#48

Brick Attack, Bloom Filters, Fast LLM streaming, T2I Models, Gemma, ElevenLabs unifying, New Here, Dimming the Sun, SSH Signs, Array.with, VSCode++, Static Roots, uv, Kitto Katto, Make or Buy and more

Welcome to the 48th! (4 more to complete our deck? 🃏)
We are planning to shuffle the deck after 52 editions.
So, we want some feedback on what you would like to see more and less of.

📰 Read #48 on Substack for the best formatting.

🎧 Podcast version of this edition is available here → #48 | Recast

📢 Get access to bonus links and discussions with fellow Nibblers 👇🏻

---

🙌🏻 We have some Cosmos (an aesthetically pleasing and modern version of Pinterest) invites. The 3 can be claimed here using → this link. (two more are in our Discord 😜)

---

What’s happening 📰

• 💰 Sam Altman wants the US Govt to give him the green flag to pursue his $7T fundraising for AI chips from UAE (possibly 🤫).

• 📈 Reddit is doing its IPO but uniquely, apart from the regular IPO thing. They’ll allocate some shares to their most active and valuable users and moderators. (lol, it’s just like Crypto Airdrops da!).

  • Can you guess what’s the relation between this and the previous news? Sama of-course! Turns out Sam Altman has a 9.2% voting power in Reddit, even after he left the board in 2022 (Right before launching ChatGPT which BTW was also the reason why Reddit restricted free API access. What a guy!)

  • Google didn’t want to miss on the “front page of the internet” as well and announced a partnership (symbiosis?) of sorts where Google would give Reddit its new Vertex AI platform to integrate into its platform and in return get access to Reddit’s Data API to improve its search results, with the deal being around worth $60M per year.

• 😎 US Tech is catching a hold of “Bhai” Is this the start of an expansion in the Indian markets or is it just another publicity stunt pulled off by Carl Pei to attract Indian buyers ahead of their Nothing Phone (2a)?

• 🐢 Node.js announced their new Mascot, the Rocket Turtle (oxymoron?). We think they could have used Gemini DALL-E instead 😛

  

  It’s cute and has the 00s animation aesthetics

✨ AI Digest

• ⚡️ Groq, the new cool kid in town has made quite a name for itself by providing insane latency, thanks to its inference engine called the LPU™ Inference Engine (language processing units). It uses a compiler-first methodology which allows custom optimizations that balances flexibility with performance. Read more about it in this thread by @IntuitMachine.

  

  X-axis: Output Tokens Throughput (tokens/s)

• 🙌🏻 Google had a mixed kind of week (not the first time though):

  • Google’s released Gemma, a new family of “open” models of 2B and 7B sizes and they claim the 7B version to outperform the Mistral-7B on many benchmarks, most evidently Maths/Science and Coding benchmarks.

    

  • 👎 Gemini Advanced Image generation was a major letdown. Apparently it was trying so hard to not be racist that it actually became racist. Google says that they are working on a fix (given how long it took them to get here, expect weeks atleast).

• 🧚🏻‍♀️ Masayoshi Son (founder of Softbank) is looking for a $100B fundraising to build “Project Izanagi¹”, a chip (AI) company to bring AGI closer to us. BTW Son thinks, we’ll achieve it in the next 10 years.

• 🆕 Mistral secretly (literally a single discord message) released a new model called “next”. It is only available on the lmsys (the model feels to us like something between the mistral-medium and the mistral-7b model).

• ♟️ DeepMind has done it again!!! Their new Chess engine doesn't calculate a single move yet it reached a 2895 Lichess blitz rating. This is super-exciting because they reached the GRANDMASTER level rating without “SEARCH”². Imagine if we can convert other hard problems to this and maybe we can solve more things without searching?!?

• 🏞️ Text to Image Diffusion Model saw a couple of new releases this week:

  • 3️⃣ Stability AI announced Stable Diffusion 3, their latest family of text to image generation models that range from 800M to 8B parameters. Well, it is only in the waitlist right now (that asks for discord username hinting at a midjourney-style interface to access it). The released images look good but they are yet to release the technical report or even open source the model architecture and weights (which is quite unlikely right now given Stability’s recent stance with gatekeeping their new models and slapping a non-commercial license on them). Both the images below are the ones generated by SD3.

