At the NYT's DealBook event, Nvidia CEO Jensen Huang says the US is 10 years to 20 years from chip independence and reaffirms Nvidia's commitment to China (@ianmking / Bloomberg) https://t.co/CUPTPO19vf 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/j8BNADFNrC
— Techmeme (@Techmeme) Nov 30, 2023
11-30-2023 An Interview with Roblox CEO David Baszucki About Advertising and AI https://t.co/vt6tlR4488
— Stratechery (@stratechery) Nov 30, 2023
Here are my favourite lectures on YT to learn distributed systems - - MIT 6.824 Distributed Systems (Spring 2020) https://t.co/dZZwnZ1i6T - UWaterloo CS 436: Distributed Computer Systems https://t.co/egrF6LhNt0 - lecture series by Martin Kleppmann https://t.co/HsvkRu1sKp
— Vipul Vaibhaw (@vaibhaw_vipul) Nov 30, 2023
Q&A with Sam Altman on why he was fired from OpenAI, returning to the company, improving the governance structure, the Q* model, lessons from the saga, and more (@alexeheath / The Verge) https://t.co/s02Zyi4kwe 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/XUXAef9Jla
— Techmeme (@Techmeme) Nov 30, 2023
Awesome round-up on synthetic data by @natolambert! Related: we found a massive increase in synthetic data use in 2023, enabling training on more diverse tasks w/ longer outputs. 🔗: https://t.co/guvIMARwJB https://t.co/xpFZ9HPXGa https://t.co/m9S5rBDYZd
— Shayne Longpre (@ShayneRedford) Nov 29, 2023
This is a good, and well written read on Nvidia and Jensen. Also, now I need a silicon component profile picture. https://t.co/biy3IB1RcO
— Ben Bajarin (@BenBajarin) Nov 29, 2023
You should get < 1 second an image on a M3 Max If only there was a super lightweight upscaler coming 🤓 Or you could do hi-res tile based upscaling with… stuff Some crazy UIs are incoming that work on your laptops all https://t.co/SVmvXco9z1
— Emad (@EMostaque) Nov 29, 2023
Before I wrote a book on the way Charlie Munger thinks I created a collection quotations which is available for free here. https://t.co/idH0DFKFPB My book on Charlie has a footnote and source for every quotation since the publisher was Columbia Business School Publishing.
— Tren Griffin (@trengriffin) Nov 28, 2023
steve jobs' advice to a young entrepreneur https://t.co/NufMPg8WZT
— Z Fellows (@ZFellows_) Nov 28, 2023
One of the most powerful aspects of our RAGs project is that you can build not only top-k RAG, but also start to build advanced RAG with natural language. For instance: dynamically handling summarization questions 📚 (in this case, naive top-k would fail). Check out the video… https://t.co/5Hzxootsah https://t.co/kQOiQJDiff
— LlamaIndex 🦙 (@llama_index) Nov 29, 2023
🦜🫅AI Driven Consultant 🚀Great example of orchestrating LLMs to accomplish complex tasks - s/o Gil Fernandes Check out the architecture diagram below 🤯 Lot's of different sub-components GitHub: https://t.co/iC2Prt85W5 Blog: https://t.co/agiMY947Qa https://t.co/gxPpP10FWL
— LangChain (@LangChainAI) Nov 28, 2023
It was an honor to give a guest lecture yesterday at Stanford’s CS330 class, "Deep Multi-Task and Meta-Learning"! I discussed a few very simple intuitions for how I personally think about large language models. Slides: https://t.co/NmusNTUVXb Here are the six intuitions: (1)… https://t.co/tPPo2eyAen https://t.co/qjmy7FGWWv
— Jason Wei (@_jasonwei) Nov 28, 2023
New deep dive on tokenization! 🔎 I've done a deep dive into the various aspects of tokenization, organized as bite-size chapters, with code and notebooks. Check it out: https://t.co/Lq3504Dskj
— Sumanth Hegde (@sumanthrh) Nov 27, 2023
How to build a DALL-E + GPT-3 fully integrated AI Agent that texts your friends back picture replies to their texts instantly without you having to do anything. You can do this in minutes with https://t.co/gdUk480IyT, no code. A 2-minute video tutorial. Timestamps: 0:00:… https://t.co/mVRrbDRc9l https://t.co/4jpaRIdbj4
— Steve Moraco (@SteveMoraco) Jun 28, 2023
Reason number 4247 LM evals are cursed: evaluation papers that are self-contradictory about how models are prompted! There is no official code for this task AFAIK, but the Big Bench implementation disagrees with both of the ones shown in the screenshot! https://t.co/2XCUPWbFhW https://t.co/gHNesQ74rX
— Stella Biderman (@BlancheMinerva) Nov 27, 2023
Bank of America Upgrades MercadoLibre $MELI Price Target to $2,000 With the stock up a cool 33% over the past month, now the big guys are about to start chasing it https://t.co/BUjDfdFgl1 https://t.co/fLWqBPOMoc
— Wolf of Harcourt Street (@wolfofharcourt) Nov 28, 2023
A big issue with trying to improve your RAG pipeline is that advanced techniques can require a ton of setup time. We spent this past weekend packaging 7+ advanced techniques in @llama_index, so that you can use all of them through a standardized interface - simply load in your… https://t.co/mJHkoZMNAc https://t.co/FnanyoQTfC https://t.co/Pave05z3Ld
— Jerry Liu (@jerryjliu0) Nov 28, 2023
