixhofer and colleagues converted subword language models to read raw bytes using 49.1 billion training tokens, under 1% of a typical pretraining budget. Byte-level models have long trailed subword systems, and starting from a trained model is how this work closes most of that gap.
Reality
Carter Church used OpenAI's GPT-6 Astra to decode an 1809 Napoleonic cipher letter in six hours from one prompt and one scan. Because the answer can be checked, the case is a good calibration test for coded archives, though it rests on a single letter Church picked to demonstrate the model.
Reality
Alexandr Wang says the new model costs developers no more than 1.2 and finishes the same work on about a quarter fewer tokens. That is a real saving on high-volume code generation and a rounding error most other places.
Perspective Coverage
3 publishersReality
OpenAI says better caching and inference let it cut API prices for Sol and Luna by half, and the cost advantage it claims for the cheap tier over the old top tier comes in at one tenth on the benchmark it published and one hundredth in its summary.
Perspective Coverage
8 publishersReality
OpenAI says artificial general intelligence may have arrived with Astra, and the researchers Scientific American quoted put that threshold either several years behind us or still out of reach.
Reality
A University of Pennsylvania group read five years of GLP-1 discussion from nearly 70,000 Reddit users and came out with two symptom families worth a real study, plus a nausea signal that shows the method finds what is already known.
Reality