TLDR 2023-03-16

Inside GPT-4 šŸ¤–, Samsung's $230B factory šŸ­, Wikipedia's redesign šŸ‘Øā€šŸ’»

šŸ“±
Big Tech & Startups

Interview with OpenAIā€™s Greg Brockman: GPT-4 isnā€™t perfect, but neither are you (6 minute read)

OpenAI's GPT-4 is a significant improvement on its predecessor, GPT-3. It allows developers to prescribe its style and behavior more easily and produces more factually accurate statements. GPT-4's image analysis capabilities mean that it can caption and explain the contents of photos. The model shows nearly human-level skill in domains like calculus and law. It still shares limitations with GPT-3, such as hallucinated facts and basic reasoning errors.

It Sure Seems Like Amazon Is Making a New Web Browser (2 minute read)

Amazon sent a survey to its users asking detailed questions regarding what features they would like to see in a new desktop/laptop browser from the company. Users who completed the survey rated the importance of features like text-to-speech, extensions, browser data syncing across desktop and mobile devices, and blocking third-party cookies. Amazon's potential entry into the browser market comes at a time when the ad industry is preparing for a massive change. Google plans to eliminate third-party cookies in the Chrome browser, a primary tool used by businesses for consumer tracking in ads.
šŸš€
Science & Futuristic Technology

South Korea to build 'world's largest' chip center in greater Seoul with $230 billion investment (3 minute read)

South Korea and Samsung Electronics plan to construct an enormous facility to manufacture computer chips in the greater Seoul area. The plant will join the existing memory semiconductor manufacturing complexes, creating the world's largest Semiconductor Mega Cluster. It will cost around $230 billion and be located in the Gyeonggi province. The project will take around 20 years to complete.

Hereā€™s the new spacesuit that astronauts will wear when they return to the Moon (3 minute read)

NASA has revealed a new spacesuit for the Artemis program created by Axiom Space. The Axiom Extravehicular Mobility Unit (AxEMU) has improved mobility and safety requirements, including added joints and thermal requirements for the Moon's south pole. The prototype features a dark gray color but the real version will be white for thermal reasons. AxEMU will enable movements like walking and picking things up from the ground that were impossible in the older design. Pictures of the suit are available in the article.
šŸ’»
Programming, Design & Data Science

Design notes on the 2023 Wikipedia redesign (25 minute read)

The Wikipedia desktop interface has been redesigned. The last significant change to the interface was made in 2004. Since then, usability for new editors has been improved and the editing experience was updated. However, the general layout and reading experience remained largely unchanged until the redesign began in 2019. Devices have changed significantly since 2004, resulting in the need for a modernized interface. This article contains notes on the redesign of the Wikipedia desktop interface by Alex Hollender, who led the project.

The Multi-modal, Multi-model, Multi-everything Future of AGI (8 minute read)

OpenAI's latest artificial intelligence language model, GPT-4, has improved significantly from its last iteration, GPT-3. It can perform new tasks like math and help with taxes. GPT-4 is also considered safer, with 20-30% less hallucinations and unsafe content. It is a multimodal model that can read images and text, convert sketches of websites into code, and offer summarizations of images. Currently, access to the API is on a waitlist. The multimodal visual API capability is exclusively available only for BeMyEyes, an assistive app for the visually impaired.
šŸŽ
Miscellaneous

OpenAI checked to see whether GPT-4 could take over the world (7 minute read)

OpenAI allowed an AI testing group to evaluate the risks of its newly released GPT-4 AI model. The testing group found that GPT-4 was ineffective at autonomous replication. The potential for models capable of being an existential risk to human society is not a new concern. The AI alignment community, fearing that powerful AI systems with superhuman intelligence may be close, wants to mitigate all possible AI harms. The emergence of large language models such as Bing Chat and ChatGPT has given these efforts renewed urgency.

How Siri, Alexa and Google Assistant Lost the A.I. Race (8 minute read)

Siri, Alexa, and Google Assistant had more than a decade to become indispensable, but they were hampered by clunky design and miscalculations. The tech world is now gushing over AI-powered bots such as ChatGPT. ChatGPT has been used to handle complex tasks like coding software, drafting business proposals, and writing fiction. The excitement around chatbots illustrates how Siri, Alexa, and other voice assistants have squandered their lead in the AI race. Many big tech companies are now racing to come up with responses to ChatGPT.
āš”
Quick Links

Docker is deleting Open Source images - here is what you need to know (8 minute read)

Docker Hub users who have created an organization will have their account and all images deleted if they do not upgrade to a paid team plan.

SpaceX Starlink rolls out $200 per month ā€˜globalā€™ satellite internet package (2 minute read)

Starlink Roam, which has replaced Starlink RV, is available anywhere in the world for $200 per month.

U.S.Threatens Ban if TikTokā€™s Chinese Owners Donā€™t Sell Stakes (7 minute read)

TikTok says that a forced sale wouldn't address perceived security risks and that it plans to spend $1.5 billion on a program to safeguard US user data.

Mirrorful (GitHub Repo)

Mirrorful is a simple design system infrastructure that can generate colors and other design tokens and then import those tokens directly into apps.

In pursuit of lunar oxygen, firm discovers recipe for net-zero steel (2 minute read)

Sodium can be used instead of coal for manufacturing steel in a process that reduces direct carbon emissions to zero.

OpenAI co-founder on companyā€™s past approach to openly sharing research: ā€˜We were wrongā€™ (6 minute read)

OpenAI hasn't released any information on the data used to train GPT-4, its energy costs, or the specific hardware or methods used to create it.
šŸ’¼
Jobs

Sourcegraph - Security Engineer (Fully Remote, $136,000+)

Sourcegraph is hiring a Security Engineer to build world-class security into Sourcegraph's codebase, product, cloud, and on-premise deployments.

Sourcegraph - Head of AI (Fully Remote, $235,830+)

Sourcegraph is hiring a Head of AI to help create the most powerful coding assistant in the world for a company that specializes in web-based code search and navigation tools for dev teams.

bex technologies - Head of Engineering (Fully Remote, $90 - $120)

bex technologies is hiring a Head of Engineering to lead the engineering team for a logistics platform in the construction industry.
Get the most important tech, science, & coding news in a free daily email. Read by +1,250,000 software engineers and tech workers.
Join 1,250,000 readers for