🔥 https://t.co/6k4t020Nim
— Nat Friedman (@natfriedman) Nov 28, 2023
Today, @lightspeedvp is announcing that we’re leading a $35M Series A in @pika_labs, the company that’s redesigning the entire video-making and editing experience with AI. Founded by @demi_guo_ and @chenlin_meng, Pika is making it possible for anyone to create high quality video… https://t.co/VXkaYe7YJT https://t.co/LEQrvrmeC4
— Michael Mignano (@mignano) Nov 28, 2023
A profile of Jensen Huang, the Taiwan-born, US-educated Nvidia founder, who started the company in 1993, and a look at Nvidia's rise, built on GPUs and CUDA (@stephenwitt / New Yorker) https://t.co/T4UKh6XzsI 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/eugyXNpynC
— Techmeme (@Techmeme) Nov 27, 2023
A new Chinese app recently hit the top of US app charts: ReelShort. WTF is it? The app offers bite-sized English scripted video content. Like Quibi but with a catch: instead of Hollywood hitters, it is cheesy D-grade soap operas meets Hallmark films. The shows are about… https://t.co/NqHdcozmGW https://t.co/hQSf6mo34u
— Trung Phan (@TrungTPhan) Nov 28, 2023
Better than an AI wrapper raising millions to call GPT-4 over an API https://t.co/HVwJLmK9yK
— anton (@abacaj) Nov 28, 2023
A whopping 54% of all online shopping sales on Black Friday came from smartphones, accounting for ~$5.3B in sales. $AAPL $GOOG $AMZN $SHOP https://t.co/7kOwBjWE6A
— Beth Kindig (@Beth_Kindig) Nov 27, 2023
Self recommending! https://t.co/jnKkG0IpQI @VitalikButerin
— Marc Andreessen 🇺🇸 (@pmarca) Nov 27, 2023
10 Book Recommendations by Mark Zuckerberg: 1. Dealing With China "I've been personally interested as a student of Chinese culture, history, and language. I'm looking forward to reading Paulson's perspective on what China's rise means for the world." 2. Orwell’s Revenge "An… https://t.co/x0p8iszjew https://t.co/1nJplL5Zhm
— The Startup Archive (@StartupArchive_) Nov 27, 2023
Citations are super important when building any RAG app. Now, take it to the next-level by line-by-line citation matching between the LLM generated response and source context! Check out the thread below 👇 LlamaPack: https://t.co/8u978vSHwC Notebook: https://t.co/mhQSly2seu https://t.co/a5gHUkaMgN https://t.co/c08yTD67GK
— Jerry Liu (@jerryjliu0) Nov 27, 2023
New (synthetic) RLHF/RLAIF dataset 🚀 A team of researchers from @Berkeley_EECS just released Nectar, a synthetically created RLHF dataset using LLMs (GPT-4) for ranking and scoring example.🥇 🧶 https://t.co/tag3vxlYhk
— Philipp Schmid (@_philschmid) Nov 27, 2023
New RLHF 7B model: * Trained with RLAIF dataset (many sources of prompts + many models) https://t.co/wSu8l7SmHB * Released RM (fine-tune from Llama-2-chat) https://t.co/Hxvw5srIUh * Trained with different policy optimizer (APA) https://t.co/yqqMFidBKX * SOTA on MT Bench 7b (8.01)… https://t.co/MHE9OzmvDu https://t.co/b93x7W2iRM
— Nathan Lambert (@natolambert) Nov 27, 2023
The world in five years: 1. 75%+ white collar activities will be automateable 2. There will be no way to push unwanted content in people's face except by legal means 3. There will be no way to infer any truth from digital media 4. Media will be individual-tailored 1/2
— François Fleuret (@francoisfleuret) Nov 27, 2023
WE ARE SO BACK!!! 🔥 Finally! We are launching Magnific AI 🔥 The image Upscaler & Enhancer that feels like Magic 🪄 Link & details 🧵👇 https://t.co/CuKiyronnZ
— Javi Lopez ⛩️ (@javilopen) Nov 27, 2023
MercadoLibre's $MELI Record-Breaking Black Friday An 80% increase in Gross Merchandise Volume (GMV) marked the largest in the company's history. When analyzing the performance of the month of November to date, the platform showed an increase of 39% compared to 2022.… https://t.co/71HUJ1TRhT
— Wolf of Harcourt Street (@wolfofharcourt) Nov 26, 2023
This video by @pradip_nichite is one of the most comprehensive video tutorials we’ve seen on using RAGs 🔥 Learn how to easily build/customize an chatbot over your data (capable of advanced summarization) - even if you’re not a developer! 💡 👉: https://t.co/bqiBKUPBXL RAGs:… https://t.co/Whog4ePQyA https://t.co/DNUOKaeTAY
— LlamaIndex 🦙 (@llama_index) Nov 27, 2023
The advantages of this system over GPT-4, Claude, and similar models are not limited to specialized domains. It often represents the only feasible option, particularly when considering factors such as response time, scaling, privacy, and the speed with which new knowledge can be… https://t.co/PQulcbjulj https://t.co/vKsGPRkakz
— Artem Andreenko (@miolini) Nov 26, 2023
I've read a lot of books but have never felt so grateful coming across this one. Many great questions about happiness, meaning of life and more answered in the perspective of Adlerian psychology. Read this 🤝🏻 https://t.co/FIo6WiN1Cj
— merve (@mervenoyann) Nov 26, 2023
