#56
Meta Horizon OS, Atlas goes electric, Llama3, OpenAI.in, Wizard vanishes, AI playlists, Mario meets Pareto, ZKasino's rug-pull, Telegram's wallet, Worldcoin L2, Hard Centering, JS Naked Day and more
👋🏻 Welcome to the 56th! Please accept our apologies for the delay in the form of the lovely postcard from Nibbler A’s recent trip to Dubai attached at the end. now onto the edition…
📰 Read #56 on Substack for the best formatting
🎧 Podcast version of this edition is available here → #56 | Recast
📢 Get access to bonus links and discussions with fellow Nibblers on Discord
What’s happening 📰
🥽 Meta opens up its Mixed Reality Ecosystem to third-party developers, the same operating system that powers its Meta Quest headsets. ASUS, Lenovo, and XBOX are already working with it. Mark is as always pumped up about his open model to win in the Mixed Reality space against Apple’s closed one. Given how Apple is doing in the EU and Mark’s top-notch posting game, we’re gonna side with Mark for this.
🇮🇳 OpenAI has appointed Pragya Misra, a former WhatsApp and Truecaller executive, as its first employee in India. Pragya will lead public policy affairs and partnerships, leveraging her extensive experience to shape OpenAI's engagement with the Indian government and other stakeholders. (BTW she also has a fun YT channel. OpenAI is strong on vibes game, but where the hell is Ilya?)
✨ AGI Digest
🤖 The OG humanoid robot maker is retiring its trademark android Atlas with hydraulics and replacing it with a brand new all-electric version boasting reduced noise, lower maintenance costs, and improved energy efficiency. Atlas specializes in industrial inspection and they aim to run it autonomously for extended periods in partially unknown environments. All we want to know here is if it can shake a leg better than its ancestors did. [video attached]
🤪 Happy Llama 3 week! We saw some crazy model drops:
🧙♂️ The Wizard team at Microsoft released the Wizard LM 2 with the permissive Apache 2.0 License and just when we were about to test it, *poof* it had disappeared. They say they forgot some toxicity testing and hopefully, it’ll be back soon.
👀 Reka AI Labs recently showcased their Core, Flash, and Edge models, a family of multimodal language models with vision capabilities that you can try right now at their playground or API. The Core model comes pretty close to GPT-4V already and it is still training and is improving as you read this.
🎥 Microsoft Research Asia announced VASA-1, which converts single portrait photo + speech audio to a hyper-realistic talking face video with precise lip-audio sync, lifelike facial behavior, and naturalistic head movements, generated in real-time (on an NVIDIA RTX 4090 GPU ofc). No need to fret though because you may not see this anytime soon as the team says they won’t release it unless they are absolutely sure that it won’t be misused.
🦙 And the talk of the own, Meta finally dropped and released Llama 3, and boy did it take the internet by storm. They released the 8B and 70B models (while the much-powerful 400B model is still in training). Remember the WhatsApp and Instagram assistant we talked about in the last edition? They were using this new model all this time along with their image generation model (which you can use to craft new GM messages for your family group chat btw)! The 8B model is already the favorite local model (performing close to the llama2 70B) but it’s the 70B that truly shines, becoming the first open-weight model to beat Claude Opus in English!
Llama3 70B beats both Gemini Pro 1.5 and Claude 3 Sonnet | Source: Llama 3 announcement blog
🧱 Companies are more than excited to bring GenAI to their apps and users:
🎶 Spotify is happy with its previous AI DJ and is now bringing AI Playlists to Premium users, who can just describe what they want and the app will spin up a customized playlist for you based on that description. That’s one way to go from RAG to riches for Spotify!
💬 Huggingface released its iOS Chat App through which you can chat with OSS models hosted on Huggingface for free! This is Nibbler P’s fav item this week because they integrated Llama3 already and he happily unsubscribed his ChatGPT Plus subscription (until GPT5 drops haha).
🏞️ Meta's Segment Anything Model (SAM) can now run in your browser with/ WebGPU. If you want to know how useful SAM is, GO CHECK DINGBOARD RIGHT NOW!
📊 OpenAI added Projects to their API dashboard bringing a granular way to manage individual projects with the ability to scope roles, generate API keys, set billing and usage restrictions, and more. Thank God for this because before, it was a pain managing keys because they were linked across all your accounts, and if you forget to mention ORG-ID in your API request, you may get charged on a different account which was so confusing.
🔐 0x Digest
💸 Bitcoin halving happened and someone paid $700k to have their $0.97 transaction included in block 840,000. The miner rewards are supposed to be slashed to half after this “halving” but a lot of abnormal prioritized transactions skewed the first block’s fee to the higher end.
🐶 Shiba Inu raised $12M to build its own Layer 31 chain.
