#34
OpenAI Coup, Google's and Meta's new models, Apple RCS, new Font(), GPT-5 & AGI, Huberman Protocols, Custom GPTs
👋🏻 It’s 34th, we hope this one finds you well.
New here? Don’t forget to 👇🏻
🎧 Podcast version of this edition is available here → #34 | Recast
💬 Few important things today, we have an OpenAI fiasco going on & Cricket World Cup Finals (60fps link) will start in an hour max.
We wanted to cover the whole OpenAI thing in a separate Nibble, but since that was not possible we created → #tech-gossip, to discuss the technology behind gossip protocol interesting stuff. (unleash the tech “bhuaji” in you)
What’s happening 📰
✨ OpenAI confirms GPT5 is a work in progress and AGI is the eventual goal.
🛑 They also halted onboarding new ChatGPT Plus users due to higher demands, an ongoing DDoS, and more. Some users were found selling ChatGPT accounts on eBay (no kidding)
🎭 Also THE DRAMA → they fired Sam Altman, Greg resigned and Mira Murati is the new CEO (temporarily).
Considering the current state of this drama,there is a good chance that they’ll resolve all this and he’ll be back by the time you are reading thisA day later, they are calling him back because the major investors are urging the board that fired him to reverse the decision and the members of the board to be replaced.Some are saying Sam lied to the board, and some are saying Ilya planned it all, some are also saying Microsoft (Mr. Satya) didn’t like him, but who knows what happened? Their earliest believer Khosla Ventures also clarified that they want Sam back on board.
The only thing we know is job Security is a myth and Sam is either going to in his Steve Jobs Arc or Gavin Belson arc (also, was it not OpenAI that was also luring top Google researchers some time ago?). And it’s not hard to find out what Sam & Greg are up to they are tweeting constantly.☀️ Enough with OpenAI, let’s move to the rivals. So Google introduces Mirasol, a multimodal model for learning across audio, video, & text that decouples the modelling into separate autoregressive models to process the inputs according to the characteristics of their modalities.
And this 3B model beats their previous 80B Flamingo (thanks to @cto_junior for this insight). But the elephant in Google’s room still is:🎼 YouTube partnered with sister company DeepMind (alphabet’s D, ah! not that way) to transform the future of
deepfakemusic.
They announced a new AI Model called Lyria and two experiments (not open to all) so you can see how powerful they are:Dream Track – YouTube Shorts feature that lets you generate music and select things like Artists (eg: Narendra Modi)
Music AI tools – like a hacker dashboard but for music producer, director & composer.
That’s not all, they also took the opportunity to add two more things to YouTube.
🤖 They will soon start showing labels/notes on videos that use AI
✨ And remember SynthID (we covered in #23)? the thing that Google created to identify AI-generated images? They have implemented the same for AI Audios.
🤡 Talking about Deepfakes, Microsoft said “Regulations, my arse!” and released a deepfake creator, lol! Looking forward to more funny powerful politician memes. 👍🏻
▶️ Shipping scammy apps on Android just became harder as Google Play tightened up rules for Android app developers to require testing with at least 20 people for a minimum of two weeks before applying for access to production. Their global app review team will now also spend more time reviewing new apps. (a small price for devs to pay for an overall better app environment for the users)
🍎 Apple to support RCS messages from Android phones next year. Wait, what? So the Nothing experiments worked? Huge win for inter-device messaging if this becomes a reality.
🪽 Meta demos Emu Edit and Emu Video, its text-to-edit and text-to-short-video model foundation model, and these are some of the smoothest we have seen so far. Until it gets released, we'll stick with Runway (also, love that Meta uses cute animal names for its models).
🎥 Nepal bans TikTok and says it disrupts social harmony. Basically, they follow the mainland nation India. (no more “Badal barsa bijuli”)
What brings us to awe 😳
🎨 makereal.tldraw.com makes creating UI components a breeze by just drawing it on the sketchpad and letting GPT do its magic ✨. Here's a tweet with a demo
🆕 Vercel released a new font and called it Geist & Geist Mono.