🦜Build your own RAG with Mistral-7B and LangChain Using local models can enable chatbots over your data while ensuring your data never leaves your computer Great for enterprise or privacy centric use cases This blog post by Madhav Thaker is one of the best resources we've… https://t.co/zrQamzBe30 https://t.co/MMc9ZP6wC8
— LangChain (@LangChainAI) Nov 26, 2023
Peter Thiel describes a startup category that is continually underestimated In the clip below, Peter Thiel explains why he believes “complex coordination” is a company category that’s continually underestimated. “The question we always focus on is ‘can this company become a… https://t.co/zF8xXfxUQC https://t.co/00FYcL5VEc
— The Startup Archive (@StartupArchive_) Nov 26, 2023
I summarized Andrej's talk and added my personal anecdotes to write this curious beginner's guide on: - How LLMs work - How LLMs are trained - How LLMs can be tailored for each company - How LLMs can evolve into an operating system 📌 Here's the link, I hope you enjoy it:… https://t.co/N6sNNAhPUY https://t.co/41qEkE8hcP
— Peter Yang (@petergyang) Nov 26, 2023
Shane Legg, cofounder of DeepMind, explains the importance of adding Search to Neural Networks. Background: "Move 37" was a pivotal play by AlphaGo during the 2nd of its 5-game series against Go Champion Lee Sedol. AlphaGo placed a stone in a highly unconventional position that… https://t.co/L9ILypdDSZ https://t.co/aGIDchgkiN
— Jim Fan (@DrJimFan) Nov 26, 2023
Someone asked me if there’s any need for paper books on AI because things are moving fast and the book may be outdated before it ships. I like reading paper books (especially on flights). Some books are timeless and some are a bit outdated but fun to read. These are some of… https://t.co/2t3ThnLnAn https://t.co/fDNcbfT3VI
— Kris Kashtanova (@icreatelife) Nov 26, 2023
Shopify Merchants drove 4.1b in sales this Black Friday. Up 22% from last year 💪 Amazing job entrepreneurs. https://t.co/9W7kcEdynM https://t.co/eEgHyBXZUY
— tobi lutke (@tobi) Nov 25, 2023
Our RAGs project enables a broad swath of users to ask complex questions over their documents without having to write a single line of code ☄️ Check out this recent article by @khemiri_ranya - uploaded a research paper to RAGs to help with a school assignment 👇. As a bonus ⭐️:… https://t.co/IrLTq2KQnr https://t.co/5BfqyQ1vSD
— LlamaIndex 🦙 (@llama_index) Nov 26, 2023
A superb book, with a prescient ending. https://t.co/yIz97BSPuD
— Marc Andreessen 🇺🇸 (@pmarca) Nov 26, 2023
Right now building an LLM app still requires a decent amount of software engineering. With our @llama_index RAGs project (and OpenAI GPTs) I'm interested in how application development can be pushed to higher levels of abstraction, through prompting vs. pure code. imagining a… https://t.co/ykwR5JALzI https://t.co/Kde0MyODY6
— Jerry Liu (@jerryjliu0) Nov 26, 2023
Thanks @martin_casado! @Intel has been making AI more accessible through rich SW portfolio and diverse Intel HWs ! We also released the high-perf LLMs and high-quality datasets for LLM training! People can easily create their own Chatbot through Intel Extension for Transformers! https://t.co/E8jayalI1A
— Haihao Shen (@HaihaoShen) Nov 25, 2023
Talked to some of the "AI engineers" working on LLM based apps inside Goog and Meta Realised how far ahead is twitter discourse on AI is compared to what normies talk about (e.g. not knowing even basics of similarity seach and just barely understand prompting even months into… https://t.co/yser2JVFKT
— TDM (e/flλ) (@cto_junior) Nov 26, 2023
Discord faced a monumental scaling challenge with the exponential growth of Midjourney, one of its most populer servers. Midjourney quickly bumped up against the then-hard limit of 1 million members per server (or guild, in Discord’s term). This is the story of how Discord… https://t.co/mOGm6V7RWh https://t.co/6BF9RWDIwT
— Sahn Lam (@sahnlam) Nov 25, 2023
I have been consulting with a Hollywood studio, producer and director who has made great progress on a film that was guided primarily by AI. Most of my work was on MonomythGPT (Hero’s journey) AI. All actors and scenes is AI generated. Out in 2024. https://t.co/PGKsMU9eg6
— Brian Roemmele (@BrianRoemmele) Nov 25, 2023
This is pure gold 🤯 $12,880 worth of Storytelling Degree in just 5 minutes. Just watch this video. Kurt Vonnegut explained stories better than anyone I’ve ever heard. [Bookmark] https://t.co/Sj11dPbthL
— Alamin (@iam_chonchol) Nov 25, 2023
𝗙𝗿𝗲𝗲 𝗘-𝗯𝗼𝗼𝗸: 𝗣𝗼𝘀𝘁𝗴𝗿𝗲𝗦𝗤𝗟 𝟭𝟰 𝗜𝗻𝘁𝗲𝗿𝗻𝗮𝗹𝘀 This book is meant for people who, while dealing with databases, will not accept a black-box solution. After a quick review of PostgreSQL's core ideas, the book delves deeply into data consistency and… https://t.co/Bbdc07pDf2 https://t.co/GrDLrEAlWg
— Dr Milan Milanović (@milan_milanovic) Nov 24, 2023
Musicians have nothing to fear https://t.co/o4LXwAv50C
— Aaron Levie (@levie) Nov 24, 2023
