Patrick Collison

𝗠𝘆 𝗝𝗼𝘂𝗿𝗻𝗲𝘆 𝗳𝗿𝗼𝗺 𝗥𝗲𝗮𝗱𝗲𝗿 𝘁𝗼 𝗟𝗲𝗮𝗱𝗲𝗿 (𝘂𝗻𝘀𝗰𝗿𝗶𝗽𝘁𝗲𝗱!) As the CEO of RivetAI, a question I encounter frequently is, "Why this path? Why you?" Given my background is deeply rooted in fields outside of the traditional film industry, the question is not only fair but also crucial because it underlines a fundamental expectation from any startup: a deep understanding of its users' needs and challenges. My journey to RivetAI began far from Hollywood's glitz, so let me share a story not of where I am now, but of where I started. I was born and raised in a world where content was heavily censored, so books and television (often sourced from the black market) were my only gateways to the world. I devoured fiction...it was my way of immersing myself in stories that took me beyond my immediate environment. Beyond entertainment, these narratives broadened my horizons, teaching me not only English, Spanish, and French, but also the universal languages of empathy, dreams, and possibilities. They provided a glimpse into realities far beyond my limited surroundings, and completely shaped my life's trajectory by allowing me to dream of what might be possible. This is the MAGIC of storytelling. While numbers always dance in my head (a CS degree from Caltech will do that!), my calling lies at the intersection of storytelling and AI. I have always sought to merge my love for technology with my passion for creativity. I believe in the power of well-crafted narratives to move us, inform us, and make the world a better place. However, the current system can stifle that creativity. Pre-production bottlenecks drain time and resources, leaving creators with less room to focus on the magic itself. Quality content requires not just brilliance but also efficiency. At RivetAI, we curate high-quality data, train customized models on that data, and believe this targeted approach will empower creators by streamlining pre-production. Imagine – more time and resources to focus on what truly matters: crafting groundbreaking narratives! That's why I'm the CEO of RivetAI. Because stories matter. Because AI can be a powerful tool for storytellers. And because RivetAI is dedicated to creating a future where more stories, diverse and impactful, can come to life. #leadership #storytelling #futureOfFilm #filmMaking #ai #contentCreation #disruptiveTechnology #innovation

30

6 Comments

We're thrilled to announce that Epsilla (YC S23) now supports Claude 3.5, the latest LLM by Anthropic that’s set to outperform GPT-4o! 🚀🤖 How does Claude 3.5 really perform against GPT-4o? We've recorded a fun video where Claude 3.5 and GPT-4o go head-to-head, tackling the question: Compare GraphRAG with traditional RAG, grounded by 361 research papers on Retrieval-Augmented Generation published since 2024. Curious about which model reigns supreme? Watch the video and join the conversation! For more in-depth insights, check out the detailed comparisons here: Claude 3.5 Insights: https://lnkd.in/gYQBKqrz GPT-4o Insights: https://lnkd.in/gyq38cwA Which one do you think is better? Please share your thoughts in the comments. #Epsilla #Claude35 #GPT4o #AI #ML #Anthropic #LLM

51

I just released the April edition of my AI & the Web newsletter. Nathaniel Simard, founder & CEO of Tracel Technologies and inventor of the burn ML framework, spoke with me about how he got into ML, burn, Tracel and the future of AI. WebGPU, the successor of WebGL, will soon also be supported in Firefox and Safari. Small, powerful LLMs running right in the browser is not a distant future, but are the next big thing in cost-effective and privacy-friendly AI. Google is working hard on better WASM support in the browser that makes ML models running fast even on low-end devices without GPU. With Llama 3 by AI at Meta and Phi 3 by Microsoft AI we got some exciting new, small modells. Llama 3 was already ported to run in the browser and we'll see Phi 3 soon as well. MusicGen has been ported to the browser in transformers.js by Hugging Face engineers and developed an online Jukebox running right in the browser. #webgpu #webml #ai #ml #llama3 #wasm https://lnkd.in/drJCY7Ew