💌 Telegram announced some cool end-user stuff at Token 2049 built on their TON. Some of them are tokenized stickers, in-app wallets & creator tips.
🤡 ZKasino just did a rug-pull, redirecting $33 million worth of users’ bridged ETH to Lido. Think of it as another sign of being careful where you put your money.
🌎 Worldcoin (yup! Sama’s project) launched a blockchain designed for humans. They launched an Ethereum Layer 2 chain (joining the Superchain ecosystem) which is claimed to have ~0 bots (coming this summer), as the humans with verified iris using their orbs or “proof-of-personhood”, will get prioritized block-space and gas fee allowances.
🛠️ Dev & Design Digest
🆕 Updates from the 101st TC39 meeting by Hemanth. (better to read this, can’t shorten this post for you). BTW, you can test JS Signals, which are in Stage 1 on JSBin.
🔫 iOS 404, a curated list of missing web features in iOS.
💿 Our favorite PostgreSQL adapter Supabase is now General Availability phase.
What brings us to awe 😳
✉️ Why Can't My Mom Email Me?: What is the OpenPGP Web Key Directory and how does it have your email’s public keys?
🏎️ Finding the best Mario Kart builds using Pareto: Do you suck at playing Mario Cart? Now you can pick the Pareto Optimal build for your kart and still suck at the game. Can’t fix the skill issue, but hey, you at least got a new build.
🤞 Hardest Problem in Computer Science: Centering Things: Are you even a programmer if you haven’t failed at centering stuff on the web? (If you have OCD, just do not open this site. It just has way too many uncentered elements to give you panic attacks.)
Today I (we) Learnt 📑
🫵 A general rule to help you know when to use ”me” and when to use “I” in a list of names is to remember that if an action is being done to or for you, then use "me." Here, you're the object of the sentence. If it's you doing the action, then use "I." In this case, you're the subject of the sentence.
Source: Sketchplanations 🏃 PDFs can run JavaScript! (in limited functionality though)
🍃 April 24 is celebrated as JS Naked Day to promote websites to work with basic functionality without JavaScript enabled. The reason for choosing April 24 is that “4/24” is also the HTTP error code for a failed dependency2
👗 The cheaper the clothing, the more polyester it contains (mostly ofc!)
Source: Post by DLVermeer
🤝 You have read ~50% of Nibble, the following section brings tools out from the wild.
What we have been trying 🔖
🪪 What’s in a Name?: Been handling the second hard thing from “There are only two hard things in computer science: cache invalidation and naming things.” in our codebases recently
⚓️ localport: A tool to securely access your computer from any device, anywhere (think
ngrok, but with the terminal, filesystem, and your Jupyter server access)🌐 The Internet used to be fun: Exploring fun blogs that are made for fun
Builders’ Nest 🛠️
🃏 pgmock: An in-memory PostgreSQL mock server for unit and E2E tests
🚪 AI Gateway: A core infrastructure stack for building production-ready AI Applications
🍱 bentocache: Robust multi-tier caching solution for Node.js applications
🤡 faces.js: A JavaScript library for generating vector-based cartoon faces
Meme of the week 😌
Off-topic reads/watches 🧗
🪪 Anatomy of a credit card rewards program: The biggest takeaways for me personally → Credit card interchanges (swipe fees/bank fees) are the biggest source of income for Credit card companies and not the gargantuan interest rates on payment defaults. Therefore rich spenders bring in more to them than broke people who keep converting their payments to high-interest EMIs.
🦠 Two lifeforms merge in a once-in-a-billion-year evolutionary event: The last time this happened, Earth got plants. How would this turn out to be? We’ll pretty much find it out in a couple of billion years.
🏪 The Arc PMF Framework by Sequoia outlines three distinct archetypes of PMF which help you understand your product’s place in the market and determine how your company operates.
Wisdom Bits 👀
"Do something well, and that is quickly enough."
— Baltasar Gracián
Wallpaper of the week 🌁
🌌 Grab the week’s wallpaper at wow.nibbles.dev
Weekly Standup 🫠
Nibbler A was traveling & meeting interesting folks for the week, here’s a Postcard he made for you (DM for more pics and rishtas ofc). He didn’t get much time to read and write this week, of course, so ended up leading this edition to a delayed one. 🫠
Nibbler P is trying to get back to running and get himself into the morning-risers club. Would he be successful? Stay tuned to this section and find out soon.
If you liked what you just read, recommend us to a friend who’d love this too 👇🏻
Layer 3 chains are settled on Layer 2 chains, which extend some functionality and provide better throughput and lower fees, but eventually settle on Layer 1 chains (like Ethereum)
Thanks to our friend Stefan from Web Weekly for sharing these on his Twitter and newsletters.