Multi-document RAG is hard. The article below shows that the OpenAI Assistant API does much better on single documents vs. multiple documents. We have a ton of resources in @llama_index for modeling multiple documents, from most complex to least complex - check it out 👇 -… https://t.co/JxFwVc9TQD https://t.co/gHfCMNg2Ef
— Jerry Liu (@jerryjliu0) Nov 24, 2023
Generative AI for Beginners It's nice to see that Microsoft created a neat website to organize these introductory lessons on generative AI. They are easier to browse and study. The lessons cover topics that range from prompt engineering to building chat apps to working with… https://t.co/Qwzho6Czwd https://t.co/etLfVuzHi4
— elvis (@omarsar0) Nov 24, 2023
step back in time by a few decades and watch @stripe *live* for the first time ever. https://t.co/cLLKLiP5HV https://t.co/oTUF2q9cEg
— sam mcallister (@sammcallister) Nov 24, 2023
Based on the data, the OpenAI saga actually spiked interest in GPTs. This is from data on my GPT directory, which is currently the most popular worldwide. Links below of the top 10 community favorites this week:
— Rowan Cheung (@rowancheung) Nov 24, 2023
"Breathing Life Into Sketches Using Text-to-Video Priors" They distill their model out of a pretrained text to video model. Their model learns a transformation on the lines of the sketch itself! Clever https://t.co/2UCLrylrG6 https://t.co/STjfjgPxrU
— kache (yacine) (@yacineMTB) Nov 24, 2023
Google Bard makes viral posts on X. No more headaches thanks to the new update. Make a long post on X in 2 minutes (Free)🧵: https://t.co/PsiJGtn52j
— Florian Camiade 🗝️ (@FCamiade) Nov 23, 2023
Aitana is the first Spanish model created by artificial intelligence. The virtual model can earn up to € 10,000 a month, according to her creator, but the average is around € 3,000. How do you feel about it? Source: https://t.co/BRJn2zmLps https://t.co/3JAYTXh4hL
— Kris Kashtanova (@icreatelife) Nov 23, 2023
Create Netflix like hover effect on your cards with HTML and CSS 🚀 Code in the comments 👨💻 https://t.co/yADg5MdQJS
— David Mraz (@davidm_ml) Nov 23, 2023
New YouTube video: 1hr general-audience introduction to Large Language Models https://t.co/Bl4WNuNyFJ Based on a 30min talk I gave recently; It tries to be non-technical intro, covers mental models for LLM inference, training, finetuning, the emerging LLM OS and LLM Security. https://t.co/JHOa2mqjdh
— Andrej Karpathy (@karpathy) Nov 23, 2023
This large language model 101 by @karpathy is technically comprehensive and easy to grasp. https://t.co/tdtZCvKKY1
— Lenny Bogdonoff (@rememberlenny) Nov 23, 2023
Any recommendations for an API that can surface high-signal news and updates about a specific company?
— Ryan Hoover (@rrhoover) Nov 23, 2023
Build AI agent in minutes to turn podcasts into insighful summaries 🤯 Meet n8n, the ultimate NoCode platform for building LLM apps with LangChain and ChatGPT. With simple Drag & Drop, you can build an LLM app that can convert lengthy podcasts into concise summaries. https://t.co/w4kxkzi7sh
— Shubham Saboo (@Saboo_Shubham_) Nov 23, 2023
Immediately following @KTmBoyle's rousing Call to Arms at the Defense Ventures Summit, @johncoogan delivered the most complete, visually stunning History of Defense Technology I've ever seen https://t.co/y8aw9zxSVS https://t.co/WuaD7I8S1V
— Mike (@MikeSlagh) Nov 23, 2023
Prompting AI to build itself is probably a step towards AGI. Philosophical questions aside given the past week, we looked into a simple version of how this works. We built this in a very constrained setting through our RAGs project to create a RAG agent purely through natural… https://t.co/Xjf6rEdKmF https://t.co/IZVawZQ0mi https://t.co/VX5qi39UOR
— Jerry Liu (@jerryjliu0) Nov 23, 2023
Keith Rabois: "Most people will solve problems that they understand how to solve. Roughly speaking, they will solve B+ problems instead of A+ problems. A+ problems are high-impact problems for your company but they're difficult--you don't wake up in the morning with a solution… https://t.co/MBJG9HSFC3 https://t.co/nx4Fc2qwlx
— BUILD OR DIE (@BUILD_OR_DIE) Nov 23, 2023
Soft disagree with this pearl clutching tone. And since I know every one of these tech leaders well, I have to say that @sama is no different than most of the talented ones, which is to say, aggressive, sometimes imperious and, yes, self-serving. Wot? https://t.co/GmWcBGhpvQ
— Kara Swisher (@karaswisher) Nov 22, 2023
wtf is this, i love it https://t.co/NCDuGO7GDA
— Pedro Duarte (@peduarte) Nov 22, 2023
Entangled #fxhash https://t.co/SrD0oAQpUJ
— 𝕭𝖏ø𝖗𝖓 𝕾𝖙𝖆𝖆𝖑 (@_nonfigurativ_) Nov 22, 2023
Why is 𝗠𝗼𝗱𝗲𝗹 𝗥𝗲𝗴𝗶𝘀𝘁𝗿𝘆 so important in your MLOps Stack? Let's look again into the Machine Learning Model Training Lifecycle. 𝗟𝗲𝘁’𝘀 𝗿𝗲𝘃𝗶𝗲𝘄 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝘀𝘁𝗲𝗽𝘀: 1: Version Control: Machine Learning Training Pipeline is defined in… https://t.co/aSzLBDrh9K https://t.co/FGdSKZe5iL
— Aurimas Griciūnas (@Aurimas_Gr) Nov 22, 2023
Ooh, a new sampling method just dropped https://t.co/IAWOgtD0gC