19

2 Comments

Humans have historically found the secret to some natural phenomena and put it on steroids. 💉 Consider the aerodynamic efficiency of a bird's wings – a concept we've amplified to create the marvels of airplanes and rockets. ✈️ 🚀 We're seeing something similar right with LLMs. The transformer approach to linking knowledge seems to produce an eerily similar result to what we do. I've seen LLMs hallucinate in a similar way to my 5 year old kid! 😵‍💫 We took the body of knowledge on the internet and created 100+ billion parameter models that hold information that any human mind can. 🧠 But there's more to the natural phenomena of intelligence. There's a few more unlocks that need to be discovered. Like the mechanisms behind novel problem solving. 🧮 It's only a matter of time – in my opinion – until we crack those, put those on steroids and then we're no longer in the pole position of the most intelligent entities. 🤖

3

6 Comments

5 hours long Lex and Dario Amodei, the co-founder and CEO of Anthropic, a company behind Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus LLMs, speak about how AI is shaping the future and how it could impact the way we live and work. They discuss how far AI can go. Amodei thinks AI can match human intelligence and even surpass it in areas like biology, where complexity overwhelms humans. But in other areas, like speech recognition or conflict resolution, AI might hit limits close to ours. One of the challenges is the fact of running out of quality data and the cost of scaling compute power. Enjoy the entire talk! https://lnkd.in/dRhpafzU

9

2 Comments

Recently I learned that LLMs show ability to acquire rare linguistic competence (e.g. "a beautiful five days" is acceptable, but "a blue five pencils" not so much), from a tech talk by Prof. Kyle Mahowald from UT Austin (thanks Dr. Steve Kramer for organizing the guest speaker series). A small, high quality dataset BabyLM was sufficient to study the effect of training data on the model's linguistic ability. By surgically removing different types of phases from training data and observing the degradation model's ability, it can be revealed which part of data is responsible for the particular linguistic ability. This is a great example of doing AI research without breaking a bank. As someone not familiar with linguistics, I take linguistic competence to be a linguistic "value function" or "reward function". Can similar emergence of rare competence (not in training data but model learned it anyway) be shown in other domains? Reference: https://lnkd.in/g4cAPUYk

16

1 Comment

Validation is always good, and this time it came by noticing that our in-house system prompt engineer at Plino (👋 Viola Bonesu) has the same background in philosophy as the system prompt engineer at Anthropic, creators of Claude.ai. In our collective company experience, to be successful at LLM prompt engineering it is critical to  1. have a good understanding of the architecture of the AI engine 2. frequently interact with the models 3. intelligently iterate on the prompts Viola is good at all of these, and has an edge on the last step, where as Amanda from Anthropic puts it: «So in philosophy, what you’re trying to do is convey these very hard concepts. One of the things you are taught is, I think it is an anti-bullshit device in philosophy. Philosophy is an area where you could have people bullshitting and you don’t want that. And so it’s this desire for extreme clarity. […] All of the terms are defined, every objection’s kind of gone through methodically. And it makes sense to me because I’m like when you’re in such an a priori domain, clarity is sort of this way that you can prevent people from just making stuff up. And I think that’s sort of what you have to do with language models. Very often I actually find myself doing sort of mini versions of philosophy.» If you want to experience the quality of Viola’s prompts in our generative-AI powered accounting finance & control solution, book a demo with us → https://plino.ai

21

5 Comments

The newest issue of The Pragmatic Engineer found a surprise uptick in eng hiring in the typically slow summer season - I'm curious to see how the macro environment impacts this uptick as we head into the fall, but I remain hopeful. I asked Gemini to give me the tldr of the article to share with my team (sorry, Gergely Orosz) and found it missed one of the most interesting pieces of the article. Gemini's highlights - Unexpected surge: Tech recruitment has seen an unusual spike in June and July, contrary to typical trends. Increased demand: Startups, especially AI-focused ones, are driving the hiring uptick. Global phenomenon: This trend isn't just limited to the US; Europe is also experiencing increased recruitment activity. Reasons for hiring: Companies are backfilling positions, investing in growth due to improved financials, and responding to increased VC funding. Challenges remain: While the job market is improving, finding qualified candidates and offering competitive compensation remains difficult for many companies. What I found most interesting - The nature of the backfills: the article finds that a lot of new headcount are actually backfills of folks that were "over-slashed" during rounds of layoffs. Not entirely surprising with the "trim the fat" and "rebuild lean" mentality, but interesting nonetheless. Either way, the uptick is encouraging (and don't always rely on Gemini's tldrs). #hiring #startups #VCbacked #softwareengineering

