Dave Deriso

Made an ultra fast DTW solver w Stephen Boyd @StanfordEng 🚀 Contributions: 1. Linear time 2. Continuous time formulation 3. Tunable regularization prevents overfitting 📄 rdcu.be/cT5dD 👁 Demo dderiso.github.io/gdtw/ 👾 C++ github.com/dderiso/gdtw 🐍 pip install gdtw

Not saying I miss the pandemic, but I do miss the hermitage. If coding agents were around back then, I wonder how much brighter this graph would be

I’m thinking a lot about code architecture. This is a higher level design process that takes a top-down view. Good architecture is intuitive. You know it’s good when it’s easy to determine the purpose of every file, where it fits in, and categorize it as a definition, utility, process, state machine, or manager. In contrast, if you can’t quickly categorize a file and visualize it in the tree, the architecture itself probably needs to be redesigned before any specific file is optimized.

Oxford neuroscientists making epic hardware in Silicon Valley 🦾🧠 Awesome work Andre @TheFrontalLobe_

This is huge. Remember how latent diffusion cracked the image generation problem? Since then, researchers have been trying to apply it to language through a variation called “discrete diffusion.” This architecture has been somewhat of a holy grail since it’s 10x faster and far more accurate than existing models. Current LLMs use autoregression, which case errors compound as the sequence is generated. Hence GPT will start to hallucinate or veer off course later in the output. Discrete diffusion does not suffer from this drawback since all tokens are generated at once and then iteratively refined into a final product. That simultaneous action is also the reason it’s so much faster and scalable. @volokuleshov and @StefanoErmon quietly cracked the problem last year and have been putting in work to stabilize it and make it accessible to developers through Inception Labs. Congrats @volokuleshov and @StefanoErmon for ushering us into a new era 🚀🚀🚀

Delighted to find this NN based on my GDTW algorithm from #ICLR Deriving the gradient of a DP is often a tedious / pedestrian activity, even with autodiff. Love how they use DDN to do it implicitly github.com/anucvml/ddn arxiv.org/pdf/2303.10778 @sgould_au

One of my favorite moments from #deeptechweek @realGeorgeHotz @DivGarg9 @kylejohnmorris @stevewattsfrey @eraqian

Awesome longevity panel with @lindaavey @bryan_johnson @celinehalioua covering everything from testing / interventions to how society will change if we live to 120 🙌🏼 hosts @Andercot @sampritibh @timshi_ai #deeptechweek sf-deep-tech-week.com

Lost track of how many times I’ve seen AI written on healthcare slides at #JPM24 Also great to meet the man himself

🗽The AI for Public Good Hackathon @cornell_tech was a success! Here are the incredible projects by theme: 🗽1. AI for Gov github.com/dbieber/city-311 AI for routing complains to agencies @flore75711500 @theandrewsiah github.com/samob917/311Weekl… AI for summarizing 311 data @samob917 @d_r_farrell & Odaine Forbes github.com/xukai92/subway-ba… AI for load balancing subway @xukai92 & Victoria Liao 🗽2. AI for Open Data github.com/ankitshah009/NYC_… CitySavy: summarizes data on NYC Open Data @ashah0052 @noahpunintended @iamjdeleon, Edward Plata &Jon Chin github.com/EdwardPlata/AIGov AI for ranking NYC Open Data @nitanisha github.com/ryanhlewis/Crowds… AI for crowdsourced map data @rhardestylewis 🗽3. AI for Citizen Life github.com/Kaya-AI/ai-hack Demo: kaya-ai-demo.streamlit.app/ Agent AI that finds/applies to jobs @rolandgvc Zameer Raizada, Ojonimi Bako, Joshua Cohen github.com/hahnandrew/ai4pg AI for sealing criminal records @jenny_wang23 @azizalto @adityachavan & Andrew Hahn github.com/vishnuravi/openha… AI Agent for lowering medical bills @rahul_sompuram @vishnuravi & Chethan Sarabu github.com/richabatra/nyc-ai… AI chat for metabolic health @datarequena & Richa Batra github.com/LearnQuantum/feyn… Generative AI for learning STEM @richa_lq 🗽 🙌🏼 to our rockstar cohosts and speakers @volokuleshov @Bieber @2plus2make5 @samgbafa @npparikh @tanzeemc @soumithchintala @Mascobot 🗽and 🫶🏼 to our Incredible sponsors @AnthropicAI @a16z @Techweek_ @645ventures