— Hassan Hayat 🔥 (@TheSeaMouse) Nov 21, 2023
The "3D Cartoon" style in @RunwayML gen2 is insane. https://t.co/CQDVOW3Xj0
— AI & Design (@AIandDesign) Nov 22, 2023
Deconstructing RAG It can be hard to follow all of the RAG strategies that have come out over the past months. I created a few guides to organize them into major themes and show how to build multi-modal / semi-structured RAG on complex docs (w/ images, tables). Here's a few… https://t.co/R6KVvkOBS5 https://t.co/M53nA9tMge
— Lance Martin (@RLanceMartin) Nov 21, 2023
I’m excited to introduce RAGs - create a “ChatGPT over your data” purely with natural language 💬 Setup and use a full RAG pipeline over files/web urls in three simple steps: 1️⃣ RAG Builder: Describe your RAG pipeline in natural language (data, parameters like top-k, system… https://t.co/YqwORIo1Vf https://t.co/Zs6Y58fbpm https://t.co/0geisdw8rI
— Jerry Liu (@jerryjliu0) Nov 21, 2023
This is a really neat paper. Uses "implicit" CoT reasoning where the model is trained to see intermediate steps but does not need to produce them during inference, benefit being faster inference! In explicit chain of thought, the model generate reasoning before the answer https://t.co/bq7ksq19yK
— anton (@abacaj) Nov 20, 2023
Having RAG deal with complex embedded objects (tables, charts, diagrams) is a huge enterprise use case. Standard top-k RAG breaks apart on this data. The @llama_index team has invested in ~a dozen features/guides/videos to tackle this, and we're excited to present some… https://t.co/CXwZN6usEs https://t.co/LkLiKLeKXw
— Jerry Liu (@jerryjliu0) Nov 20, 2023
Microsoft launched the best course on Generative AI. The free 12 lesson course is available on Github and will teach you everything you need to know to start building Generative AI applications. Each lesson includes: - a short video introduction to the topic - a written lesson… https://t.co/T1DfeNqtii https://t.co/0k0mfrjlu1
— Lior⚡ (@AlphaSignalAI) Nov 20, 2023
An educated explanation of OpenAI's RAG based improvements. Cosine Embedding: Explanation: Cosine embedding involves computing the cosine similarity between vectors. In the context of RAG, this method can be used to improve the retrieval of relevant information by comparing… https://t.co/TDT2EOSWab https://t.co/TiwvglRhsN
— Carlos E. Perez (@IntuitMachine) Nov 20, 2023
OpenAI’s Misalignment and Microsoft’s Gain https://t.co/FZWaYaOjlI
— Stratechery (@stratechery) Nov 20, 2023
Sam Altman warned us: “Everyone thinks you can hire someone to do this." "That fails 100% of the time." A board of directors won't cut it. The secret to building a generational company, according to him? The founder doing this every day: https://t.co/uhQAmx7O9Z
— Brandon Arvanaghi 🐱 (@arvanaghi) Nov 20, 2023
holy based I am in awe at the precision of this speech. 20 minutes of pure, densely packed truth. extremely honored to have @KTmBoyle as a mutual. let's just go ahead and make her president https://t.co/ItXKgoJxtM
— bayes (e/acc) (@bayeslord) Nov 20, 2023
@elonmusk Most picturesque launch ever. Here's shot from a guy who went to his first rocket launch... and he captured this! From https://t.co/v4FtvstlZF https://t.co/ClbJOJb8Zz
— Steve Jurvetson (@FutureJurvetson) Nov 18, 2023
Everything you need to know about $SHOP's Investment Case, Prospects & Risks: https://t.co/fHuZ7N3iiF
— Brad Freeman (@StockMarketNerd) Nov 19, 2023
We have over 250+ guides to help you build simple-to-advanced LLM/RAG apps over your data. But if you’re looking for a unified resource, this guide by @nanonets is one of the best that we’ve seen 👇 It’s concise yet comprehensive, covering 12+ parts including: ✅ Data… https://t.co/HunnWK1hIo https://t.co/HfKw5ezXYM
— LlamaIndex 🦙 (@llama_index) Nov 19, 2023
THE DESIGN OF POSTGRES (1986) https://t.co/DDRnNfRekA https://t.co/DypYNkxjIM
— Phil Eaton (@eatonphil) Nov 19, 2023
The first-ever deep dive on the engineering culture at OpenAI, after long discussions with @E0M, one of the first engineers on ChatGPT. Regardless of board drama, this is stuff that distinctively sets OpenAI apart. How the eng team ships so fast: https://t.co/5s5Kfx5SS1 https://t.co/xXDo3rud68
— Gergely Orosz (@GergelyOrosz) Nov 19, 2023
What makes a company succeed, according to Satya Nadella? It's not product. It's not distribution. It's something entrepreneurs often underrate: https://t.co/1SGKtMWY57
— Brandon Arvanaghi 🐱 (@arvanaghi) Nov 18, 2023
🧩Multi-index retrieval templates Sometimes you want to or need to use information from multiple indexes in your application. There's two basic ways to do this: Route: given a new input, choose (or route to) the most relevant index and retrieve from it. This is most useful when… https://t.co/A6evz5FEtB https://t.co/snxpkuQoPG https://t.co/maTw796KXS
— LangChain (@LangChainAI) Nov 18, 2023