11

1 Comment

💬 Stuck on what to type next? Meet Inputron’s AI-powered fields - The new co-worker that gets your thoughts before you do! 1️⃣ Predict – Think faster, type smarter. 2️⃣ Enhance – Make your ideas pop effortlessly. 3️⃣ Suggest – AI suggestions to keep you on point. Ready to simplify your writing? Visit Inputron.com and see it in action! #AIPowered #GenerativeAI #WorkSmarter

2

Solitary LLMs are impressive, but several can work together as a Multi-Agent System to astounding effect and overcome solo LLM (GPT-4o, Claude 3, Gemini) limitations. Hear all about it in today's "Five-Minute Friday" episode. In a bit more detail: • In recent weeks, OpenAI's GPT-4o and Google's Project Astra showcased A.I. agents that engage in humanlike conversations and analyze real-time video. • Despite such impressive capabilities, single A.I. agents still struggle with many categories of complex tasks. • Thoughtfully linking individual LLMs together into a Multi-Agent System (MAS) enables them to work together, assign tasks, build upon each other's work, and deliberate to find solutions, in aggregate exceeding the capabilities of a single LLM on its own... often without any decrease in compute efficiency. • As an example, a DARPA experiment demonstrated 3 agents (named Alpha, Bravo, Charlie) collaborating to defuse (virtual!) bombs efficiently. • In an Massachusetts Institute of Technology study, two chatbots solving math problems outperformed a single agent through dialogue and updating each other's answers. • Microsoft developed a software-writing MAS with specialized roles (Commander, Writer and Safeguard; see the image for this post) wrote code 3x faster without sacrificing accuracy. • As always, there are potential risks, e.g.: illogical solutions cascading through the team, agents getting stuck in loops, and malicious exploitation. • Commercial interest in MAS is accelerating, with frameworks like AutoGen and Camel making it straightforward for you to get started with MAS today yourself. What in your industry could be automated or improved by MAS where single-LLM approaches aren’t sufficiently effective? 🤔 For more detail on all of the above, check out today's episode (#788)! The Super Data Science Podcast is available on all major podcasting platforms and a video version is on YouTube. #superdatascience #machinelearning #ai #llms #multiagentsystems

52

4 Comments

AI Innovation Taking Center Stage The recent news of /dev/agents raising a staggering $56M seed round at a $500M valuation is a testament to the transformative potential of AI-driven developer productivity tools. This massive funding round highlights: -The growing demand for AI agents that streamline workflows, enhance efficiency, and supercharge development teams. -How investors are betting big on technologies that redefine traditional software development processes. As someone passionate about leveraging AI for impact, this news reinforces the importance of continuous innovation. Tools like these aren’t just about productivity—they’re shaping how the world builds technology. Kudos to the /dev/agents team for this milestone! What are your thoughts on the future of AI in development? #AI #StartupFunding #Innovation #TechTrends https://lnkd.in/gXnd5DaB

4

Great talking about Cicero and our digital twin tech with the legendary Cal Fussman. The future of AI is human—experience ultra-realistic digital twins at HeyCicero.com.

22

3 Comments

Terence Tao on LLM There is too much hype, hucksters and dumb money poured into GenAI and LLM now. It’s refreshing to hear from expert voice for once. Terence Tao is a UCLA math professor, Fields Medal winner and is considered the world’s greatest living mathematician. I overlapped a year with him at UCLA in 1996 and I met him when he was TA for a Linear Algebra course I took my senior year. Terence recently gave his honest feedback on GPT-o1 here https://lnkd.in/gq_6bcwM, and was subsequently interviewed by The Atlantic https://lnkd.in/gv6fgqrB I will summarize this interview: 1. OpenAI o1 is a mediocre or incompetent research assistant. 2. It does not “reason”, is not a source of knowledge or ideas, but is a really useful glue. 3. We will always need humans and AI. They have complementary strengths. AI is very good at converting billions of pieces of data into one good answer. Humans are good at taking 10 observations and making really inspired guesses. I cannot agree more with the “Mozart of Math”. Especially on point number #2. LLM is very good at doing relational joins of non-relational unstructured data. Once you realize its true strength and what it isn't, LLM can be a very powerful tool. To read more of my ponderings on "AI" and other tech stuff, subscribe to my substack: https://lnkd.in/gtknApfw

