Someone is going to make $100M in 2024 with a TikTok Shop. They'll probably have no inventory, no VCs and be 24 years old. This stat is crazy. 1 out of every 10 Americans are going to buy something on a TikTok shop this year. I don't know how to say this any other way, but… https://t.co/BJweuNfFhQ https://t.co/nWvmzo6dRb
— GREG ISENBERG (@gregisenberg) Oct 30, 2023
These AI baby chibi monsters got over 20k views I made them in 45 minutes Here’s my scary good playbook so you can animate your own cute creatures https://t.co/w1hPxbdz0B
— Dave Villalva (@DaveJWVillalva) Oct 30, 2023
What @TrungTPhan is describing here I've first learned it with the concept of Ontological Design "Everything that we design, in turn, designs us back" https://t.co/7NkxzyN7AU https://t.co/w1i4zpFic5
— Roberto Ansuini (@robansuini) Oct 29, 2023
A very nice visual explanation of how Gradient Checkpointing works is in this blog post by @yaroslavvb. https://t.co/IZLkEYf8nZ A brief summary from the blog on how GC stores some activations and uses partial forward passes for backprop. (Visualizations are from the blog) https://t.co/tM3pXBYioY https://t.co/0UL0ez9avJ
— Prateek Yadav (@prateeky2806) Oct 29, 2023
That’s what I am hearing also. Open source models are gaining in quality and obviously much, much cheaper. No wonder the players that raised $Billions are spending money to try and knee cap Open Source. https://t.co/eTiovrmrf2 https://t.co/AD45xAuUe0
— Bill Gurley (@bgurley) Oct 29, 2023
An outstanding story of a true hero. Kelly Johnson was the leader of the @LockheedMartin Skunk Works and the designer of SR-71 Blackbird. Thanks @johncoogan for an amazing video! https://t.co/d8TC8H4vSN
— Starson (@DrStarson) Oct 28, 2023
Almost every open-source LLM struggles with structured outputs. Here's a comprehensive, easy-to-follow guide on how you can use fine-tuning as a solution 👇 Our goal is to give every AI engineer the tools to do this easily without needing to learn PyTorch/heavy ML concepts.… https://t.co/sOs3p3UY5y
— Jerry Liu (@jerryjliu0) Oct 28, 2023
In spite of Sam Altman’s play for regulatory capture, looks like the Biden administration is going to call BS and issue an executive order correctly identifying the more prevalent risks as “concentrated corporate control” https://t.co/mpu3Us6ldx https://t.co/S5IOjPqKPp
— Kevin Fischer (@KevinAFischer) Oct 28, 2023
This summary of the various common definitions looks about right to me https://t.co/pmAsmglHqi
— Simon Willison (@simonw) Oct 28, 2023
If you’re an incumbent and believe AI is the next platform shift then the question is not “how do you infuse AI into every level of the stack”. Instead it is “how do we build a new stack that starts with AI”. The biggest mistake is treating disruptive tech as an ingredient.
— Steven Sinofsky (@stevesi) Oct 28, 2023
Here is a user interface I built to rapidly review LLM data for curation/eval for one of my clients @rechathq (sharing with permission). Sharing b/c folks said this is interesting https://t.co/xZCydn24J3
— Hamel Husain (@HamelHusain) Oct 28, 2023
🚨 THE PIRATE WIRES WEEKLY PODCAST IS HERE 🚨 Episode 20 — Tech Media Meltdown This week, the Pirate Wires crew joined Solana to discuss the tech press's reaction to @pmarca's techno-optimist manifesto, new developments in San Francisco's war on self-driving cars, the… https://t.co/PaEOuG8eOG https://t.co/7qwwojRIR8
— Pirate Wires (@PirateWires) Oct 27, 2023
The meta provided code for fine-tuning llama models using lora + fsdp works really well, minor adjustments to tokenizing might be needed depending on your datasets but I would just use their code https://t.co/hO0Kfq6bzC
— anton (@abacaj) Oct 27, 2023
RLHF is powerful; it lets us fine-tune LLMs to be more useful. What if we could do RLHF… without fine-tuning??? Excited to share Emulated Fine-Tuning (EFT)! EFT lets us “emulate” what we would have gotten if we did RLHF on a new model, without actually doing the RLHF! https://t.co/wRQU8BKjax
— Eric (@ericmitchellai) Oct 27, 2023
@Jake_Browning00 You are biased as hell. Some papers demonstrate something quite the opposite. For example, this one: https://t.co/VUSSMIXAra. Reasoning and planning in Minecraft done by GPT-3.5 One of the problems is that a lot of people don't know how to use LLMs efficiently.