Google is launching a price transparency widget as part of Chrome. Basically, it will tell you the price of the item in the past to let you know whether you're getting the best deal or not. Too early to tell the ramifications (but coupon codes become even more important).… https://t.co/1LWEIWUUx4 https://t.co/2YAUovuFyM
— Moiz Ali (@moizali) Nov 18, 2023
One of the many fine books written by Keiron Le Grice [I think I own all the ones published so far.] Here is some more from this book [The Rebirth of the Hero] which is the one I would recommend to you as a starting point. https://t.co/7UUWe7Wjyf
— SophiaCycles (Béa Gonzalez) (@SophiaCycles) Nov 17, 2023
Scoop: There are about to be a lot more major departures of top folks at @OpenAI tonight and I assume Altman will make a statement tonight. But, as I understand it, it was a “misalignment” of the profit versus nonprofit adherents at the company. The developer day was an issue. https://t.co/eIE4zctv4X
— Kara Swisher (@karaswisher) Nov 18, 2023
5 Must-Read Books about AI Chips: 1. “Chip War” by @crmiller1 2. “The Hardware Lottery” by @sarahookr 3. “The Sentient Machine” by @amirhusain_tx 4. “Fabless” by Daniel Nenni and Paul McLellan 5. “Crystal Fire” by Michael Riordan and Lillian Hoddeson 🧵 https://t.co/NzCsJEazya
— TuringPost (@TheTuringPost) Nov 17, 2023
A Survey on LLMs for Code Great paper providing an overview (50+ pages) of LLMs for Code. https://t.co/3jiXCX5fkN https://t.co/IWv4ZmEOxG https://t.co/hdzdMDx2My
— elvis (@omarsar0) Nov 17, 2023
Pretraining Data Mixtures Enable Narrow Model Selection Capabilities in Transformer Models https://t.co/XYMBSaDbPo
— /MachineLearning (@slashML) Nov 17, 2023
One concept I formalized from reading @ShaneAParrish's Clear Thinking was putting together a list of exemplars who make up my personal board. Now Shane says you don't have to meet these people. Well, I met two of them this week...@philwinkle and @brianjlange My heart is so… https://t.co/cyD5ohJz35 https://t.co/BG4iAV0FwP
— Doz (@DozieAnyaegbuna) Nov 17, 2023
When I built gpt-chessbot, I was shocked how often the LLM would suggest an illegal move, I’d give it a list of legal moves to pick from, and then it would pick a different illegal move. https://t.co/lbsUeBgoEv
— // teej // (@teej_m) Nov 16, 2023
Fine-tuning a custom @cohere Reranker for better RAG Adding a reranker is underrated but crucial for improving retrieval: filter out irrelevant context, reduce LLM hallucination The @cohere reranker is popular. Thanks to @ravithejads we have a new guide 💖 showing how you can… https://t.co/D1rqishPSV https://t.co/4759g3sTDX https://t.co/bleJlBTIbP
— Jerry Liu (@jerryjliu0) Nov 16, 2023
This is so good. Live the "We don't regulate the C language because it can make malware" (though personally, someone should have stepped in when C++ added templates, but I digress). https://t.co/r9ZAY0c6YD
— Steven Sinofsky (@stevesi) Nov 16, 2023
I met @christinacaci back in 2008, when I bumped into her in a class at the @stanforddschool. I've learned so much from her, and really admired her journey over the last 15 years, from investor to operator to founder. There's a lot of learnings packed in this Episode 3 of First… https://t.co/PRBqlZo9H7 https://t.co/5XuqWp7UB1
— Akshay Kothari (@akothari) Nov 16, 2023
DeepMind just dropped Lyria, a state-of-the-art music generative model. Last year, I said 2022 was the year of pixels and 2023 would be the year of soundwaves. We are making great progress here! The most impressive demo is converting humming to a full instrument suite. I think… https://t.co/EQdTU13gLa https://t.co/oHmcn5mtBk
— Jim Fan (@DrJimFan) Nov 16, 2023
On this years bingo card didn’t have the @GoogleDeepMind x @TPAIN collab Wonderful work, what an amazing time to be alive 🚀 🌟 https://t.co/APSGAUYxff https://t.co/jWB83XirY7
— Emad (@EMostaque) Nov 16, 2023
extremely good podcast episode on neural networks designed to discover geometric equivariants love how @naturecomputes succinctly breaks down technical terms she uses few of my takeaways below https://t.co/FZI8bqFap7
— murat 🍥 (@mayfer) Nov 16, 2023
Want to compare RAG approaches on your own data to see what performs best? With high quality and low cost? Incredible work—paper and code—by Jon to automate this. Lots of great new ideas and strong results here. https://t.co/oOYabrjvOW
— Omar Khattab (@lateinteraction) Nov 16, 2023
😀Great tutorial @icreatelife ! Here is mine! The ability to blend video with the image is cool! Prompt for the image (#DALLE3) An enchanting forest scene under the gentle embrace of twilight. The atmosphere is serene, filled with the soft glows of bioluminescent plants and… https://t.co/59g5NpYlrw https://t.co/yFiaA6lOBF https://t.co/HWQnyKCFSj
— Learn AI (@LearnAI_MJ) Nov 15, 2023
Move faster. Slowness anywhere justifies slowness everywhere. 2021 instead of 2022. This week instead of next week. Today instead of tomorrow. Moving fast compounds so much more than people realize.