    

    Prompt: cinematic photo of a red apple on a table in a classroom, on the blackboard are the words "go big or go home" written in chalk

    

    Prompt: studio photograph closeup of a chameleon over a black background

  • ⚡️ Bytedance introduced SDXL-Lightning that achieves new SoTA in one-step/few-step 1024px text-to-image generation. It is distilled from SDXL combining progressive and adversarial distillation techniques (SDXL-Turbo only uses the latter). This leads to better image quality among the few-step generation models.

• 📡 Perplexity released some new model families in their API called sonar, namely sonar-small-chat and sonar-medium-chat, along with their search-enhanced “online” versions, sonar-small-online and sonar-medium-online. They also increased the context length of their models, reduced API limits, and integrated the new 2B and 7B Gemma models in labs.pplx.ai.

• 🎤 Just last week we had OpenAI release Sora and now ElevenLabs have started adding generative audio to all these generated videos. It’s all coming together guys! Just about time, somebody makes that immersive spatial audio, and boom, you are now a step closer to building industry-level cinematic videos from the tips of your fingers (quite literally by hitting those keys on your keyboard).

  • 🤝 Oh! And on that note, ElevenLabs is on fire, they are also now a part of the 2024 Disney Accelerator. (we’ll hear “Yamete Kudasai” in diverse voices)

  • Aaaand they also had a partnership with Perplexity to launch a podcast called Discover Daily. This takes the Discover section of Perplexity and forms a podcast out of it using ElevenLabs.

🔐 0x Digest

• 📈 After Sora’s announcement, WLD (Worldcoin) tokens surged by ~200% and could benefit backers from Three Arrow Capital & FTX.

• 🎌 Japan proposed a draft bill to allow VCs to fund Crytpo startups in exchange for tokens.

• 🚨 [SCAM ALERT] Earlier it used to be asset stealing due to random link clicks, but a new attack has spawned “Brick Attack”, and in this, it reassigns ownership of your wallet ☠️. So, not only your current assets are compromised, but all your future airdrops are not yours anymore either. 🥹

• 🚀 BTC hits an ATH³ in 14 countries this week. (no more asking “moon wen?”)

• 🎬 Arbitrum Foundation will fund a movie on how artists leverage AI & Crypto. It’s named “New Here”. It’ll be a series of 11 short films that will air-drop throughout the year. (Lights, Camera, Liquidate? HODL)

🛠️ Dev Digest

• ⚡️ V8’s Static Roots makes the whole V8 engine faster and brought a 1% bump in score on Speedometer 2.0.
  The JS Core Objects like undefined, true, and more are stored in a read-only heap and are called immovable immutable roots. Since they are part of every user-defined heap also, they need to be quickly accessible. This feature which was shipped in Chrome 111, achieves a way to correctly guess the memory address of these roots.

  So, what they do is they compile V8 to make a d8⁴ binary with fixed (static) root addresses. (for deep dive, read the blog linked on “Static Roots”)

• 🆕 Array.prototype.with(index, value) is now available across all major browsers. The method returns a copy of the array it's called on with the index set to the new value you provide. [Read More]

• 💔 Apple being Apple, confirmed that it might be breaking some PWA (progressive web apps) on the iPhones intentionally in the EU. (*sad DHH noises*). BTW, they said “And so, to comply with the DMA’s requirements, we had to remove the Home Screen web apps feature in the EU”

---

What brings us to awe 😳

• 🏀 While half the world is busy revolutionizing the software, Willson 3D-printed a prototype of Airless Basketball. ah! MKBHD got his hands on it and quickly explained how it’s surprisingly competent and full of holes.

• 🛍️ UAE will pay $35bn to Egypt to buy the town of “Ras el-Hekma” on its northern coast. (when Habibs go shopping they don’t get plots, they get towns)

• 🔌 Story of getting SSH Port 22: It dates back to the early days of the internet. Tatu Ylonen was testing his new creation that was a replacement for both telnet (port 23) and ftp (port 21) and asked for port 22 between them which was unoccupied at that time. As simple as that. Go ask for what you want and you shall get it.