— Leszek Bukowski 🧠💻🏛️👾 (@LeszBuk) Oct 27, 2023
Alright folks! "Luna Whisk" has landed in the Runway Studios Gen:48 finals!🚀 If you've got a minute, hop over and give us a vote. Luna promises endless interdimensional cuddles in return. 🐱 https://t.co/j0wsj2BW94 @runwayml @runwaymlstudios #gen48 #ai https://t.co/mO1ZkQM2vG
— Sid Audhinarayanan (@sidaudhi) Oct 27, 2023
🚨#PromptShare🚨 NIJI Hey folks! 🌟 Since you all loved the NIJI-themed #prompt I shared the other day for showcasing your pop culture heroes, I thought why not up the ante? Here's another one, but this time with a more epic, action-packed vibe. Dive in and let your heroes… https://t.co/1e45N7uuGV https://t.co/Y8V6t2wlTz
— Pierrick Chevallier | IA (@CharaspowerAI) Oct 24, 2023
I’ve been sleeping on Canva 😴 I just discovered how to easily stage product photos on Canva and enhance them with Stable Diffusion In just three simple steps,… https://t.co/q648GJOKmS https://t.co/Ig4KbY6GUG
— Salma - Midjourney & SD AI Product Photographer (@Salmaaboukarr) Oct 24, 2023
RT @sidaudhi: An AI generated short film using Dall-E 3 and Runway Gen-2, for Runway Studios Gen:48 https://t.co/djSkiGMjGB #ai #artificialintelligence #Runway #Gen:48 #luna #space #cat #3d #pixarlike
— Gopi Vikranth (@GopiVikranth) Oct 24, 2023
An AI generated short film using Dall-E 3 and Runway Gen-2, for Runway Studios Gen:48 https://t.co/djSkiGMjGB #ai #artificialintelligence #Runway #Gen:48 #luna #space #cat #3d #pixarlike
— Sid Audhinarayanan (@sidaudhi) Oct 24, 2023
https://t.co/Bhr8QBK38A https://t.co/MiksLbDjT1
— Marc Andreessen -- e/acc (@pmarca) Oct 23, 2023
✨ Early sparks of a working proto JARVIS 🤩 Daily Uber ride: (now reaching ~99% accuracy on this task with @MultiON_AI) Watch in audio 👇 https://t.co/n3WDU7NxhH https://t.co/AIeKzBDGga
— Div Garg (@DivGarg9) Oct 22, 2023
Magnificent shot of the Andromeda-Galaxy (M31) over pine trees. Image by Ritzelmut https://t.co/cCpka2mBoV
— Amazing Physics (@amazing_physics) Oct 22, 2023
The Experience Machine – fascinating read on the power of expectation and how the mind renders reality https://t.co/5VJb0K66Pa
— The Marginalian (formerly Brain Pickings) (@brainpickings) Oct 22, 2023
Microsoft is FULL steam ahead on regulatory capture: — Co-design the laws that dictate capability thresholds of AI (aka you can only compute so much before the government goes after you) — Co-design government license for “too powerful” AI (aka totalitarian state apparatus on… https://t.co/3A9neDdngO https://t.co/p2r59QxcDs
— JJ — e/acc (@JosephJacks_) Oct 22, 2023
@misha_saul https://t.co/JBOitMpRyy and the pro version: https://t.co/wDZcooTIvT Biased as we invested 5 years ago to build music AI free of training on other artists music. @SplashMusicCo
— Vinod Khosla (@vkhosla) Oct 21, 2023
OpenAgents - an open platform for using and hosting language agents in the wild. Includes three agents: - a Data Agent for data analysis - a Plugins Agent with 200+ daily API tools - a Web Agent for autonomous web browsing paper: https://t.co/6DqPR0ed7Z code:… https://t.co/Mne8ETinyn https://t.co/pNh4UaXNgm
— elvis (@omarsar0) Oct 20, 2023