43

9 Comments

LLMs are designed to make stuff up. In most applications this an acceptable risk, after all, people make stuff up too. But in regulated, risk averse industries (finance, healthcare, insurance, legal) even a small mistake can cause big problems. Manual evaluation and monitoring is a crucial yet often overlooked component to catching errors and mitigating risk. Automated oversight can help, but there's no such thing as "Goodbye Hallucinations" unless humans are firmly in the loop.

10

1 Comment

My epic co-founder Tim Cheung did an excellent job (as always) in his latest post for Factory AI. The article is about how manufacturers can leverage the goldmine they’re sitting on—their data. First, he clears some of the fog on today’s hot topics: AI, ML, Deep Learning, and Gen AI. Second, he shares six relevant use cases, each accompanied by a case studies for each. Check out the link to the post in the comments. Give it a read and let us know your thoughts as always. #Manufacturing #FoodAndBeverage

16

3 Comments

Today, Vincent Weisser and I are excited to share what we’ve been working on. Prime Intellect is building the infrastructure for decentralized AI development at scale. We aggregate global compute resources and enable researchers to collaboratively train state-of-the-art models through distributed training across clusters. https://lnkd.in/d39i9ux8 We’ve raised a $5.5m seed round, co-led by Distributed Global and CoinFund, with participation from Compound, Collab+Currency, Protocol Labs, and an incredible group of angel investors incl. Clem Delangue 🤗 (HuggingFace Founder), Dylan Patel (SemiAnalysis) and many others. After reading the Scaling Hypothesis in 2020, I shifted my entire focus to LLMs. I joined Aleph Alpha over two years ago, helping build a new, distributed training framework with state-of-the-art efficiency. Following endless discussions on the latest AI research, traveling together to ICML, NeurIPS and more, Vincent Weisser and I decided four months ago to go all in on Prime Intellect to build our vision of an open and decentralized future of AI. Seeing that even the largest organizations begin to train their models across multiple clusters and scale up to hundreds of thousands of interconnected GPUs, it became obvious to us that the open-source community must overcome the limitations of traditional compute infrastructure and find ways to train on thousands of distributed clusters across the globe. Learn more about state-of-the-art decentralized training techniques and our development plans on our blog: https://lnkd.in/dCRjQwfF Our masterplan is as follows: 1. Aggregate global compute (live in beta) 2. Enable globally distributed training across clusters 3. Collaboratively train open AI models in high-impact domains 4. Create a decentralized protocol for collective AI model ownership Our platform aggregating the most cost-effective GPU providers is live in private beta! Sign up for early access at https://primeintellect.ai/ Next, we'll build an open, decentralized training framework to enable collaborative model development across globally distributed GPUs. We’ve started assembling a world-class team across AI foundation model training, novel transformer architectures and distributed systems to execute on this mission. If you’re excited about what we are building, we are actively hiring for founding team members! https://lnkd.in/dxH8WPj8 I'm immensely grateful for the hard work of our early founding team Sami Jaghouar, Kemal Erdem, Johannes Weniger, Mario Sieg and contributors Leonardo (Openprest), Max H., @goth600 over the last months! Back to building 🫡

66

13 Comments

François Chollet is a prominent French software engineer and artificial intelligence (AI) researcher, currently serving as a Senior Staff Engineer at Google. Chollet is best known for creating Keras, a widely-used deep learning library that was released in 2015. In 2019, he devised ARC. The Abstraction and Reasoning Corpus (ARC) is a benchmark designed to measure AI's ability to acquire skills and progress towards human-level intelligence. Learn more about it in this guest post with Jurgen Gravestein of Teaching Computers How to Talk. We need to continue to devise new benchmarks for our reasoning and agentic machines and tools. The ARC prize and context relating to AGI is of particular interest to me. Read more and check out some of the links in this article. Including videos by Machine Learning Street Talk (MLST) and Dwarkesh Patel.

20