Saturday's digest #15

Weekly curated articles for the busy AI & data professional

Author

Jeremy Lezac
August 10, 2024

Hello,

Saturday. Digest. 5 articles. For you. Let’s roll.

#1 - Ben Dickson from TechTalks - Don’t evaluate compressed LLMs based on accuracy

Don't evaluate compressed LLMs based on accuracy

Compressed LLMs maintain their accuracy metrics in comparison to the baseline models. But their behavior changes dramatically, according to other metrics.

bdtechtalks.substack.com/p/dont-evaluate-compressed-llms-based

Understand why accuracy alone isn't enough when evaluating compressed LLMs. Compression can change model behavior, even if accuracy stays the same. Learn about "flips" and better metrics to ensure compressed LLMs still perform like the originals.

#2 - David Pereira (and Stefan Wolpers) from Untrapping Product Teams - Overcoming Dangerous Scrum Anti-Patterns with Stefan Wolpers

Overcoming Dangerous Scrum Anti-Patterns with Stefan Wolpers

Beyond Doing Scrum for the sake of it

dpereira.substack.com/p/overcoming-dangerous-scrum-anti-patterns-260

If you're using Scrum but not getting the results you want, this article is for you. It breaks down how rigid Scrum practices can hinder progress and offers practical tips to refocus on creating real value. Learn how to avoid common pitfalls and make Scrum work for your team’s success.

#3 - Anne-Laure Le Cunff from NessLabs - The Curse of Knowledge

The Curse of Knowledge

The curse of knowledge is a cognitive bias that occurs when someone incorrectly assumes that others have enough background to understand.

nesslabs.com/curse-of-knowledge

The "curse of knowledge" happens when experts (you) assume others understand their jargon, leading to poor communication. This article explains how to overcome this bias by simplifying language, knowing your audience, and using visuals and storytelling. It’s a must-read for improving communication skills and making your explanations clear and relatable.

#4 - John Cutler from The Beautiful Mess - Stop the (Goal) Cascade Madness

TBM 305: Stop the (Goal) Cascade Madness

Goal cascades are compelling, but they don’t work nearly as well as we think they do.

cutlefish.substack.com/p/tbm-305-stop-the-goal-cascade-madness

Keep on with the “let’s mimic this process because everyone is doing it and let’s do it poorly (no biggies, everyone is also doing it poorly): Goal cascades.
Goal cascades don't work as well as we think.
This article explains why—starting with front-line teams. Instead of intricate goal cascades, focus on stable, team-level metrics that really matter. Simplify your approach, align teams around core goals, and avoid getting lost in meaningless mapping exercises.

#5 - Eric Newcomer and Madeline Renbarger from Newcomer - Investors Ask What’s Next As FOundation Model Mania Recedes

Investors Ask What's Next As Foundation Model Mania Recedes

Plus, supply chain wreckage, and Google's antitrust troubles

www.newcomer.co/p/investors-ask-whats-next-as-foundation

The AI investment frenzy is cooling off. After a huge initial surge in AI-related stocks like Nvidia, private market enthusiasm is waning, and investors are getting cautious. The AI bubble may not have popped yet, but the easy gains are gone. Now, it's all about finding real, valuable applications for AI. It’s time for investors and startups to focus on building practical products that people actually want. The hype may be fading, but the AI revolution is far from over.

See you next week!

Reply

or to participate.