Such a powerful AI use case. Without AI this would take FOREVER & cost a ton. Going to use this to translate my Midjourney course into multiple languages. Portuguese will likely be my first so I can offer the course w/ price parity for that market. https://t.co/DMWT5AKuP7
— Nick St. Pierre (@nickfloats) Oct 19, 2023
Me 🤝 OpenAI Writing long-ass few shot prompts with instructions in bullet points and simply trusting the demigod to adhere to them (it does) https://t.co/ZFJPojxAa1 https://t.co/wFqV7EAM6h
— TDM (e/flλ) (@cto_junior) Oct 19, 2023
Secret techniques in the room strengthening the waist and kidney;👇👇👇 - male over 40 years old must practice https://t.co/B8amHnMQgH
— Life Pad (@The_Life_Pad) Oct 19, 2023
Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection Outperforms ChatGPT and retrieval-augmented Llama2-chat on Open-domain QA, reasoning and fact verification tasks, https://t.co/G5u4hvv1Jo https://t.co/xXQX6HjxYv
— Aran Komatsuzaki (@arankomatsuzaki) Oct 19, 2023
You can get big performance gains for RAG if you fine-tune the LLM + retriever with retrieval in the loop. @manelferreira_ has an awesome post on the RA-DIT paper (@VictoriaLinML et al.) + impl in @llama_index. Check it out! https://t.co/51B2pXHDL0  https://t.co/NkgnVxskXA
— LlamaIndex 🦙 (@llama_index) Oct 19, 2023
This a great read We were lucky enough to have Andrew on our webinar talking about how he built this - it’s a great watch for anyone looking to build agents: https://t.co/TYAeWzC3B4 https://t.co/bRodRw6U6I
— Harrison Chase (@hwchase17) Oct 18, 2023
State of AI Report: very nice snapshot of the AI ecosystem across research, industry and (geo)politics (as usual each year :)). https://t.co/Jw2hNV52Oh https://t.co/mtV0GlMh71
— Andrej Karpathy (@karpathy) Oct 17, 2023
Want to build ChatGPT for your own data? LLaMa 2 + RAG (Retrieval Augmented Generation) is all you need! But what exactly is RAG? - Retrieve relevant documents from an external knowledge base. - Augment the retrieved documents with the original prompt. - Generate output text… https://t.co/CYg35IDxKm https://t.co/RuZGvpvYPG
— Shubham Saboo (@Saboo_Shubham_) Oct 17, 2023
THE TECHNO-OPTIMIST MANIFESTO part 1 “You live in a deranged age — more deranged than usual, because despite great scientific and technological advances, man has not the faintest idea of who he is or what he is doing.” — Walker Percy “Our species is 300,000 years old. For the… https://t.co/DgXm08Rgyb
— Marc Andreessen -- e/acc (@pmarca) Oct 16, 2023
The Techno-Optimist Manifesto -- please read and Ask Me Anything! Post questions as replies to this xeet. https://t.co/5ZKjiUG8de
— Marc Andreessen -- e/acc (@pmarca) Oct 16, 2023
Short Book Review 🌟Building An Event-Driven Data Mesh🌟 by @AdamBellemare through @OReillyMedia TLDR: a practical overview of data mesh architecture using event streams for inter-domain communication based on Apache Kafka and related projects; while also addressing… https://t.co/s1DOmRwbAi https://t.co/bupQUvoqfO
— Bilgin Ibryam (@bibryam) Oct 15, 2023
LLMTime: “Large Language Models Are Zero-Shot Time Series Forecasters” LLMTime encodes numbers as text and samples possible extrapolations as text completions. (NeurIPS 2023) Paper: https://t.co/SmHI5tTSHk Python GitHub / Open Source Code: https://t.co/ZGuXQNVN9e https://t.co/Lp6ZxJSqTK
— Carl Carrie (@🏠) (@carlcarrie) Oct 15, 2023
