HackerNews Summary (2023-04-26 12PM)

Source: The Full-Source Bootstrap: Building from source all the way down — 2023 — Blog — GNU Guix

Summary
The Guix Full-Source Bootstrap successfully reduced binaries down to a 357-byte program, achieving the goal of building everything from source, all the way down. This milestone addresses supply chain security concerns, such as the “trusting trust” attack.

Fact
💻 The stage0-posix package is built from the 357-byte hex0-seed binary provided by the bootstrap-seeds, building all the way up to M2-Planet. The new GNU Mes v0.24 release can be built with M2-Planet.

---

Source: HDR-NeRF

Summary
This paper introduces HDR-NeRF, a method to recover an HDR radiance field from a set of LDR views with different exposures. The method accurately models the imaging process to generate both novel HDR and LDR views under different exposures, with experimental results showing its effectiveness.

Fact

🔍 HDR-NeRF recovers an HDR radiance field from LDR views with different exposures.

📸 The method models the physical imaging process to generate novel HDR and LDR views.

🖼️ The color of the ray is predicted by feeding the radiance and corresponding exposure time into the tone mapper.

🏭 The proposed method is tested on both synthetic and real-world scenes.

---

Source: Training ChatGPT with Custom Libraries Using Extensions — Release

Summary
Release is exploring the power of embeddings, vector databases, and language models to create innovative product features. After experimenting with ChatGPT Plugins, the team is now leveraging embeddings and vector databases to generate prompts for ChatGPT.

Fact

🤖 Using ChatGPT Plugins, Release explored how to give ChatGPT outside knowledge from its training set.

📝 Instead of ChatGPT directly accessing their documentation, Release decided to feed snippets of docs to ChatGPT in the prompt.

🧠 Both the ChatGPT Retrieval Plugin and Supabase’s AI Documentation Search rely on generating, storing, and searching embeddings as a way to represent text, images, or other types of data in a numerical format.

---

Source: TSMC’s 3-nm Push Faces Tool Struggles

Summary
TSMC is struggling to meet Apple’s demand for 3-nm chips, as yield and tool issues are hindering its ramp to volume production. The foundry is racing with Samsung to be the first to offer 3-nm production for customers such as Apple and Nvidia in HPC and smartphones. TSMC currently dominates the foundry market, and is at present preferred over rivals Samsung and Intel for leading-edge nodes.

Facts

🤔 TSMC’s yield and tool issues on 3-nm production are hindering its ramp to volume production.

🏎️ TSMC is racing with Samsung to be the first to offer 3-nm production for customers such as Apple and Nvidia in HPC and smartphones.

💰 For decades, the foundry business has outpaced the growth rate of the overall semiconductor industry.

📈 TSMC dominates the foundry market, as for now it keeps the top spot, according to Mehdi Hosseini, senior equity research analyst with Susquehanna International Group.

💻 TSMC will ramp Apple’s A17 and M3 processors at the N3 node in H2 2023, as well as security-based server CPUs at N4 and N3.

---