Today we’ve lost a giant and a visionary: Prof. Daniel Kahneman has passed at the age of 90. The Princeton SPIA community shares our condolences with his family and all who knew him, learned from him, and loved him. bit.ly/4cy0arM

Our friends Bob Sutton and Huggy Rao—two professors who have been working with the d.school from the start—share their work with @HarvardBiz. bit.ly/3GHJi3h @work_matters @huggyrao @StanfordGSB

Netflix has shipped 5.2 billion DVDs over the last 25 years — here's a timeline of the most rented movies throughout those years.

🎇 Big news this week! @DagWorks is open for anyone to sign up. It's free to get started! 🍾 TL;DR: Build better data/ML pipelines. Faster. DAGWorks is your unifying platform layer for data & ML pipelines. 1/n

Here's how I think of open-source LLM vs top proprietary ones: 1) Open LLMs will improve at an accelerating pace. 2) However, their gap from the best commercial models will continue to *widen*. This is because OSS community is decentralized and chases different objectives, but big companies are able to concentrate their massive GPU & talent firepower at babysitting a single model, while also learning the best tricks from OSS. GPTs, Claudes, and DeepMind Gemini's will improve at an even faster rate. I assign < 5% probability that by the time OpenAI releases GPT-5 (let's say in late 2024), any OSS model can even truly match GPT-4 on all tasks that matter. I'd be very, very happy to be proven wrong! 3) OSS LLMs will always have much more diversity. Examples: - Domain-finetuned models for medical, legal, finance, etc. - Country-specific models that cater to local cultures and languages. - Unconstrained models for more creative use cases, like Wizard-7B-Uncensored.