Victor Geislinger
@victorsvector.com
Combination of machine learning engineer / data scientist and teacher (depending on what hat I'm wearing that day)
Focusing on ML/AI @Google
(obligatory 'opinions are my own')
github.com/MrGeislinger
Focusing on ML/AI @Google
(obligatory 'opinions are my own')
github.com/MrGeislinger
Pinned
Victor Geislinger
@victorsvector.com
· Oct 28
MrGeislinger - Overview
Combination of machine learning engineer / data scientist and teacher (depending on what hat I'm wearing that day) - MrGeislinger
GitHub.com
Hi! I'm Victor!
Here's a quick intro :D
These days I focus on ML/AI at Google but have been a teacher, data scientist MLE, etc.
I try to be genuine & positive so when I leave this world it's in a better state than I found it
You can likely find my other online presences on GitHub.com/MrGeislinger
Here's a quick intro :D
These days I focus on ML/AI at Google but have been a teacher, data scientist MLE, etc.
I try to be genuine & positive so when I leave this world it's in a better state than I found it
You can likely find my other online presences on GitHub.com/MrGeislinger
mood is set
Let's read the last book of Twilight
vampires by the campfire
Let's read the last book of Twilight
vampires by the campfire
November 9, 2025 at 2:50 AM
mood is set
Let's read the last book of Twilight
vampires by the campfire
Let's read the last book of Twilight
vampires by the campfire
Reposted by Victor Geislinger
One of these pitch ideas might be of interest to linguists who are looking to write for a general audience!
November 8, 2025 at 5:07 PM
One of these pitch ideas might be of interest to linguists who are looking to write for a general audience!
Super interesting to me on the model results in this new Oolong dataset re: questins that involves aggregation over a long context (love the name)
I wondered along the lines of how well LLMs do in long-contedt— not generally amazing in some contexts according to this XD
arxiv.org/abs/2511.02817
I wondered along the lines of how well LLMs do in long-contedt— not generally amazing in some contexts according to this XD
arxiv.org/abs/2511.02817
November 8, 2025 at 4:23 PM
Super interesting to me on the model results in this new Oolong dataset re: questins that involves aggregation over a long context (love the name)
I wondered along the lines of how well LLMs do in long-contedt— not generally amazing in some contexts according to this XD
arxiv.org/abs/2511.02817
I wondered along the lines of how well LLMs do in long-contedt— not generally amazing in some contexts according to this XD
arxiv.org/abs/2511.02817
Frankly haven't watched this yet but plan on it— Sebastian does an amazing job knowledge sharing & educating about LLMs
My "The Building Blocks of Today’s and Tomorrow’s Language Models" talk at the PyTorch Conference is now up on YouTube! youtube.com/watch?v=nDl6...
The silver lining of my late arrival and rescheduling: There was no talk after mine, it's followed by a 30 min Q&A instead of just the usual 5 :)
The silver lining of my late arrival and rescheduling: There was no talk after mine, it's followed by a 30 min Q&A instead of just the usual 5 :)
The Building Blocks of Today’s and Tomorrow’s Language Models - Sebastian Raschka, RAIR Lab
YouTube video by PyTorch
youtube.com
November 8, 2025 at 4:11 PM
Frankly haven't watched this yet but plan on it— Sebastian does an amazing job knowledge sharing & educating about LLMs
Reposted by Victor Geislinger
DataRater can continuously+adaptively learn which examples will improve an ML model the most during training.
Delighted to have worked on this with @luisazintgraf.bsky.social, @dancalian.bsky.social, @gregfar.bsky.social, Iurii Kemaev, et al. (see full list at the of Luisa's thread).
Delighted to have worked on this with @luisazintgraf.bsky.social, @dancalian.bsky.social, @gregfar.bsky.social, Iurii Kemaev, et al. (see full list at the of Luisa's thread).
Excited to share our new paper, "DataRater: Meta-Learned Dataset Curation"!
We explore a fundamental question: How can we *automatically* learn which data is most valuable for training foundation models?
Paper: arxiv.org/pdf/2505.17895 to appear at @neuripsconf.bsky.social
Thread 👇
We explore a fundamental question: How can we *automatically* learn which data is most valuable for training foundation models?