GitHub followed suit and released Monaspace, a beautiful family of fonts, named after inert gases.📸 Opal Tadpole - a camera built for the notebooks (weird but yeah, we get it)
⭐️ Ship Shape by Canva Engineering - An interesting blog on how Canva does hand-drawn shape recognition in the browser in its draw tool.
💪 Huberman Protocols - A website to keep track of all Andrew Huberman's protocols shared in the Huberman Lab podcast
🧥 Exploring GPTs: ChatGPT in a trench coat? - It's been two weeks since DevDay and we have been seeing so many fun GPTs coming along. In this article, Mr. Simon takes us through the various experiments he did and shares his insights on creating better GPTs for yourself.
Our main takeaway?
1) Assume your prompts will leak. Don’t bother trying to protect them.
2) In fact, take that further: lean into it and share your prompts.
Today I(we) Learnt 📑
Remember those detailed AWS Infra Architecture Diagrams that make you go “That person must be a genius! This thing certainly goes over my head.” Well, it may be simpler than that and it is just a “tech greeble”1 in play.
💾 MongoDB reserves a default 100MB memory for sorting, and sometimes that is not enough (don’t you dare to ask me why).
So, you have a flagallowDiskUse
that can leverage disk, if memory is not enough in sorting.👐🏻 Fun fact: One of the designers of the font Nokia Pure is the designer of Comic Sans.
🤝 You have read ~50% of Nibble, the following section brings tools out from the wild.
What we have been trying 🔖
✨ BestGPTs - Top-ranked OpenAI GPTs
🆓 adsby Free Keyword Generator - Provide two details for 15 free tailored keywords, including exact, phrase, and broad matches.
📟 16bitGPT by
- A Custom GPT that creates images for you in the 16-bit graphic aesthetics. He has been killing it on Twitter with all these creations.👀 Screenshot to Code - a crazy app that kinda does something, if you want to prototype really quickly, but honestly, it’s not on par with humans in terms of adding complexities where not needed.
Builders’ Nest 🛠️
🔥 hotkey: Trigger an action on an element with a keyboard shortcut.
✨ inshellisense: IDE style command line autocomplete
👊🏻 ts-chan: Concurrency primitives for TypeScript and JavaScript.
⚡️ pacquet - experimental package manager for node.js (rewrite of pnpm in rust)
Meme of the week 😌
Off-topic reads/watches 🧗
📏 Measure Twice, Cut Once by Matt
🔢 Why the Human Brain Perceives Small Numbers Better - The Quant Magzine, apparently “4” is a neuron’s favorite number. And we might have found something revolutionary or as the article suggests it might be a complete miss.
🥷 Ask vs Guess Culture by Jean Hsu
😏 things that aren't doing the thing, a small note that says “JUST DO IT”.
Wisdom Bits 👀
"Quality is not an act, it is a habit."
— Aristotle
If you liked what you just read, recommend us to a friend who’d love this too 👇🏻
Weekly Standup 🫡
Both will be glued to TV after this goes live to cheer for India 🇮🇳🏏
Nibbler A had a decent week with Diwali celebrations in the first half and tons of work in the second. Everything was pretty normal until he woke up one fine Saturday, saw Sama was fired, grinded on 𝕏 for 5 hours, and figured out its politics or strategy, those two words led him to open chess.com, he won a few games, and just like how it ends, he kept playing lost some, still trying to recover from the lost ratings. Legends say he is still playing. (& losing)
Nibbler P was at his parents’ celebrating Diwali and then went off to his (our) alma mater to receive his bachelor’s degree at the convocation. Also, here’s a postcard from his visit.
“tech greeble” - Yes, I just now made that up. And I’ll be using it everywhere now :P
hey thanks for the shoutout.