— Sam Altman (@sama) Jan 1, 2021
Fine-Tuning LLMs To Be More Factual -Learning from automatically generated factuality preference rankings significantly improves factuality -58% reduction in factual error rate when generating biographies -40% reduction when answering medical questions https://t.co/OfFitSuKLy https://t.co/9koigMOlds
— John Nay (@johnjnay) Nov 16, 2023
8 open-source tools for RAG implementation: 1. RAG on Hugging Face Transformers 2. REALM library 3. NVIDIA NeMo Guardrails 4. LangChain 5. LlamaIndex 6. Weaviate Verba: The Golden RAGtriever 7. Deepset Haystack 8. Arize AI Phoenix https://t.co/8pPH0qLV3q
— TuringPost (@TheTuringPost) Nov 15, 2023
The Severe Gap in LLM Capability wrt to Agent Environments A recent paper demonstrates the severe limitations of current LLMs when deployed as agents in complex, interactive environments like WebArena. Despite their impressive capacities in constrained domains, leading models… https://t.co/SpYJRMVKYu
— Carlos E. Perez (@IntuitMachine) Nov 15, 2023
NVIDIA basically compressed 30 years of its corporate memory into 13B parameters. Our greatest creations add up to 24B tokens, including chip designs, internal codebases, and engineering logs like bug reports. Let that sink in. The model "ChipNeMo" is deployed internally, like a… https://t.co/ivR61wJeKs https://t.co/maK7Kcc5nj
— Jim Fan (@DrJimFan) Nov 15, 2023
Feels weird, wasn't Satya just literally like 2 days ago arguing not to trust open source models like Meta's? Now they will be serving them for profit? lol cc @pmarca https://t.co/dMnvxW5btr
— Teknium (e/λ) (@Teknium1) Nov 15, 2023
Very interesting results from @GoogleAI The new Mirasol model both more powerful and more lightweight. Models like Mirasol focus on multi modal inputs like text, sound, and video inputs https://t.co/qsX2QcfANj https://t.co/kSRXek9P3P
— Passio AI (@passio_ai) Nov 15, 2023
Have been binge-watching the breakout session videos from OpenAI Dev Day for the past 3+ hours. Really great content. If you're not a developer, I'd start with this session with product builders from Shopify and TypeForm: https://t.co/xCdkCSlZFT
— dharmesh (@dharmesh) Nov 14, 2023
A solo founder created an Instagram scheduling tool. That tool is now making $74,000/mo. This is his story:
— Indie Hackers (@IndieHackers) Nov 13, 2023
1/n Logical reasoning is a pillar of intelligence, but parsing errors have severely hampered solver-augmented models. LoGiPT provides an elegant and effective solution. Rather than treating solvers as black boxes, LoGiPT opens the box to reveal the reasoning. By formalizing… https://t.co/ArTdDuYtTb https://t.co/FBfnSd48mk
— Carlos E. Perez (@IntuitMachine) Nov 13, 2023
As multi-modal use cases take off, more and more storage systems will need to account for indexing text and images. Excited to partner w/ @trychroma early on in this fast-moving space https://t.co/VrRksmG9OL
— Jerry Liu (@jerryjliu0) Nov 14, 2023
Great news for devs who were not at @OpenAI Dev Day, the breakout sessions are now live on YouTube! 🎉 Check out sessions on: - Maximizing LLM performance - The New Stack and Ops for AI - The Business of AI And more! Details below 👇
— Logan.GPT (@OfficialLoganK) Nov 14, 2023
Midjourney hit an Annual Recurring Revenue of $200M. And is valued at $1B. This is a company that has less than 100 employees (86). They’ve raised no funding. And don’t plan to. And was founded mid-2021. Founder David Holz has his priorities. And it’s refreshing to see.… https://t.co/TV7CvDH9lX https://t.co/cczb1HwS1S
— Parves Shahid (@parves) Nov 13, 2023
This is one of the most comprehensive AI reports that we've seen - with a 1500+ sample size, there's tons of good info about AI+LLM use cases, stages of maturity, and pain points: - 66% of companies have at least one AI use case live - Accuracy is #1 concern - RAG is 2nd most… https://t.co/DL7UcEIUH8 https://t.co/aw5ClNLiEr
— Jerry Liu (@jerryjliu0) Nov 13, 2023
Lumos by AllenAI seems to be really good. Like make everyone-put-oss-in-production good. Everyone tries to go too wide with agents, there's a limit to what you can achieve with api over closed LLMs. Now solve self hosting and inference and we ball hard. https://t.co/Whs2kH1dDL
— 🪔TokenBender (e/xperiments) 🪔 (@4evaBehindSOTA) Nov 13, 2023
Mirasol3B: A Multimodal Autoregressive model for time-aligned and contextual modalities Achieves the SotA perf on well established multimodal benchmarks, outperforming much larger models. https://t.co/m7X6zGqmMU https://t.co/Y8BZ965mJx
— Aran Komatsuzaki (@arankomatsuzaki) Nov 13, 2023
Anthony Alcaraz describes how you can address the limitations of RAG in this excellent article with a bold title: - hybrid approaches, - multi-step reasoning modules, and - ranking algorithms. Read more on @TDataScience: https://t.co/pYUt6qARG6 https://t.co/uEjVMcOVSl
— Leonie (@helloiamleonie) Nov 12, 2023
⛪️Adding Structure-Aware Retrieval to GenAI Stack A great post by Joshua Yu takes a look at improving retrieval performance by using the GenAI stack: @Docker + @neo4j + @LangChainAI + @Ollama_ai This stack is (1) fully local, (2) uses advanced retrieval methods that encode… https://t.co/flDcTX5HYP https://t.co/6tb89ZkOGN
— LangChain (@LangChainAI) Nov 12, 2023
It’s been a week since dev day, and we’ve shipped 10+ features + guides + analysis allowing developers to use and understand these new features ⚡️ Features + Guides ✅ Parallel function calling: structured data extraction, agents ✅ Assistant API Agents: use in-house or custom… https://t.co/Sc8OR7GWvq https://t.co/n2Bt7MEZwj
— LlamaIndex 🦙 (@llama_index) Nov 12, 2023