Google released 10 free courses on AI. No payment needed. Here are 10 FREE Google courses to master AI in 2023. [ Thread ] 🧵 https://t.co/zxCfiWssde
— Alamin (@iam_chonchol) Oct 14, 2023
🔴 BREAKING: META have developed a more advanced Generative AI model than SDXL for image generation! It's called ✨ Emu ✨ and it comes with a new technique called "quality-tuning". I think we are going to hear a lot about it in the coming months! 💡 Here's everything you need… https://t.co/xQZSRgrkfi https://t.co/qfIjhJaAon
— Javi Lopez ⛩️ (@javilopen) Oct 13, 2023
The bot that cares: the future of scalable, accessible, affordable psychotherapy as I predicted in 2012: https://t.co/8qOQns55V8 . The team at @limbic_ai is doing just that: https://t.co/sG4oQy0IU0 https://t.co/iBhhE65YZ7
— Vinod Khosla (@vkhosla) Oct 13, 2023
MemGPT: Towards LLMs as Operating Systems website: https://t.co/ABC0DBLhH9 abs: https://t.co/CYCTVENqC9 Inspired by operating systems, a fixed-context LLM processor is augmented with a tiered memory system and a set of functions that allow it to manage its own memory. https://t.co/FvXad9Lp14
— Tanishq Mathew Abraham, PhD (@iScienceLuvr) Oct 13, 2023
finally published our latest research on text embeddings! TLDR: Vector databases are NOT safe. 😳 Text embeddings can be inverted. We can do this exactly for sentence-length inputs and get very close with paragraphs... https://t.co/susxYN4ODe
— jack morris (@jxmnop) Oct 12, 2023
A bit about the use of GenAI in legal e-discovery. tl;dr super excited. At the @Everlaw Summit today, CEO @AJShankar showed off some of the new generative AI advances that have been in beta for some time already. These are ground-breaking efforts that really show how GenAI can… https://t.co/uOblBS3LwY https://t.co/2UzQv0TNpZ
— Steven Sinofsky (@stevesi) Oct 12, 2023
I've been digging in kaggle competition solutions involving LLMs and OCR-free doc understanding transformers, you find some really interesting solutions (and a lot of solutions using synthetic data) One that was interesting using 70B llama: https://t.co/IhXjDIxaSk
— anton (@abacaj) Oct 12, 2023
𝗪𝗵𝗶𝗰𝗵 𝗰𝗹𝗼𝘂𝗱 𝗽𝗿𝗼𝘃𝗶𝗱𝗲𝗿 𝘀𝗵𝗼𝘂𝗹𝗱 𝗯𝗲 𝘂𝘀𝗲𝗱 𝘄𝗵𝗲𝗻 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮 𝗯𝗶𝗴 𝗱𝗮𝘁𝗮 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻? The diagram below illustrates the detailed comparison of AWS, Google Cloud, and Microsoft Azure, created by Satish Chandra Gupta. They all use… https://t.co/PjUZw9v34P https://t.co/ZCVDTMabzS
— Dr. Milan Milanović (@milan_milanovic) Oct 11, 2023
Midjourney vs. Dalle-3 vs. Adobe Firefly Same prompt for all. 10 spicy examples. 1. https://t.co/364dNk4egV
— Alie Jules (@saana_ai) Oct 11, 2023
Apples for apples comparison of cost and performance: pgvector vs Pinecone it's not easy to figure out accuracy using Pinecone so we have detailed our methodology in the blog post - @supabase https://t.co/9PBuGcNlEr
— Paul Copplestone (@kiwicopple) Oct 10, 2023
Let's reverse engineer Disney's adorable, lifelike robot! I couldn't find a whitepaper, but this is how I think it's trained: 1. The emotional behaviors are curated by Disney animation artists, keyframe by keyframe. But it cannot be "rendered" directly on the robot because it… https://t.co/5REBYN2Hgi https://t.co/l62yeiNdGk