---

Today I(we) Learnt 📑

• ⛅️ Someone is going to dim the Sun and it’ll happen soon. - The article is the best place to read it in detail, but here are some key points:

  • We might inject tons of sulfur into the atmosphere to reflect sunlight and make the planet cooler, which will fix the heating of the ball we live on.

  • But, it won’t solve all the issues that we have with Climate. It’ll not decrease the CO2 we produce and the sulfur damages the marine life along with CO2

• ✍🏻 You can sign commits with SSH keys (from git 2.34 release onwards), of course, it's not as good as (or recommended) GPG Keys, but you can. It’s as easy as setting gpg.format → ssh

• 🌻 Have you ever been in a “maybe” state? (We all are from time to time). But imagine a Data Structure that’s like that. Boom! We have “Bloom Filters” (it’s like a “maybe” Set) and here comes an interactive explainer about it, by our friend Sam. Why Bloom Filters, though?
  → Well with 0.0001% chances of a false-positive, you can save up to 82% of storage.
  → Real-world users of bloom filters include Akamai, who use them to avoid caching web pages accessed once and never again & Google's BigTable.

---

🤝 You have read ~50% of Nibble, the following section brings tools out from the wild.

---

What we have been trying 🔖

• 📧 Unspam.email: Email Spam Checker & Deliverability Test

• 🦙 raycast-ollama: Loving how we can conveniently chat with ollama models using the familiar Raycast UI.

• 💙 Apc Customize UI++: a VSCode extension that allows you to completely change the way it looks (including default native settings). [Shared by Kautuk]
  Here’s an example of how you can change your Editor’s font to JetBrains Mono and make it look like an IDE. Install & add this to your settings.json
"apc.font.family": "JetBrains Mono",

  

  VSCode with JetBrains Mono as APC Font Family

---

Builders’ Nest 🛠️

• 🦀 uv: An extremely fast Python package installer and resolver, written in Rust. (yes, a drop-in alternative to pip and pip-tools)

• 💨 incredibly-fast-whisper: Replicate has OpenSource versions of Whisper API that run 9x faster and 10x cheaper than OpenAI's Whisper API

• 🎫 minbpe: Minimal, clean code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization (at this point, karpathy is building the real “open” AI after he left OpenAI).

• 🎂 firstpr: GitHub’s PR is 16 years old today. Take a look at what your first PR on GitHub was.

---

Meme of the week 😌

Source: Tweet by @tunguz on 𝕏

---

Off-topic reads/watches 🧗

• 🍫 “Why Japan <loves> KitKat (Kitto Katto)?” from Chitthi by MoM, it’s almost part of the culture there. This article digs into “Why”.

• 👷🏻‍♂️ Make or Buy by Seth Godin, on how sometimes you should take a step back that buying a particular piece of your product is better than making it or not.

• ✍🏻 For the writers here, an article urging to write for humans and not be sadists, do some page pacing.

• 🤡 Apple is playing corporate tactics for you to buy into the most expensive device they sell, not the first time. But this one is rude considering the devices already cost ~$2500. You’re being misled.

---

Wisdom Bits 👀

“The main thing is to keep the main thing the main thing.”
― Kevin Kelly (kinda person, you should take life advices from)

---

Wallpaper of the week 🌁

🌌 Link to Wallpaper → wow.nibbles.dev

---

Weekly Standup 🫠

• Nibbler A will have a busy coming weekend (equivalent to a weekday). He’s been in the FAFO phase (again) in a new space he is trying right now, baking and breaking new stuff every day. (his way of “in my arena, not moisturized, but FAFOing”).

• Nibbler P has been trying to create latte art for his daily coffee cup but has been failing miserably. He has a deadline coming and will be toiling away the rest of his weekends.

If you liked what you just read, recommend us to a friend who’d love this too 👇🏻

1. For people who haven’t watched/read Naruto, Izanagi is the creator deity (kami) of both creation and life in Japanese mythology

2. A “searchless” chess engine can assess the current state of the game and suggest optimal moves without exhaustively analyzing potential outcomes via extensive search trees. It directly suggest the best move based on the current state of the game. Kind of like how a human has an intuitive understanding of the game through experience, rather than exhaustively analyzing potential future moves.

3. All-time high

4. V8's developer shell