#39
MMMU, MidJourney v6, no AGI in 2024, LLM in a flash, Hyperloop One shuts down, Rusty love, Hallucinate, PWAs in 2024, Barnum Effect, Broken OAuth, Sama & Gates's blogs, useWhisper and more
👋🏻 Welcome to 39ᵗʰ Nibble and wish you all a Merry Christmas 🎄
This is the penultimate edition of 2023.
We tried to split the 2023 sum-ups between this and the next one. Take your time it’s a little bigger than it should’ve been.
🎧 Podcast version of this edition is available here → #39 | Recast
📢 Get access to bonus links and discuss with fellow Nibblers
🎄 If you are doing Advent of Code 2023, we have a discord channel for discussions.
What’s happening 📰
🪑 Massive Multi-discipline Multimodal Understanding and Reasoning (MMMU), a new benchmark designed to evaluate multimodal models on massive multi-discipline demanding and reasoning tasks is released. While the MMLU benchmark only had texts, this one also contained images such as charts, diagrams, maps, music sheets, and chemical structures. Even GPT-4V only achieves a 56% accuracy on this.
📸 Midjourney released V6 with in-image text (finally) and much more photorealism in generations. Check some of the comparisons on this Twitter thread. (v5 was already nice, v6 goes above and beyond, long live 3D women!)
Sourced from same thread. 💎 Talking about Midjourney it’s one of the rare gem startups you’ll find in 2023. They generated $200 million in revenue without raising any capital from VCs and with just 40 employees.
🤖 FYI, Sam said don’t expect AGI to be shipped in 2024, one more year to work in peace as a Software Engineer.
⚡️ Apple is making advancements in AI, they recently published a paper “LLM in a Flash” which shows a way to optimize LLMs to run on edge and in memory. This enables a way to load models twice the size of available DRAM with increased inference speeds. (funny they said “in a flash” considering their history, lol)
🚄 Hyperloop One is selling its assets and planning to shut down by EoY. The company was trying to cut transit time by throwing people (which they later shifted to cargo) into the vacuum at high speed (ofc, a crazy idea raised by Elon back in 2023). They tried for 9 years to make a production-grade train, but couldn’t. Although this soldier falls the idea raised by Elon still lives with another similar venture The Boring Company. Let’s see how it goes, TBC says that they can build one by the end of the decade.
🦀 For the 8th year in a row, Rust has topped the charts for being the most “desired” programming language. GitHub wrote a blog post around “Why so much love for Rust?” and turned it into a Co-pilot Ad. You can find more such posts over the internet boasting about performance, control over memory, and some other cool stuff in Rust. There are also some things to consider before you make a move that it’s HARD.
What brings us to awe 😳
🔗 Google’s OAuth is broken and it allows you to keep access to apps like Zoom & Slack even if you leave your organization (with one trick ofc). [Google fixed it is what we are hearing from some updates]
👗 ChatGPT can now answer the 2015’s famous “The Dress” visual illusion. [Reddit Thread] (eyes ain’t moat anymore you humans!)
Sourced from the same thread as above 💰 Why are things getting expensive?: A highly detailed and interactive infographic on how things across the world are getting expensive. (spoiler alert: people think it’s “food & drink” in the USA and “appliances & tech” in India, but both places are acing in the consumption of this expensive stuff aren’t they?)
🥴 For language nerds out there, Dictionary.com named “hallucinate” the word of the year 2023.
🔭 The folks at Runway ML announced Telescope Magazine, “An Exploration of Art, Technology and Human Creativity” which will be released in January (excited about this because generative image genuinely needs you to be well versed with the artistic forms to get good creations and this might be a way to make people more conscious about it?)
🕸️ DHH made the most reasonable (we think so) prediction for WebDev in 2024, Native mobile apps are optional for B2B startups in 2024.
TL;DR 2024 is going to be the year of PWAs.
The reasons are that Apple is now finally catching up with PWA specs in Safari, the cost of native app development was always a problem, then paying cuts to the app store and being at their mercy for approvals, then the EU always comes up with weird data laws.
It makes sense for folks to switch to PWA now and avoid all these issues.
Today I(we) Learnt 📑
🫣 Barnum Effect: when you attach high personal meaning to statements that are applicable generally (so now you know Astrology is the art of creating the Barnum Effect).
🎭 The emotional arcs of stories are dominated by six basic shapes. (same as what we covered about “Archetypes” in #35)
📐 What Every Developer Should Know About GPU Computing: A detailed guide on how GPU works beyond the “Oh a GPU? It’s just a device good at parallel processing”
🤝 You have read ~50% of Nibble, the following section brings tools out from the wild.
What we have been trying 🔖
🔗 better-commit: A CLI for creating better commits following the conventional commit guidelines.
🧪 Perplexity Labs: Perplexity’s labs now have
mixtral-8x7b-instructalong with their existingpplx-70b-onlinemodels and they are blazingly fast! (how?) 🆓 Oh! yeah, and Santa came early and said “HOHOLIDAY”. yup! get two months of Perplexity Pro free using this link.🔎 Fooocus is an image-generating software based on Gradio. The good things are it’s offline, open-source, and free & aimed so that you can FOCUS just on prompting
and making your AI influencers.♊️ Gemini Pro Chat: Chat app based on Gemini Pro API [GitHub]
Builders’ Nest 🛠️
🎙️ useWhisper: React hook for OpenAI Whisper with speech recorder, real-time transcription, and silence removal built-in.
⚒️ redom: Tiny (2 KB) turbo-boosted JavaScript library for creating user interfaces.
🤏 microagents: Agents Capable of Self-Editing Their Prompts / Python Code
Meme of the week 😌
Off-topic reads/watches 🧗
🔮 Serial predictor Bill Gates shares “The road ahead reaches a turning point in 2024”
🤥 “Yes” Actually Means “No”: The Curious Language of VCs: VCs are difficult to understand (just like my ex) but at least there’s a primer to understand their language now. (yay!)
🪖 95%-ile isn't that good: @danluu argues that reaching 95-percentile mastery in a skill is not that difficult and using a strategic approach inculcating feedback and iterations can be just as effective.
📣 What I Wish Someone Had Told Me by Sam Altman: Sama posted after a year on his blog (actually, two years because the last one was more of a recruitment call). Wonder how much of it is influenced by the November Coup. (you know how one starts prepping a resume and blogs, once they see they are replaceable, lol, he ain’t)
Wisdom Bits 👀
“Never, ever, think about something else when you should be thinking about the power of incentives.”
— Late Charlie Munger
If you liked what you just read, recommend us to a friend who’d love this too 👇🏻
Weekly Standup 🫡
Nibbler P met a few friends, did some work, and handled lots of logistical work. *Busier week incoming for him*
Nibbler A did ship some pre-holiday season codes to prod, now cleaning up his inbox for Christmas. As always he’s getting his hands dirty with new tech & fields in his free time. He was busy so didn’t catch up with animes, and AoC backlog yet.
can't post on main Twitter account because of a personal ban but I enjoyed this :)
becoming my favourite weekly treat