Tobie

Thesis case study drop! Read here: diana.lu/point-n-talk AI Is Smart Enough, Let's Just Point and Talk to it ;)

Making commercials and films with AI isn’t just exciting — it’s the future. That said, the hardest part isn’t the technology, it’s finding people you genuinely love working with, and sharpening your storytelling skills. Those two things will never be replaced by “clicking buttons.” The fundamentals that carry over from traditional filmmaking remain the same: 📕 Strong stories 🤝 Strong relationships If you treat these as secondary, you’ll eventually feel stuck and frustrated. But if you keep them front and center — valued, protected, and constantly nurtured — you’ll thrive in this new era. I’m deeply grateful to Julien Vallée and Google for giving me the opportunity to be part of this project. Here’s to many more. Director: Julien Vallée (@julienvallee) Composer: Blackpaw AI artists: Julien Vallée, Jordan Daniel Chesney Motion design: Kyle Harter, David Urbinati Flame artist: Mathieu Arvisais VFX artists: Le Jumper, Empty Frame (Jeremie Drapeau & Francois Careau) Designer: Kelsey Lim Sound design: David Urbinati Producer: McKenzie Hayes Lead producer: Angela Long Google Maps Brand Creative Senior Brand Marketing Director: Max Kaplan Group Creative Lead, Search & Maps: Daniel Chandler

If you're building a PDF RAG pipeline: Should you be using OCR and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 methods, or just 𝗲𝗺𝗯𝗲𝗱 𝗶𝗺𝗮𝗴𝗲𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 using late interaction models? This paper says the answer might actually be 𝘣𝘰𝘵𝘩. My colleagues at Weaviate released IRPAPERS, a benchmark comparing 𝗶𝗺𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱 and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 retrieval over 3,230 pages from 166 scientific papers. The setup: Take the same PDFs and process them two ways. For text, run OCR with GPT-4.1 and embed with Arctic 2.0 + BM25 hybrid search. For images, embed raw page images with ColModernVBERT multi-vector embeddings. Test both on 180 needle-in-the-haystack questions. 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝘀: Text edges out images at the top rank: 46% vs 43% Recall@1 But images match or exceed text at deeper recall: 93% vs 91% Recall@20 But text and image based methods actually fail on 𝘥𝘪𝘧𝘧𝘦𝘳𝘦𝘯𝘁 𝘲𝘶𝘦𝘳𝘪𝘦𝘴. At Recall@1: • 22 queries succeed with text but fail with images • 18 queries succeed with images but fail with text This complementarity is what makes 𝗠𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗛𝘆𝗯𝗿𝗶𝗱 𝗦𝗲𝗮𝗿𝗰𝗵 work. By fusing scores from both text and image retrieval, they achieved: • 49% Recall@1 (beating either modality alone) • 81% Recall@5 • 95% Recall@20 More in the video below 🔽 Dataset: huggingface.co/datasets/mteb/… Paper: arxiv.org/abs/2602.17687 Code: github.com/weaviate/IRPAP…

x.com/i/article/2036…

we’re looking for founding generalist to join our team at @rightfitso 🏂 people who figure things out on their own are usually the best people to work with in early stage startups. there’s no perfect playbook in the beginning. things break. priorities change. new problems appear every week. the people who thrive in this environment are the ones who don’t wait for instructions. they observe, they think, and they move. that’s the kind of people i want around @rightfitso. this is not a comfortable role. there’s no fixed lane. one day you’re talking to founders about hiring strategy + another day you’re evaluating a candidate’s work + another day you’re researching companies, understanding markets, figuring out how early teams think and build + working with your domain to improve rightfit. you will learn how hiring actually works at the highest level. how founders evaluate talent. how great engineers and designers think about their craft. how early teams make decisions when every hire matters. but this role is not for everyone. you need to have taste. you need to care deeply about good work. you need to be curious about different domains. you need to be able to work long hours when the problem demands it and stay calm when things are uncertain. most importantly, you need to have a founder mentality. the ability to take ownership, think long term, and treat the work like it’s your own company. now i am planning to move fast. extremely fast. the environment here is very honest, if things are working we double down, if they’re not we change direction quickly. what i want to build here is a small team of people who are hungry to learn, hungry to build, and hungry to operate at a very high bar. > people who can think independently. > people who care about craft. > people who can move without waiting. > people who think like founders and take real ownership. if you’re an amazing engineer / designer / marketer who also wants to understand the hiring ecosystem deeply and help build the future of how great teams find great talent, this role might be for you. we’re building rightfit with a very simple belief. “great teams deserve great people + great people deserve great companies to build with” and i want to build a team that reflects that belief from the inside 🏴