Roger L. Cauvin
@rcauvin.bsky.social
640 followers 930 following 300 posts
♟️ #prodmgmt | Positioning | Strategy 👨‍🔬 Consumer Science | ML | AI 🏢 Downtown Austin dweller 🏘️ Advocate for inclusive neighborhoods
Posts Media Videos Starter Packs
rcauvin.bsky.social
When someone asks for a report, don't ask, "What do you want in the report?" Instead, ask what decisions it will help them make.

"Nobody wakes up in the morning and says today's going to be a great day, I'm going to produce 15 reports." - @jmspool.bsky.social
rcauvin.bsky.social
Today's #AI and #ML models rely on memorization and "local generalization". Massive scaling has led to smarter models. But getting models to adapt to novel situations, where there is little or no data, will require more than just memorizing more data.
deeplearningwithpython.io/chapters/cha...
Deep Learning with Python, Third Edition
Deep Learning with Python is written for anyone who wishes to explore deep learning from scratch. This new edition adds comprehensive coverage of generative AI and modern deep learning frameworks. It ...
deeplearningwithpython.io
Reposted by Roger L. Cauvin
downtownaustin.org
To honorable City Council, Commissioners, & Staff:

DANA opposes height limits and other exclusionary zoning downtown, and we specifically oppose the base height limit proposed in the C20-2025-010 Central Business District (CBD) amendments. 🧵
rcauvin.bsky.social
Sometimes we need to fundamentally rethink the existing journey instead of selectively fixing points of frustration within it. Either way, breaking down the larger aspirational vision into visionettes makes a lot of sense.
rcauvin.bsky.social
Me: "I start walking north at constant forward pace of 2 mph. I continue walking at the same pace, but each minute, I turn 1 degree clockwise. After 360 minutes, where do I end up relative to my starting point?"

Gemini: "After 360 minutes, you will end up at your starting point."
rcauvin.bsky.social
Wilson scoring is a simple way to score the popularity of items (such as comments on a Reddit post) based on up-votes and down-votes. As far back as 2009, Evan Miller pointed out the flaws in other methods (e.g. average rating).
www.evanmiller.org/how-not-to-s...
H/T @kennyd.bsky.social
How Not To Sort By Average Rating
Users are rating items on your website. How do you know what the highest-rated items are?
www.evanmiller.org
Reposted by Roger L. Cauvin
downtownaustin.org
Join in for September's volunteer Shoal Creek clean-up session!

WHEN: Saturday, Sep 13 (8:45-9a)
WHERE: outside The Independent (301 West Ave)

The monthly collaboration continues with APD, DAA, & the Downtown Austin Community Court.

www.downtownaustin.org/september-20...
September Shoal Creek Clean-Up
WHAT: September Shoal Creek Clean-Up WHEN: Saturday, Sept. 13, 8:45 am WHERE: The Independent @ 301 West Ave. Volunteers will meet Saturday, September 13, 2025 outside The Independen…
www.downtownaustin.org
rcauvin.bsky.social
Concerns about "timing" are also frustrating. I remember a time when, in a meeting, everyone supported a position, but a few people said it was "too early" to publicly express our support. Then, at the next meeting, the same people said it was "too late" to express our support.
rcauvin.bsky.social
One of the most frustrating situations in civic advocacy is when potential allies think the cause has merit but is not "politically viable". Yet the reason it's not viable is that they, themselves, haven't publicly voiced their support.
rcauvin.bsky.social
Also, strategy is largely about aligning current and future decisions with deliberate choices you've already made. #prodmgmt
rcauvin.bsky.social
I hope pro-housing advocates will embrace and apply this mindset to density bonus programs. Inclusionary zoning and density bonus programs are part of a scarcity mindset, not an abundance agenda. We should advocate for better alternatives.
Reposted by Roger L. Cauvin
downtownaustin.org
Come out to DANA's End of Summer Urban Core Happy Hour this Sun, Aug 24 at The Stay Put (73 Rainey St.) from 4-7pm!

Our guest speaker will be Davon Barbour, CEO of @DowntownATXInfo! Entry comes with 2 free drink tickets & appetizers.

More info and RSVP here:
www.eventbrite.com/e/dana-end-o...
rcauvin.bsky.social
"It combines the deep language understanding capabilities of a large language model with the speed and reliability of a dedicated classification head. This leads to a more accurate, consistent, and cost-effective solution."
rcauvin.bsky.social
I asked Gemini (in the context of classification problems), and it agrees:

"For most practical classification problems, the embedding-based approach is unequivocally better."
rcauvin.bsky.social
An approach that leverages the LLM's first three layers to extract the meaning of the input, bypasses the final output layer, and instead applies task-specific machine learning layers to make the final predictions, is likely to perform better.
rcauvin.bsky.social
The fourth "layer" merely outputs ("predicts") a likely "next token". It doesn't directly predict an outcome. You can certainly ask the LLM to output "next tokens" that effectively predict or classify, but it's not what the output layer is designed to do.
rcauvin.bsky.social
A typical LLM (1) tokenizes the input, (2) encodes the tokens into "embeddings" (semantic representations), (3) transforms the tokens and encodings to further capture the meaning of the input, and (4) outputs the most relevant and likely "next token".
rcauvin.bsky.social
Increasingly, I've noticed vibe coders and product people suggesting that prompts and LLMs can replace specialized machine learning models for making predictions. We only have to look at how LLMs work to recognize the flaw in this belief. #AI #ML 🧵
rcauvin.bsky.social
"Google seems to be working on bringing NotebookLM to Gemini, combining two of its most powerful AI apps.

The option to import notebooks will reportedly appear in the same menu where you currently add images, files, or Google Drive documents to Gemini."

www.androidpolice.com/googles-best...
Google's best AI tools could finally join forces
NotebookLM integration might turn Gemini into a serious research tool
www.androidpolice.com
rcauvin.bsky.social
Without much fanfare, @divyasreepat.bsky.social and her team have released Keras 3.11. It includes support for Grain, a declarative data processing solution similar to TensorFlow datasets, yet not dependent on TensorFlow or any particular backend.
github.com/google/grain
#MachineLearning #ML #AI
GitHub - google/grain: Library for reading and processing ML training data.
Library for reading and processing ML training data. - google/grain
github.com
rcauvin.bsky.social
"A very thin line separates the 'good' kind of personalization from the creepy kind."
thedecisionlab.com/insights/tec...
by Preeti Kotamarthi
#ML #MachineLearning
thedecisionlab.com