TL;DR summary of stories on the internet
Tenstorrent, led by chip architect Jim Keller, has launched the Grayskull RISC-V alternative to GPUs. The Grayskull processors, available as e75 and e150 DevKits, excel at handling run-time sparsity and conditional computation, making them great for AI development. The Milestone launch comes after a partnership with Japan’s LSTC to build a 2nm AI Accelerator. The […]
Read more »
Kākāpō is a web bundler for Nix strings that adds context, inspired by the flightless parrot. The author’s motivation stems from an illness and an interest in quirky hacks like //users/sterni/nix/html/README.md. Basic usage involves writing a bundle from a derivation. Bundling a file tree from an attribute set can be done using kakapo.bundleTree. The tool […]
S3, hailed as an “object store,” is actually a cloud filesystem for files, not objects. The unix file API consists of five basic functions handling numerous concerns in a “deep” module fashion. S3’s API, simpler than the unix file API, lacks partial overwrites crucial for databases, limiting its capabilities. While S3 excels in high bandwidth […]
LlamaGym simplifies fine-tuning Large Language Model (LLM) agents with reinforcement learning (RL) in a Gym-style environment for web data extraction. Although LLM-based agents typically don’t learn online via RL, LlamaGym’s abstract class streamlines the process, allowing for easy experimentation with agent prompting and hyperparameters across different environments. By following a few simple steps to implement […]
Andrej Karpathy’s groundbreaking tweet in 2023 highlighted the desire for programming in English over traditional languages like Python and C++. The dream of specifying abstractly without writing code brought about Intentional Software, led by Charles Simonyi, but didn’t fully succeed. With the rise of Large Language Models (LLMs) like GPT-4, programming in English seems achievable, […]
Reading Day is a pivotal moment in a child’s life, marking the transition from illiteracy to literacy within a single day. On this day, children are subjected to various tests and procedures to determine their reading abilities. George Platen, an eight-year-old boy, undergoes these tests and is transformed from a non-reader to a reader within […]
In Fall 2022, the documentary “The Bleeding Edge” revealed shocking information about the FDA’s medical device clearance process, highlighting how devices can be fast-tracked without clinical trials, leading to patient injuries. The process includes 510(k) clearance, which allows new devices to be marketed if equivalent to existing ones without trials. Digging into the data, the […]
The Yi model family, created by a team of authors including Alex Young and Shawn Yue, features language and multimodal models with impressive capabilities. These models are built on pretrained language models and expanded to include chat models, long context models, depth-upscaled models, and vision-language models. The models have shown strong performance on various benchmarks […]
At a fork in the road of programming language semantics lies the choice between call-by-value (CBV) and call-by-name/call-by-need (CBN) evaluation strategies. Languages like Rust, Java, JavaScript, Python, and C/C++ opt for CBV, while Haskell stands as the sole proponent of CBN. However, a recent breakthrough in the form of Call-by-Push-Value (CBPV) introduces a third option […]
Author presents protodump, a tool for extracting protobuf definitions from compiled binaries, useful for reverse engineering closed-source APIs. By analyzing the byte array generated by protoc (Google’s protoc compiler), the field names and types can be extracted and decoded. A search strategy involving the ASCII string “.proto” is employed to locate FileDescriptors, which can be […]