Joseph Dormani’s Post

How can you transition to a career in responsible tech? 🤔 We've curated four tips from last week's livestream about how you can transition to a career in responsible tech! All Tech Is Human Program Manager Sarah Welsh, PhD moderated a panel featuring Kyle Boerstler, PhD (Senior Data Scientist, Activision), Elizabeth Thompson (MS Data Science @ NC State | Researcher @ UNC), Daniel Siegel (AI & Malicious Actor Analyst, Stealth), and Megan Shahi (Director, Technology Policy at Center for American Progress). Read more tips and insights: https://lnkd.in/gbpvYr38 Do you have additional advice for people looking to transition to the field? Share it with us in the comments!

Get in touch with us for your AI Advisement related needs!

📢 Throwback Interview: "Oracle of Silicon Valley" Tim O'Reilly on Doing Work That Matters Six years ago, I had the privilege of interviewing Tim O'Reilly, dubbed the "Oracle of Silicon Valley" by Inc Magazine and "The Trend Spotter" by Wired. His insights are just as relevant today! 🎙️ Highlights: Why Tim sees himself as a change-maker The importance of storytelling and ethics in tech His views on AI, the Singularity, and why tech isn't enough Discussing his book "WTF: What's the Future and Why It's Up to Us" Rewriting the human story and serving the right algorithms 💡 Favorite Quote: "Economic determinism is probably the cardinal sin of the present moment." Note: Despite being recorded six years ago, this conversation remains incredibly pertinent. Don't miss it! 🔗 Watch the interview here https://lnkd.in/gZpTXGD

Finally we have Adam DeConinck to share his wealth of experience in an invited talk on architecting and deploying compute clusters for large language models!

This week, ByteDance US Infrastructure lab will invite Zhaozhuo Xu, an assistant professor from Stevens Institute of Technology, to talk about "Advancing Efficient Long Context LLM Serving: Challenges, Trade-Offs, and Opportunities". Here is the abstract, and we are looking forward to the talk: In recent years, large language models (LLMs) have achieved remarkable accomplishments, driven by their increasing sizes. However, the rapid growth in model scale has surpassed advancements in hardware and network capabilities, posing challenges for deploying these LLMs efficiently on existing hardware. Moreover, current LLM algorithms and infrastructure often face an inherent trade-off between accuracy and efficiency, where optimizing for computation and memory usage can compromise inference accuracy. This presentation explores recent algorithmic advancements aimed at addressing this trade-off in LLM deployment. I will introduce a taxonomy of efficient LLM serving strategies on data management, model optimization, algorithmic enhancements, and hardware utilization. Additionally, the talk will highlight emerging trends and insights for the future development of LLM serving infrastructures capable of handling long-context scenarios effectively.

This paper from deepmind is very interesting in that it clearly distinguishes between the LLM not knowing the ground truth vs the variability in the ground truth itself (caused by e.g. an ambiguous prompt like "who's the president?"). One immediate use case is when we use LLMs for data labeling. If we generate 10 responses with 10 different retrievals and one of them happens to be very similar to the ground truth, it's not clear if all the other ones were wrong. This is simply because we don't have all the possible ground truths. https://lnkd.in/efa9bcQj

Hi friends 👋 A new edition of Calculated Randomness is out! In this newsletter, I’m documenting my journey building machine learning and distributed systems projects from scratch. Just published the latest update of the first project I am tackling: a real-time traffic accident detection system.  If you’re interested in machine learning, distributed systems, or enjoy following along hands-on technical projects, you might find this interesting.

👏 important step forward in understanding "atomic" networks of concepts and features representation inside production grade foundation LLms, the implications are fascinating 1) can we imagine training millions of tiny LLms each representing a feature fine tuned for a specific task 2) can they be then "transplanted" into an existing network? 3) can a larger network be assembled from many parts ? let me know your thoughts

Now good folks at Anthropic have released "Golden Gate Bridge LLM" on Claude.ai so we can all play with it, it's spooky how every question is answered with "Golden gate bridge" as the central anchor so if you ask "What's the best coffee in Austin, Texas" is says "There are several iconic spots to get fantastic coffee in San Francisco, including: Golden Gate Bridge - This famous suspension bridge is perfect for photo opportunities and is one of the most visited attractions in the San Francisco area. It spans the Golden Gate strait and connects the city to the Marin headlands." - self-centered to the extreme :) This shows how safety and red-teaming in production grade LLMs is super critical to get right and tweaking few "features" can drastically alter the responses

Andrew Ng said, "As ML Engineer, our job is to make things work, so if the data set changes, it's our responsibility to adjust to the new distribution rather than wish for the old one." I am pleased to share that I’ve completed the Machine Learning in Production certification by DeepLearning.AI. The course underscored that machine learning in production is an iterative process, stressing the importance of thorough planning and effective communication.  📊🔄