TLDR 2023-12-26

Researchers from Apple and Cornell University quietly released an open-source multimodal LLM in October. Ferret is a system that can refer and ground anything anywhere at any granularity in an image. The model can examine a region drawn on an image, determine the elements within it that are of use to a user in a query, identify the elements, draw a bounding box around the detected elements, and answer questions about the elements. Ferret's release shows that Apple is starting to be more open with its AI work.

Humane will start shipping orders of its AI Pin device based on the date of purchase starting in Match. The AI Pin can respond to user queries using a range of AI services without forcing users to look at a screen. It features a projection system that can display things for users to interact with on their hands. Prices for the AI Pin start at $699 and it requires a $24 per month subscription for a phone number and cellular data.

Uniqlo has developed a new self-checkout technology that uses RFID chips to transmit product details and prices when shoppers are in scanning range. Customers just have to drop off their goods into a basket and the checkout machine will list the goods prices on a screen that tells them what to pay. The process takes around 30 seconds and has been highly popular with customers. The technology has been deployed at several locations, including the chain's New York shop on Fifth Avenue.

A small but vocal set of prominent skeptics in and around the emerging quantum computing industry say that the quantum computer revolution is still further off and more limited than many have been led to believe. Quantum computers have been touted as a solution to a wide range of problems and some of the more ambitious timelines have suggested that these machines could be impacting real-world problems in just a handful of years. However, there's growing pushback against these expectations as more is discovered about quantum computing.

This post presents a long and wide-ranging survey of different ways to make large language models go faster. The list is not completely exhaustive, but it can be used as a jumping point to learn more about interesting topics. It includes links to relevant papers and blog posts where possible.

qsv is a parallel, CPU-accelerated command-line program for querying, indexing, slicing, analyzing, filtering, enriching, transforming, sorting, validating, and joining CSV files. It features over 50 simple, fast, and composable commands. qsv is designed to handle very large datasets without specialized programming skills. It also supports other formats like Excel, Open Document Spreadsheet, JSON/JSONL, DataPackage, Apache Arrow/Parquet, and more.

This article presents the eight-step process that brought WP Engine from an idea to a Unicorn. It describes a formula for going from initial product to Product/Market Fit. The progression goes from personal fit to market fit, customer fit, building and shipping the SLC quickly, marketing and sales, retention-driven product development, prioritizing systematically, ruthlessly, and strategically, and managing your psychology. The roadmap isn't failsafe, but it is supported by many years of experience.

China has revealed new regulations to restrict online spending in the gaming industry. The news sent shares of several Chinese tech giants falling, wiping billions off their collective value. Online games will be required to set spending limits and in-game measures that could induce high spending will be banned. In-game measures that could lead to trading of virtual goods at high prices and large tips to livestreamers will also be banned. China is currently cracking down on the online gaming industry to reverse what it sees as a growing trend of gaming addiction among young people.

Early adopters are individuals who are willing to adopt a partially developed product, provide feedback, and trust founders when others might be suspicious of them - they are crucial in accelerating access to a broader addressable market.

Security should always be a priority, but not all companies have deep pockets - this article looks at cheap ways to implement security for young startups.

In every type of communication, there is a certain amount of relevant information (the signal) and there's irrelevant information (the noise) - good presentations maximize the signal and minimize the noise to end up with mainly relevant information.

If Apple's current schedule is maintained, the Apple Vision Pro may start getting stocked in stores beginning in late January.

Exposing a subset of SQL in an API makes the API more accessible, functional, compact, and easy to read for everyone, which can be very risky unless user input is sanitized using a white-list approach.

Wayfair employees were also encouraged to think of company money as their own and to negotiate prices accordingly.