— Jim Fan (@DrJimFan) Oct 10, 2023
it’s official - I think GitHub Copilot is the first* generative AI product to publicly claim they’ve passed $100m ARR — enough to stand alone as a publicly listed company Whenever people ask me “is AI a fad” the biggest thing I point to is “follow the money”: - revenue, not… https://t.co/tVjCzkzdpZ https://t.co/8sxqWKrOX0
— swyx @ ai.engineer! (@swyx) Oct 10, 2023
Ton of people are still waiting for ChatGPT Vision... But the few that got it.. Are scratching their heads, waiting for their brains to reboot. 10 fresh examples:
— Borriss (@_Borriss_) Oct 10, 2023
If you take anything away from my talk, I hope it’s these three things https://t.co/5CbLvtm7Ll
— Eugene Yan (SF 7-10 oct) (@eugeneyan) Oct 10, 2023
Glad to be an advisor to this new initiative in AI for Science. https://t.co/HSEhbNY41q
— Yann LeCun (@ylecun) Oct 9, 2023
Huge. Disney just introduced their new robot. It was fully trained in simulation with Reinforcement learning just like OpenAI did with their robotic hand. The sim trained policy allows for rough terrain, traversal, and bump recovery. The whole project only took a year. https://t.co/ej4g6zLaLC
— Lior⚡ (@AlphaSignalAI) Oct 9, 2023
Improving Retrieval-Augmented LMs with Compressors Presents two approaches to compress retrieved documents into text summaries before pre-pending them in-context: 1) extractive compressor - selects useful sentences from retrieved documents 2) abstractive compressor -… https://t.co/iZTPucAGLx https://t.co/V0fbO8CfAf
— elvis (@omarsar0) Oct 9, 2023
Expected revenue growth in the next 5 years: Nvidia: +369% Tesla: +170% Amazon: +77% Microsoft: +77% Meta: +70% Adobe: +69% Alphabet: +67% Netflix: +59% Salesforce: +54% Apple: +27%
— Jon Erlichman (@JonErlichman) Oct 8, 2023
1. Jump into DallE-3: Skip the expensive designers; AI has your back. → Look up "Bing image creator". → Go to the website shown in the video. → Sign up or log in if you're already a member. https://t.co/81dkcepREN
— Abhishek (@HeyAbhishekk) Oct 8, 2023
When you lose momentum, it's hard to regain it. But here we go again, building a Personal Cockpit with AI Agents. I've built a chatbot connected to the internet and various APIs, linking it to multiple workflows and AI agents. Here's the first demo: 1. AI Newsletter Agent… https://t.co/qT2F16ymik https://t.co/dp7buZTGVU
— Muratcan Koylan (@youraimarketer) Oct 8, 2023
A cheat sheet of various databases in cloud services, along with their corresponding open-source/3rd-party options. Choosing the right database for your project is a complex task. The multitude of database options, each suited to distinct use cases, can quickly lead to decision… https://t.co/k1Y3hY9Hsl https://t.co/emuIwfOsp6
— Alex Xu (@alexxubyte) Oct 7, 2023
👉 Self Supervised Learning A curated list of awesome self-supervised learning Graphs, Talks, Thesis, Blogs, Theories, surveys, papers and a lot more. 🔗 https://t.co/lEnzZzsy9m https://t.co/84gvnBhgS7
— Dan | Machine Learning Engineer (@DanKornas) Oct 7, 2023
Wtf 😱🤯 AI VFX experiment: AnimateDiff + ControlNet This is some incredible work by @CitizenPlain Check thread 🧵 https://t.co/NEKyXH8iIM
— Linus (●ᴗ●) (@LinusEkenstam) Oct 7, 2023
