Nibble #17
Meta's release spree, OpenAI pulls back, AppleGPT, fix your git fsck ups, GitHub Certifications and more
🙏 Namaste & welcome to Nibble #17 (yay! 17 weeks, consistency is hard)
If you are new to this place, don’t forget to Subscribe.
What’s happening 📰
🆕 Meta releases
Llama 2 - Zuck drops another bomb, this one is the best open-source LLM you might have ever seen. Since Llama was already good, this one is trained on 40% more data and has a double context window. And and, they said out loud that the quality of output, safety & helpfulness matters to us. They really did set a new bar for all these metrics with Llama2. [Detailed Analysis]
In light of it, PerplexityAI launched LLaMa Chat, to just brag about how fast it really is.CM3leon, a more efficient, state-of-the-art generative model for text and images (it performs well on super-resolution too, btw!)
👬 Zuck finds his MoneyPal1 with Meta and Microsoft expanding their AI partnership - So, Azure now allows you to deploy and fine-tune Llama 2-powered models. Making Azure a one-stop solution to build on AI.
💬 Remember when you wished that ChatGPT remembered at least something about you in a new conversation? So OpenAI releases Custom instructions for ChatGPT and now you can add preferences or requirements that you’d like ChatGPT to consider when generating its responses, starting with Plus users now (basically make a GPT talk formal, be your personal teacher, act cute, be helpful, split slang, always be brev - just anything you like without additional prompt messages. just remember that Brother Sam is always moderating you)
🤖 Apple is testing a ChatGPT-like AI chatbot and Apple fanbois are going crazy with the plain ol’ quote “Bro Apple’s feature releases are late but far better implemented than the competition”. We are of the opinion that we are not believing all this hearsay until we see Siri actually being useable other than setting alarms and reminders.
🆕 You will soon be able to enroll in the GitHub Certification Program to earn a certification of your proficiency with their tools. The beta version launches at GitHub Universe this November and would cost $200. (all set to pay $200 and get a GitHub Certification so that I know everything to get those 4 PRs in Hacktoberfest this year 🫡)
OpenAI holds back on GPT-4’s image recognition capabilities due to privacy concerns. As they moved towards multi-modal support, they started supporting scanning the images as part of the query and now it detects some people, raising some privacy questions.
What brings us to awe 🤪
AGI safety from first principles - So somebody on the internet in 2020 wrote about AGI safety, thinking from first principles and it’s a pretty good read and possibly you can even form your own opinions around it (this btw is even before the release of Chat GPT which makes it even cooler)
GPT4 & ChatGPT are getting dumber since March [Paper].
Yup! You heard that right, possibly we were able to poison them enough with our queries or maybe OpenAI just turned off some servers or took a step back maybe. (Eg: more than 95% drop in ability to detect prime numbers for GPT4 & more formatting errors are showing up in the generated code.)React95 is a React component library full of Win 95 elements.
We Put a Distributed Database In the Browser – And Made a Game of It!, the folks at TigerBeetle built a browser game to say out loud that TigerBeetle now compiles to WebAssembly and can be used directly in the browser.
- explains how DataDog’s 3-day outage in March caused them $5M and how a mere circular dependency can trigger sleepless nights.
LLMs open up a bunch of possibilities in science, but at the same time in history, we can now speed up the process of analyzing mysterious manuscripts and more. But before that
takes us through World History through the lens of AI, checking if any of the existing LLMs already have any biases on the present history or not.
TILs 🤯
You can develop inside the containers with VS Code Dev Containers (surprised by the flawless support & OpenSpec, it’s beautiful) - Detailed deep dive into WebContainers here by unzip.dev
The Coolest Library (not a package) on Earth, researchers store ice cores that hold the keys to Earth’s climate past and future.
So apparently can send in ANY number and WhatsApp will deactivate that account? (Hope they fix this by the time this Nibble is released. We know some of our readers well and they would be up to no good 😉)
What we have been trying 🔖
Magic app generator - Generate a full-stack web app in WASM and Tailwind in one click
onsheets.io - Convert your Google Sheets into REST APIs
shellcheck - find bugs in your shell scripts, quick & free.
pkg-size - an app that lets you inspect a node package size & deps tree in the browser itself.
GigaBrain scans billions of discussions on Reddit and other online communities to find the most useful posts and comments for you.
Builders’ Nest 💪🏻
ChromeStats - Compare and analyze Chrome extensions for competitor research, risk analysis, and growth tracking
Surprise Me - Get a different Web 1.0 website with every click
ElevateHQ - Automate any sales commission plan in minutes
RapidPages - Ship beautiful websites in a few mins (React & Tailwind landing pages)
ohshitgit: Git is hard: screwing up is easy, and figuring out how to fix your mistakes is fucking impossible
Meme of the week 👑
Off-topic reads/watches 🧗
📃 How to Do Great Work by Paul Graham - This is a pretty long read so get your maskabun and cutting chai ready before you dive in. Paul talks about a mix of choosing the right fields, working on interesting problems, working really hard and maintaining high morals (the simplest of words are the most profound init?)
🎧 600 working mothers told us how they navigate work, home, and their careers - We have featured The Ken’s Cost to Company before here and its latest episode is here again for a reason. One of the lines that stuck with us goes along the lines of “Children are the responsibilities of society and not just the parents.” Think about it.
🎵 永遠に光れ (Everlasting Shine) - and yes, in case you have watched Black Clover this should sound familiar.
📖 Do LLMs know what they talk about? A light dive into how LLMs are just letter/word manipulators.
Ponder-worthy words 🤔
"Most of the time you don't need more information, you need more courage."
- James Clear
Job Posts 💼
PS: This section is not sponsored, we hand-pick stuff by ourselves, from the week-long scrolling of the web and email reading. We will usually post Global or India-based opportunities.
Various Positions at Anthropic [SF/London] (the folks behind Claude)
SDE-2 at Sharechat [Bengaluru]
Data Scientist, Operations Machine Learning at Stripe [Remote]
Infrastructure Engineer at Ghost [Remote]
Product Engineer at Ghost [Remote]
SWE (Frontend) at Meta [Seattle, WA/ Remote]
What?!? that’s it? this one (Nibble) is smaller. that’s what she said.Well, we got some good feedback from our readers, we are trying to cut on content and build a rhythm slowly. So, we won’t spam you with unworthy stuff (until they become worthy). Do you like it this way? Let us know in the comments below!
From the Author’s Lap(top)
You can reach out to us using our nests on the Internet.
- Aashutosh - Rajasthani Baniya, has a craving for memes, loves *-wares (soft, hard, middle, ALL), can talk in italics, is a part-time cat-whisperer, and can tirelessly work towards the bloom doom of humanity.
- Pushkar, Gujarati Bihari Dikra, loves languages (programming and otherwise) and can be a part-time meme-lord, and a part-time neural network whisperer.
👋🏻 See you at the next one!
If you didn’t get it (hope you did), Check Satya Nadella’s Bachelor Alma Mater.