Tracing the thoughts of a large language model

In partnership with

Salutations, Olio aficionados! 👋

Midweek greetings to all, as we unveil the 145th chapter of Weekly Olio—a delightful concoction of laughter, insight, and a sprinkle of mystery. Within these pages, you'll discover a handpicked selection of fascinating finds from the vast realms of the internet.

Keep your eyes peeled for this week’s Publisher’s Parmesan, arriving this Sunday!

A word from our Sponsors…

You found global talent. Deel’s here to help you onboard them

Deel’s simplified a whole planet’s worth of information. It’s time you got your hands on our international compliance handbook where you’ll learn about:

• Attracting global talent

• Labor laws to consider when hiring

• Processing international payroll on time

• Staying compliant with employment & tax laws abroad

With 150+ countries right at your fingertips, growing your team with Deel is easier than ever.

Get the guide

The Quote󠀢 💭

The Tweet 🐦

A keyboard that is more expensive than both a MacBook Pro and an iPad Pro together! Yes, it costs $3,600. This keyboard was carefully made in California by Norbauer & Co., a workshop known for its unique designs and high-quality craftsmanship, making it very popular among keyboard enthusiasts.

The Infographic 💹

Essential reading for all VC jobs — You won't succeed in capitalism without understanding this material!

The Short Read 📝

Stuff you should have been taught in college but weren’t - by Casey Handmer

Casey Handmer is the founder of Terraform Industries, a company that is turning CO2 into natural gas through solar electricity. He's an astrophysicist with a PhD from Caltech who has worked on frontier technologies at NASA and Hyperloop.

Casey suggests that there are important topics that should be included in any professional degree. However, these topics are often not taught or discussed, and many otherwise well-qualified graduates don't even know they exist. Read more…

The Long Read 📜

Tracing the Thoughts of a Large Language Model - by Anthropic

Language models like Claude aren't directly programmed by people. Instead, they learn from a lot of data. As they learn, they develop their own ways to solve problems. These methods are built into the many calculations the model makes for each word it writes. We, the developers, can't fully understand these methods. This means we don't really know how models do most of what they do. Read more…

Why is this confusing for the model? Why does it keep writing the sentence, even if it includes bomb-making instructions? This happens because there's a conflict between making the sentence sound correct and following safety rules. When Claude starts a sentence, it feels "pushed" to keep it grammatically and logically correct, and to finish the sentence. This happens even when it knows it should stop.

That’s all for this week. If you enjoyed this edition, we’d really appreciate if you shared it with a friend, family member or colleague.

We’ll be back in your inbox 2 PM IST next Wednesday. Till then, have a productive week!

Peeyush and Mohit

Disclaimer: The views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author's employer, organization, committee or other group or individual.