4Mathematics is a specialist search engine for Mathematics. Discover the latest math news and mathematical content. Part of the 4SEARCH network of topic specific search engines.
An MIT-led research team has developed an artificial intelligence model that connects descriptions of materials with their three-dimensional crystal structures, offering researchers another way to search for candidates with useful properties. MIT highlighted the work on October 18. The underlying peer-reviewed…
Education ministers will soon receive advice on the findings of a national review of the maths curriculum for the first three years of school. It may be one of the most important curriculum decisions of the decade. Education Minister Jason…
I fine-tuned a toy model to vomit back arbitrary tokens. I used them as Markers tied to specific subsets of its training data to connect training to generation. Training + testing cost $0.49 on an RTX 4090. Could help with interp and safety…
My worst post on LW of all time is Boltzmann brains, like Doomsday, require no explaining. It's currently sitting at -18, which I think would be fair if my argument presented was mistaken. However, no one has of yet has shown…
This article provides a step by step guide to serving Gemma 4 12B, Google's quantization-aware trained (QAT) build, on an 8 GB NVIDIA RTX 3050 with llama.cpp, with every layer on the GPU and Google's multi-token prediction (MTP) drafter for speculative decoding. A…
A studio headshot used to cost hundreds of dollars. Now an AI headshot generator returns a gallery of studio-style portraits for the price of a takeaway dinner — anywhere from about $17 to $79, delivered in 15 minutes to a couple of hours. I…
When building retrieval pipelines for agent memory systems, the difference between a 50ms and 500ms vector search can determine whether your agent responds in real-time or feels sluggish. I spent the last two weeks stress-testing pgvector's HNSW and IVFFlat index strategies under…
Most bad AI video comes from bad prompts, not bad models. A video model has to fill in every gap you leave: what the subject looks like, what moves, where the camera goes, how the light falls, what happens first.
While integrating RAGFlow into our agent memory pipeline, we hit a wall at around 50 concurrent document insertions per second. Query latency spiked from 80ms to 4+ seconds, and recall rates dropped by 30%. The culprit? HNSW index locks and improper embedding normalization. RAGFlow…
This article provides a step by step introduction to speculative decoding with Gemma 4's multi-token prediction (MTP) drafter, served by llama.cpp on a local, desktop hosted GPU enabled system. It is the first of a short series: this part…