Peter Boshard Olson retweetledi
Congratulations @JumpAdvisorAI Jump, winner of the TECHNOLOGY PROVIDERS: Artificial Intelligence > Meeting Support category of the 2025 Wealth Management Industry Awards #Wealthies
English
Peter Boshard Olson
442 posts
Peter Boshard Olson
@peterbolson
Sr. Dev Advocate @ Snowflake - The Data Cloud. Prev: CoS @ Ponder (acq. by Snowflake), Sr. Data Scientist @ Coursera. Bachelor's in Applied Math from Harvard.
Salt Lake City, UT Katılım Mart 2013
531 Takip Edilen214 Takipçiler
Peter Boshard Olson retweetledi
I've been using @SnowflakeDB Copilot for the past two days and have successfully onboarded my entire Data Science team using Copilot. Snowflake copilot is pretty fast and way faster than OpenAi.
Text-to-SQL is likely to be one of the first challenges successfully tackled by large language models, and its direct integration into the data warehouse offers sheer convenience.
We experimented with various types of queries, and Copilot handled easy to moderately complex queries brilliantly. It struggled with larger or more complex queries, but those are not the situations where I anticipate needing Copilot's assistance.
English
@modin_project This is great -- and I like the subtle point that even when a change makes individual queries slower, if it dramatically improves the slowest portion of a workload, that can more than offset all the other changes
English
Peter Boshard Olson retweetledi
💥Performance Improvement for .merge in Modin!💥
The result? H2O benchmark (500 MB) total execution dropped by 70% (28.64s -> 8.78s)
What changed?
Before: Gather the right df into one partition
After: Repartition so we can broadcast only row partitions of the right df
English
Peter Boshard Olson retweetledi
Dmitry Chigarev at @IntelDevTools wrote a post about Modin: dchigarev.github.io/modin_perf_exa…
He takes a 4-minute pandas workload & says:
"Modin could be a solution here, offering a drop-in replacement for Pandas and efficient parallel implementations of its API."
With Modin? 55 seconds
English
Peter Boshard Olson retweetledi
Join us for @SnowflakeDB’s first-ever Dev Day! dive into app dev, #GenAI bootcamp, get career mentoring, swag, and have a blast!
Snowflake.com/summit/devday
English
Peter Boshard Olson retweetledi
We're trying to understand when, how, and why people search for datasets, to build better dataset search systems. Please RT this 2 minute survey!
cc @UCBIDS @BerkeleyDataSci @BerkeleyISchool @Berkeley_EECS
English
This is where I usually delete the repo and clone fresh from github
The Figen@TheFigen_
Kindness... twitter.com/Sadie2901/stat…
English
Peter Boshard Olson retweetledi
I just published "Testing the Snowflake Query Acceleration Service with a 17 TB table": @hoffa/testing-the-snowflake-query-acceleration-service-with-a-17-tb-table-21677a0d6de6" target="_blank" rel="nofollow noopener">medium.com/@hoffa/testing…
@SnowflakeDB has an easy way to make queries faster while using the smallest warehouses: The Query Acceleration Service. Let's test it with 17TB of @github events.
English
@AbeGong I've only experienced the first of these -- junior person (me) working with a senior person. Even a few mins of this was helpful. At @Quora, @adamdangelo would emphasize that it's good to make investments in dev velocity. I think pair programming can be a big part of this [2/2]
English
This is turning into a really interesting thread on dev culture.
Cory House@housecor
Theory: We know pair programming is effective. But, most developers prefer working alone. So chances are, we don't pair as often as we should. How do you interpret the results of these two polls?
English
@AbeGong I think there are two kinds of pair programming -- knowledge transfer PP, where a junior dev works with a senior dev, and can be like: "What was that hotkey you just used?!?" And "two heads are better than one" PP, where two senior people pool their thinking [1/2]
English
Peter Boshard Olson retweetledi
Creating personas for data engineers vs analytics engineers vs data scientists be like
Nature is Amazing ☘️@AMAZlNGNATURE
QUIZ TIME which is a Gator, Croc, and Caiman 🤔
English
@lmushin @barrnanas I read this too quickly and thought you said "ironic," and that significantly altered my viewing experience. I kept thinking this was going to turn into an Icarus moment. But no, panache won -- phew!
English
@barrnanas that and our shared love for alternative activities/events ❤️
English
allow me to be hyperbolic to say I truly believe this is one of the most iconic moments in US history
Historic Vids@historyinmemes
Back in November, 1999, John Carpenter took part in the first season of the American version of the long-running game show, Who Wants to Be a Millionaire? Hosted by the late Regis Philbin, the quiz show includes three 'lifeline' options to help contestants answer particularly tricky questions, with the most popular option giving contestants a chance to 'phone-a-friend'. Carpenter managed to make it through the game without using any lifelines, which is an impressive feat in itself. But that all changed when it came to the final, million dollar question.
English
@BetseyStevenson It's always struck me as mind-blowing that William Goldman, the author of The Princess Bride, also wrote the screenplay for Butch Cassidy and the Sundance Kid, and the screenplay for All the President's Men. Extraordinary! Inconceivable!
English
I'm considering dropping a quote from the Princess Bride in all of my tweets. I'll explain, and I'll use small words.
English
And there are 6 major architectures for AI apps, depending on how involved a human is in choosing each of 3 things: the step output, the steps to take, and the possible step sequences.
For more, watch 3:03 to 9:10 in @hwchase17's BUILD video: snowflake.com/build/americas…
🧵(2/2)
English
Clear explanation from @langchain's @hwchase17:
There are 4 main ways to give an AI app background info:
1. Instruction prompting
2. Few-shot examples
3. RAG
4. Fine-tuning
It hadn't occurred to me that these are all ways of accomplishing the same broad goal. 🧵(1/2)
English
@AbeGong I know some 10X engineers, but my sense is they also tend to be paid 10X (or at least 5X), so for me the 10X engineer conversation becomes more interesting when framed in terms of arbitrage -- Where is the optimal "X-ness-per-dollar-in-comp"? Is there an elbow?
English
@AbeGong I love the thought that there's a "Blue Ox" equivalent in the 10X Engineer's life. An unreleased M4 mac? A hemispheric computer monitor (you look up, you look back, it's there).
English