🎉Recent book in town (2023) with 🆓 OA PDF: Distributional Reinforcement Learning by Marc G. Bellemare, Will Dabney and Mark Rowland, MIT Press 2023 "focuses on the total reward or return obtained as a consequence of an agent's choices" 👉 https://t.co/Xz6dXIuL6u @mitpress https://t.co/FoSFt4ZIc1
— Frank Nielsen (@FrnkNlsn) Nov 12, 2023
Want to learn about how to fine-tune both the LLM and retriever for better RAG? Check out our webinar with @VictoriaLinML on the popular RA-DIT paper: https://t.co/mp5LeqhK1S Now that OpenAI has released token logprobs, it's potentially easier to implement the retriever… https://t.co/LFAJGZKjHk
— LlamaIndex 🦙 (@llama_index) Nov 12, 2023
I don't usually watch long YouTube interviews because they're usually fluffy and not a great investment of time, but every second of this @lennysan interview with @bchesky was gold and pushed me to rethink how I think about leadership and product. https://t.co/cK5dAC8j8Y
— Siqi Chen (@blader) Nov 12, 2023
Steve Jobs keynote for the iPod Nano is an incredible re-watch. Just toying with the audience. No crowd has ever been more hyped by the close-up of dad jeans. And then boom: pulls the Nano out of the tiny 5th jean pocket (!) and everyone loses their mind. Masterful. https://t.co/kxnpNZjJGt
— Trung Phan (@TrungTPhan) Nov 12, 2023
Google Deepmind released an album with "visualizations of AI" to combat stereotypical depictions of glowing brains, blue screens, etc. https://t.co/1iTKctTob0
— /MachineLearning (@slashML) Nov 12, 2023
Introducing MechGPT 🦾🤖 This project by @ProfBuehlerMIT is one of the coolest use cases of 1) fine-tuning an LLM, and 2) generating a knowledge graph that we’ve seen (powered by @llama_index 🦙). The end result is a system capable of understanding a diverse range of scientific… https://t.co/fJ6lBk7UeC https://t.co/8htV3GdaUi
— LlamaIndex 🦙 (@llama_index) Nov 11, 2023
One of the biggest challenges of using knowledge graphs is building it. Here we show an awesome research project using LLMs to construct knowledge graphs, allowing researchers to make conceptual connections between disparate fields (e.g. relate materials science to biology)… https://t.co/jMvgWpDwh4 https://t.co/SeJ2ohk21t https://t.co/bq4ZBXvhHK
— Jerry Liu (@jerryjliu0) Nov 11, 2023
Biggest misconceptions in AI: 1. Confusing skill and intelligence. They are orthogonal. General intelligence can be converted into skill at many tasks, but in reverse, you can achieve arbitrary levels of skill at arbitrary tasks without requiring any intelligence at all.
— François Chollet (@fchollet) Nov 10, 2023
AI in real time is 🤯 SDXL + LCM-LoRA https://t.co/FynZe1CZrP
— Kris Kashtanova (@icreatelife) Nov 11, 2023
My timeline is once again flooded by "10 GPTs that blow your mind"🤢. Let's rain on their parades! The first step to solve any problem is to recognize there is one. A recent paper systematically reveals GPT-4V's limitations: - Text prior knowledge overrules vision: GPT-4V tends… https://t.co/Xc3hJivqpv https://t.co/PMiiGncDh2
— Jim Fan (@DrJimFan) Nov 11, 2023
Is it just me or is everyone sleeping on the new copilot that just dropped? It works really well in my tests. It does so much more now: - Plan and write entire projects - Plan and write notebooks - Fix bugs, write tests etc. - Write complex terminal commands https://t.co/4hZ3SraduC
— Hamel Husain (@HamelHusain) Nov 10, 2023
🤯 A simple side-by-side demonstration of the power of Retrieval Augmented Generation in #genai. After a couple of minutes and several plugins, ChatGPT could not find Voila Davis' Bacon Number but a small @LangChainAI & @streamlit app backed by @Neo4j can do the job. https://t.co/DjPbzqsblW
— Adam Cowley (@adamcowley) Nov 9, 2023
"Geometry is the archetype of the beauty of the world" 🔷 Johannes Kepler 🔶 https://t.co/8isD2Yc2Ei
— Kazonomics (@kazonomics) Jun 5, 2017
What is Chain-of-Thoughts prompting (CoT)? We will discuss: - The origins of CoT - Chain-of-thought Lineage (Auto-CoT, PoT, GoT, ToT, AoT, etc.) - Their architecture - Why did CoT cause a wave of research https://t.co/V2e23FZpqm
— TuringPost (@TheTuringPost) Nov 10, 2023
The full story of the open-access Deep Learning course that Alfredo Canziani and I have been teaching together and refining over the last several years. https://t.co/WB301DkMif
— Yann LeCun (@ylecun) Nov 10, 2023
In a comprehensive tutorial, Rahul Nayak introduces a method that allows you to convert any text corpus into a knowledge graph by using the Mistral 7B model. https://t.co/fFtLIqaS2f
— Towards Data Science (@TDataScience) Nov 10, 2023
One of the better articles I've read over the years on the culture that is Google, warts and all. (This substantially matches my understanding from *many* conversations that do not have a publicly readable URL associated with them.) https://t.co/L5duzKFZWE
— Patrick McKenzie (@patio11) Nov 9, 2023
I’ll never quit. I’ll never stop. I’ll never feel sorry for myself. I’ll never blame anyone but me. I’ll never complain. Because even if I lose the battle I will never lose the war. You with me? 🔥⛵️ https://t.co/D258ZqLSjg
— Matt Higgins (@mhiggins) Aug 3, 2023
🦜📷Multi-modal LangSmith Tracing With OpenAI releasing GPT4-V, we've now adding support for images in LangSmith tracing! This means you can now see images as they appear in your LLM calls, making it easy to debug multi-modal RAG pipelines Example Trace:… https://t.co/YUVxQqWHCy https://t.co/N7TnNr54ty
— LangChain (@LangChainAI) Nov 9, 2023
It's an open question how much you can eventually replace existing LLM text workflows with multi-modal vision workflows Previously: Parse a web page (e.g. bs4) -> index text -> dump into LLM context Now: Screenshot a web page -> index images -> pass to GPT-4V (??)
— Jerry Liu (@jerryjliu0) Nov 9, 2023
