The Daily Index — Monday, January 15, 2024 · TLDR Index

Across today’s newsletters

Today’s Index

TLDR

APPLE VISION PRO CUSTOMERS FACE A 25-MINUTE IN-STORE SALES PITCH

The Apple Vision Pro will be available from February 2 starting from $3,499. Customers waiting to buy an Apple Vision Pro may have to sit through a lengthy sales pitch that includes a 25-minute in-store demonstration on how to use the headset. More than a dozen demo units will be available at any one time at some of Apple's largest stores to accommodate the appointments. Customers will be able to purchase the Apple Vision Pro online without visiting a store to try the headset out.

appleinsider.com · 3 min

TLDR AI

OPENAI CHANGES POLICY TO ALLOW MILITARY APPLICATIONS

OpenAI has revised its usage policy to potentially allow military applications of its technologies, a change from its previous policy that explicitly prohibited such use.

techcrunch.com · 3 min

TLDR Web Dev

HOW APPLE BUILT ICLOUD TO STORE BILLIONS OF DATABASES

Apple iCloud has an extreme multi-tenant architecture with billions of databases. The service stores at least one database for each user using technologies like Cassandra and FoundationDB. Apple's architecture focuses on asynchronous processing, stateless architecture, logical resource isolation, and layer abstractions to handle scalability and reliability needs. Apple ran one of the largest Cassandra deployments globally, but it still experienced scalability limitations. FoundationDB, a distributed key-value store, solved its scalability issues. The article also explains other problems Apple solved in its journey to supporting exabytes of data and millions of queries per second in iCloud.

read.engineerscodex.com · 10 min

TLDR InfoSec

HYUNDAI MOTOR INDIA FIXES BUG THAT EXPOSED CUSTOMERS’ PERSONAL DATA

Hyundai India has fixed a bug that exposed customer data, including names, addresses, emails, phone numbers, and vehicle details. The flaw leaked data through WhatsApp links shared after vehicles were serviced. The links were automated without human involvement. Hyundai said systems get periodic reviews and updates and that it strives to safeguard data.

techcrunch.com · 3 min

TLDR Crypto

SCARAMUCCI SAYS GRAYSCALE ETF SALES HELPED FUEL BITCOIN DECLINE

According to SkyBridge Capital founder Anthony Scaramucci, Bitcoin’s sell-off after the much-anticipated Spot Bitcoin ETFs started trading is partially due to sales of Grayscale Bitcoin Trust shares. Investors are reportedly selling to book losses and shift to lower-fee alternatives. Cryptocurrency exchange FTX's bankruptcy estate is apparently also unloading, contributing further downward pressure on Bitcoin.

finance.yahoo.com · 2 min

TLDR

HOW THREADS WILL INTEGRATE WITH THE FEDIVERSE

The Fediverse consists of small social networks that can be hosted by an individual or company whose users can communicate and interact with users on other similar networks. This article details how Meta plans to integrate Threads with the Fediverse. Meta engaging positively with its user base is difficult due to its size. Integrating with the Fediverse will let Meta step back and focus instead on engaging with organizations that represent the interests of the Fediverse.

plasticbag.org · 38 min

TLDR

SOME HIGHLIGHTS ON THE STARSHIP SIDE OF THE ALL HANDS

Starship's second flight almost made it to orbit - if it had a payload, that payload would have made it to orbit. SpaceX thinks it has a good shot at reaching orbit with the third flight. It will then focus on achieving full and rapid reusability. There are two towers at Starbase so one can be upgraded while the other is used for launches.

twitter.com · 2 min

TLDR

NEW ANTHROPIC PAPER: SLEEPER AGENTS

Anthropic recently trained large language models to act secretly malicious. It found that despite best efforts at alignment training, deception still slipped through. The alignment training only made models look safe. Anthropic's research shows how standard safety training might not actually ensure safety and might end up giving people a false sense of security.

twitter.com · 3 min

TLDR

HIDDEN CHANGES IN GPT-4, UNCOVERED

Recent changes made to ChatGPT's GPT-4 interface significantly altered the model's web browsing tools. The changes prevent the model from directly citing quotes from webpages and limits viewing of full content. This article analyzes the underlying mechanism of GPT-4 web browsing, the hidden changes, and their implications. OpenAI is increasingly under scrutiny regarding copyright laws and the reproduction of content on the internet. These changes seem to be aimed at eliminating the issue.

dmicz.github.io · 14 min