- TechTok Newsletter
- Posts
- LLM Training Revolution 🏆, Microsoft Exits OpenAI Board 💼, Check your competitors 🥇
LLM Training Revolution 🏆, Microsoft Exits OpenAI Board 💼, Check your competitors 🥇
Today's TechToks: Meta's multi-token LLM, Microsoft and Apple turned away from OpenAI borad seats, How Figma runs design critiques, Trending GitHub Repositories, Product Picks, and more!
TechTok by João Martins
July 10, 2024
Today’s Summaries
See all as stories in techtok.today
Today's Picks🍒
⚖️ Microsoft exits OpenAI board seats and Apple stops pursuing one
💡 Meta opens new and innovative multi-token LLMs for research, as opposed to standard single-token LLMs
🛸 An artificial life simulation engine for digital organisms and evolution
🤔 Understanding the reasoning of LLMs with geometry
✨ From Figma: How to run a design critique
▶️ From HuggingFace spaces: Paste a video link and get a summary
GitHub's Trending Repositories
🗜️ Data pipeline to extract graphs from unstructured data
🕹️ Simple Rust game engine focused on speed & modularity
📚 Self taught course on System Design
Trending Products of the Week
🏆 Find & research your competitors
🌌 Add animated backgrounds and stickers to your screenshots
🎯 Live Stream of VC Funded Companies
And 7 more products to automate writing, research your microSaaS ideas and much more.
Quick Note
Help us reaching more people and get your shout-out in our newsletter 🤩!
Copy and share your unique referral link:
https://newsletter.techtok.today/subscribe?ref=PLACEHOLDER
Each week we spend several hours curating and crafting the best of new products, trending repos, tech news, and engineering blog posts. Share TechTok and help others who want to stay up to date with the latest in tech.
✨ "Too long don't README"
GitHub trending repos of today
GraphRAG is a data pipeline that extracts structured data from unstructured data, by leveraging LLMs to transform unstructured text into structured knowledge graphs.
What can you use this for: extract insights, extract knowledge graphs, and apply RAG for LLMs on private data.
A quick-start is available at Azure’s Solution Accelerator package.
|
Bevy is a simple, data-driven Rust game engine focusing on speed and modularity, with a complete 2D and 3D feature set. Cross-platform, fast compile times, free and open-source.
What can you use this for: create games, rendering, and animation, with support to hot-reloading on asset changes: no restarts or recompiles required.
A quick-start is available at the docs.
A feature-rich web interface to the Stable Diffusion AI model for text-to-image generation, image editing, and more.
What can you use this for: Edit an image with text prompt— i.e. delete people from a picture, extend an image to increase the dimensions — and reuse community scripts, besides generating new images.
You can try a no-login sample at stablediffusionweb.com. For a local deploy, follow the docs.
Pongo offers MongoDB-like syntax on PostgreSQL, with strong consistency and JSONB benefits. It translates MongoDB API to PostgreSQL queries, leveraging JSONB for performance and efficiency.
What can you use this for: Use it to integrate seamlessly, migrate easily, manage nested data, boost performance, handle JSON queries.
For a quick-start, install via npm install @event-driven-io/pongo, and check this blog post.
Repos to learn something new
A guide to learn how to design scalable and reliable large-scale systems, available in 20 languages. It covers key concepts, design patterns, and best practices, providing sample solutions for common system design interview questions.
What can you use this for: Prepare for interviews with interactive coding challenges, study with Anki flashcards, and understand trade-offs in distributed system design.
You can get started following the guide on the readme.
📱 Product Picks
Curated products from Product Hunt and MicroLaunch
All-in-one tool that helps to research competitors, with SEO, marketing, pricing, product and many more insights.
Get 100+ daily leads of companies which received funding.
Grab attention of people scrolling their feed with animated screenshots.
🔁 BurstyAI
No-code AI workflow tools to automate content creation, SEO and engagement with over 30 quick-start templates.
🎙️ VoiceCheap
A web application that translates and dubs videos into 30 languages.
Research and validate you micro saas.
A user-friendly platform to launch projects directly on the blockchain.
🧠 MindPal
Build custom AI tools that automate anything
👨👨👦👦 Onyxium AI
A Hub for AI Tools.
🕵🏽 OHMNI
Anti-surveillance fashion from M.I.A, the activist rapper.
🤗 Hugging Face - Space of the Week
Check this trendy AI model online for free
Paste a Youtube link and receive a summary with a transcription.
You can edit the transcription before generating a summary. For longer videos, it’s recommended a local clone— Remove @Spaces.gpu from the code and run locally on your GPU, here’s a guide.
What can you use this for: rewrite short presentations by editing the transcription before summarizing, read a video you don’t intend to watch, extract text from audio in a foreign language and characters.
What would you like to see more often on TechTok? |
🧐 Daily Picks
Most impactful articles and news of this week on techtok.today
Meta released pre-trained models with a new prediction approach: multi-token, and it could become the new standard for developing large language models. It’s on HuggingFace. Allowing models to forecast multiple future words simultaneously, it promises enhanced performance and reduced training times— which could make AI more accessible and sustainable.
Microsoft dropped their observer seat on OpenAI's board and Apple has reportedly stopped pursuing one, amid growing regulatory scrutiny over antitrust concerns. Microsoft is their exclusive cloud partner, and UK and EU regulators have looked into their partnership. Now, OpenAI shifts to new approaches to engage key partners and investors, including regular stakeholder meetings.
Artificial LIfe ENvironment is an artificial life simulation program for digital organisms in ecosystems, simulating evolution. Each simulated body is a network of particles that can be enriched with higher-level functions, and can be understood as agents or digital organisms operating in a common environment.
A paper on exploring the reasoning abilities of large language models through their geometric properties. It demonstrates that: the density of an LLM’s self-attention graphs defines the intrinsic dimension of inputs to MLP blocks, and that a higher intrinsic dimension implies greater expressive capacity of the LLM. This relationship is seen in recent enhancement of LLM reasoning.
👩🏽💻 Engineering Blogs
Articles from engineering blogs of big tech companies
Focus on making critiques motivating, helpful, and enjoyable for the team. Here are six methods for you :
Standard critiques: Standardize the critique format, for when you’re looking for specific feedback.
Jams & workshops: brainstorms, Crazy 8’s , mood boards and etc., for when you’re early on in a project.
Pair design critique: Work in small groups of two or three, for problems that require more context and deep work.
Silent critiques: Everyone stays silent and adds feedback digitally, for when you’re looking for a large volume of feedback.
Paper (“print-out”) critiques : Print out work on paper and hang it up around the room, For ideas that are hard to navigate in a single file.
FYI critiques: Sharing quick context on a project without requiring feedback, for when you want to take advantage of the attention in the room.