Paper: arxiv.org/pdf/2505.17895 to appear at @neuripsconf.bsky.social
Thread 👇
November 6, 2025 at 3:43 PM
DataRater can continuously+adaptively learn which examples will improve an ML model the most during training.
Delighted to have worked on this with @luisazintgraf.bsky.social, @dancalian.bsky.social, @gregfar.bsky.social, Iurii Kemaev, et al. (see full list at the of Luisa's thread).
Delighted to have worked on this with @luisazintgraf.bsky.social, @dancalian.bsky.social, @gregfar.bsky.social, Iurii Kemaev, et al. (see full list at the of Luisa's thread).
Morning's a little wild so far
Woke up at 4am to bike to the bus to get to the work office (it's been a bit) only to realize ~½ there that I forgot my work ID. So biked all the way back & couldn't find it. Contemplating what next, my kid finds it under my desk!
So I'm just now at the bus stop :D
Woke up at 4am to bike to the bus to get to the work office (it's been a bit) only to realize ~½ there that I forgot my work ID. So biked all the way back & couldn't find it. Contemplating what next, my kid finds it under my desk!
So I'm just now at the bus stop :D
November 6, 2025 at 1:59 PM
Morning's a little wild so far
Woke up at 4am to bike to the bus to get to the work office (it's been a bit) only to realize ~½ there that I forgot my work ID. So biked all the way back & couldn't find it. Contemplating what next, my kid finds it under my desk!
So I'm just now at the bus stop :D
Woke up at 4am to bike to the bus to get to the work office (it's been a bit) only to realize ~½ there that I forgot my work ID. So biked all the way back & couldn't find it. Contemplating what next, my kid finds it under my desk!
So I'm just now at the bus stop :D
A movie you've seen more than seven times with a gif.
November 4, 2025 at 12:44 AM
A movie you've seen more than seven times with a gif.
I've been playing Factorio this morning while my 5yr olds watched, suggested things I do, and did 'hair salon' with me
It's been a good morning
It's been a good morning
November 2, 2025 at 5:11 PM
I've been playing Factorio this morning while my 5yr olds watched, suggested things I do, and did 'hair salon' with me
It's been a good morning
It's been a good morning
How did I utilize my extra hour?
I finally decided to give in & play Factorio for the first time, embracing what I knew would become my new addiction
I finally decided to give in & play Factorio for the first time, embracing what I knew would become my new addiction
November 2, 2025 at 4:00 PM
How did I utilize my extra hour?
I finally decided to give in & play Factorio for the first time, embracing what I knew would become my new addiction
I finally decided to give in & play Factorio for the first time, embracing what I knew would become my new addiction
completely triggering— I love my punctuation outside the quotes
But also, I'd like to take this moment to completely denounce 'smart quotes' as the plague that they truly are
And finally: ' > "
But also, I'd like to take this moment to completely denounce 'smart quotes' as the plague that they truly are
And finally: ' > "
November 1, 2025 at 8:32 PM
completely triggering— I love my punctuation outside the quotes
But also, I'd like to take this moment to completely denounce 'smart quotes' as the plague that they truly are
And finally: ' > "
But also, I'd like to take this moment to completely denounce 'smart quotes' as the plague that they truly are
And finally: ' > "
Ooo I really like this! I haven't really been using labelers but this was one that I could see behind super useful to filter out accounts if rather not even bother with
bsky.app/profile/stec...
bsky.app/profile/stec...
November 1, 2025 at 8:00 PM
Ooo I really like this! I haven't really been using labelers but this was one that I could see behind super useful to filter out accounts if rather not even bother with
bsky.app/profile/stec...
bsky.app/profile/stec...
I'm halfway through reading the Twilight saga for the first time (if you include midnight sun), and man are some of those characters just as bonkers as they are in the movies
I love it lol
I'm totally into it at this point of the story, just like when I binge watch the movies XD
I love it lol
I'm totally into it at this point of the story, just like when I binge watch the movies XD
November 1, 2025 at 3:04 PM
I'm halfway through reading the Twilight saga for the first time (if you include midnight sun), and man are some of those characters just as bonkers as they are in the movies
I love it lol
I'm totally into it at this point of the story, just like when I binge watch the movies XD
I love it lol
I'm totally into it at this point of the story, just like when I binge watch the movies XD
"Friction is an essential ingredient for learning"
To me, that's the most important point. Any technology including LLMs can make it 'too easy'. Remember that learning will feel like an effort
Also why I don't default to just videos to learn— easy to trick yourself that you understand by watching
To me, that's the most important point. Any technology including LLMs can make it 'too easy'. Remember that learning will feel like an effort
Also why I don't default to just videos to learn— easy to trick yourself that you understand by watching
When people learn with ChatGPT instead of following their own searches, they end up knowing less, caring less, and producing worse advice, even when the facts are the same.
