DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
TL;DR When DeepSeek-V4.1 Flash was released, I thought it might just be a post-training iteration version… but after using it for a while, I found it reached nearly 420 Tokens/s in speed, and then Cui said all DeepSeek-V4 Pro models would be taken offline… suddenly I felt this was no small matter… until the Technical […]
OpenSpec – A lightweight and configurable AI spec framework
/opsx:explore map the problem and understand the codebase /opsx:propose draft proposal.md, specs/, design.md, tasks.md /opsx:apply implement tasks from the specification /opsx:verify check the implementation matches the spec /opsx:archive archive completed changes
Breaking the 1.58-bit Barrier for Ternary LLMs
Abstract:Ternary Large Language Models (LLM) store every weight as one of three symbols ${-1,0,+1}$, so the cost of a ternary model is conventionally referenced to the information-theoretic $log_2 3 approx 1.585$ bits per weight. The prevailing deployment format packs five ternary weights into one byte (five-trit packing), and due to the power-of-two group sizes used […]
Backups Aren’t Simple
Aleksandar Filipovski, 2026-09-16 See also: John Salvatier’s excellent blog, Reality has a surprising amount of detail I read a comment somewhere that stuck with me, that went something like this: “There are two types of people: those who have suffered a catastrophic loss of data, and those who will.” Trying to find the source for […]
macOS 27 Golden Gate – Review
Apple hasn’t said what limits users can expect, just that they will vary “by feature, request complexity, system demand, system policies, and other factors” and that additional usage “will be available for a fee.” Apple gates some Home app features, like video summaries from cameras, to upper-tier iCloud+ plans, and I wouldn’t be surprised to […]
How good are frontier models at physics?
Abstract:Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced physics, a demanding test of their scientific reasoning and quantitative problem-solving abilities. Yet this impression does not always align with domain experts’ experiences using these models in their work. […]
Accurate Models of AMD Matrix Cores
Abstract:Matrix multipliers available on recent GPUs do not conform with the IEEE 754 floating point standard. Features of matrix multipliers differ across vendors and architectures of the same vendor, such as accumulator width, rounding behaviour, normalisation points, intermediate underflow and overflow logic, the handling of subnormals, and the treatment of special inputs. As a result, […]
Training a 4B model to produce 81% faster query plans than Postgres
How good are query optimizers, really? Leis et al. asked this exact question in 2015. Then, they asked it again 10 years later. Despite an enormous body of research spanning a decade since their original exploration, they found that query optimizers continue to leave much to be desired. I was surprised when I first learned […]
GitHub is having trouble counting things
In the past week, I’ve noticed two pretty severe UI bugs in GitHub. The first is an overcounting of PRs in the tab on a repository, this is particularly common after you merge a PR, but it’s not the only trigger of this behavior. To add insult to injury, the right count is even right […]
ER visits for gambling disorders doubled after expanded online gambling market
Emergency department visits for gambling disorders doubled following the legalization of single-event sports betting and expansion of Ontario’s online gambling market, according to a new study by University of Toronto researchers. The increase was observed predominantly among young men. Canada legalized single-event sports betting in August 2021. In April 2022, Ontario became the first province […]