@MKBHD @LinusEkenstam @omooretweets Another day, another massive AI fund. Visa has just unveiled their new $100M generative AI venture fund. The fund aims to target AI startups that could potentially redefine the future of commerce and payment systems using generative AI. https://t.co/PgHXP1lDBh https://t.co/gW2iFWGjrK
— Rowan Cheung (@rowancheung) Oct 6, 2023
Another massive week for AI. Huge developments from Canva, Google, Rewind, Apple, OpenAI, Harvard, LinkedIn, EdX, Stanford, Adobe, Epik, Arc, Meta, Anthropic, MrBeast, JPMorgan, Visa, Humane, Stability AI, and Spotify. Here's EVERYTHING you need to know (a thread):
— Rowan Cheung (@rowancheung) Oct 6, 2023
This has got to be one of the coolest pics I've made with DALL-E 3 so far. I'm still trying to perfect the prompt. I'll share it once I'm done. Then you can make these images for any character you like! https://t.co/V1Db1uEhwE
— Chase Lean (@chaseleantj) Oct 5, 2023
Oftentimes in RAG, you don’t really want to retrieve “chunks”; you want to retrieve atomic documents as context 📑 The upgraded document summary index in @llama_index does exactly this💫 - index + retrieve docs based on their summaries 📝 Guide: https://t.co/kWwT23fnPa https://t.co/GCuWVEMhc5
— Jerry Liu (@jerryjliu0) Oct 4, 2023
I built an agent that answers machine-learning questions. It's autonomous, and the best part is that I built the whole thing without writing a single line of Python code. Here is what I did and how I did it: Over a year ago, a friend and I built a site that publishes… https://t.co/8Q9KXswZcU https://t.co/VSqp7SJ8iJ
— Santiago (@svpino) Oct 4, 2023
Buckle up; this is going to be a good one! How to Generate On-Brand Stock Images using Stable Diffusion! 👇 https://t.co/YjmuowGig3 https://t.co/zWzYDM7veG
— Salma - Midjourney & SD AI Product Photographer (@Salmaaboukarr) Oct 3, 2023
Jamie Dimon’s monologue on our AI future: - 3.5 day workweeks - our kids will live to 100 and not have cancer - jobs will be eliminated but workers redeployed - “take a deep breath” https://t.co/aKaYYNqyvJ
— Emily Chang (@emilychangtv) Oct 3, 2023
There's a lot of fluff around ChatGPT + product management right now, but this is a surprisingly solid, in-depth article. I've been very active in exploring this space, but there's still a ton of ideas in here I hadn't considered. https://t.co/whJJ10IE4m
— Carl Vellotti (@carlvellotti) Oct 2, 2023
Building good RAG systems is hard, but building LLM-powered QA systems that can scale to large #’s of docs and question types is even harder 📑 We’re excited to introduce multi-document agents (V0) - a step beyond “naive” top-k RAG. Using multi-document agents allows our system… https://t.co/BQvdX5Fec1 https://t.co/rMH4jwoob9
— Jerry Liu (@jerryjliu0) Oct 1, 2023
Implementing LLaMA from scratch! 🙏 This implements a LLM in the style that Karpathy implemented nanoGPT Even though it focuses on LLaMA-1, it’s a refreshing code first read Perfect for a Sunday crispy read: https://t.co/y1fC10ogvp https://t.co/K5PKI4F5w6
— Sanyam Bhutani (@bhutanisanyam1) Oct 1, 2023
There are multiple ways to approach customer-segmentation projects. Damian Gil presents three in his clear, thorough guide: Kmeans, K-Prototype, and a combined LLM + Kmeans method. https://t.co/VPCsEpzc3L
— Towards Data Science (@TDataScience) Oct 1, 2023