Friction is an essential ingredient for learning! Convenience makes us shallow.
academic.oup.com/pnasnexus/ar...
Friction is an essential ingredient for learning! Convenience makes us shallow.
academic.oup.com/pnasnexus/ar...
Experimental evidence of the effects of large language models versus web search on depth of learning
Abstract. The effects of using large language models (LLMs) versus traditional web search on depth of learning are explored. A theory is proposed that when
academic.oup.com
October 28, 2025 at 4:24 PM
"Friction is an essential ingredient for learning"
To me, that's the most important point. Any technology including LLMs can make it 'too easy'. Remember that learning will feel like an effort
Also why I don't default to just videos to learn— easy to trick yourself that you understand by watching
To me, that's the most important point. Any technology including LLMs can make it 'too easy'. Remember that learning will feel like an effort
Also why I don't default to just videos to learn— easy to trick yourself that you understand by watching
I just hit the halfway point of the New Moon book for the first time & I can tell w/ it hit that I'm definitely fully invested in the book series now lol
October 26, 2025 at 5:43 AM
I just hit the halfway point of the New Moon book for the first time & I can tell w/ it hit that I'm definitely fully invested in the book series now lol
My wife has been gone all week so it's been solo parenting my twins. But the thing that really has saved me is the kids falling asleep to my wife's recorded reading of a 'meditation bed time story' of a platypus
Literally took them less than 2 minutes to drift off
Literally took them less than 2 minutes to drift off
October 24, 2025 at 4:30 AM
My wife has been gone all week so it's been solo parenting my twins. But the thing that really has saved me is the kids falling asleep to my wife's recorded reading of a 'meditation bed time story' of a platypus
Literally took them less than 2 minutes to drift off
Literally took them less than 2 minutes to drift off
@lingthusiasm.bsky.social if I had to bet anyone on Bluesky would know, I'd put my bet on you! Or at least knowing who might have a rec
recommendation request: looking for a novel about the impossibility of communication
October 24, 2025 at 4:24 AM
@lingthusiasm.bsky.social if I had to bet anyone on Bluesky would know, I'd put my bet on you! Or at least knowing who might have a rec
Seriously would love to read a book that hits on this
I'd even expand that to non-fiction. Just had to be how communication & conveying the exact idea in your head to another is not just difficult but practically impossible
I'd even expand that to non-fiction. Just had to be how communication & conveying the exact idea in your head to another is not just difficult but practically impossible
recommendation request: looking for a novel about the impossibility of communication
October 24, 2025 at 4:20 AM
Seriously would love to read a book that hits on this
I'd even expand that to non-fiction. Just had to be how communication & conveying the exact idea in your head to another is not just difficult but practically impossible
I'd even expand that to non-fiction. Just had to be how communication & conveying the exact idea in your head to another is not just difficult but practically impossible
Can't stop thinking about a chat I had where someone told me their ML team was using llama to generate an array of logits for a classification task
me: oh so you're using llama like you would with BERT to get a feature set?
them: no... it *just* outputs the array... as text output...
me: 😨😰
me: oh so you're using llama like you would with BERT to get a feature set?
them: no... it *just* outputs the array... as text output...
me: 😨😰
October 23, 2025 at 3:56 PM
Can't stop thinking about a chat I had where someone told me their ML team was using llama to generate an array of logits for a classification task
me: oh so you're using llama like you would with BERT to get a feature set?
them: no... it *just* outputs the array... as text output...
me: 😨😰
me: oh so you're using llama like you would with BERT to get a feature set?
them: no... it *just* outputs the array... as text output...
me: 😨😰
Was driving the kids to school today (not my typical) and we talked about Jumanji (from the 90s) which led about big bugs in the Carboniferous, then to how bees didn't exist, and then how flowering plants are relatively new
I love this era where these are the conversations we have w/ my 5yr olds
I love this era where these are the conversations we have w/ my 5yr olds
October 23, 2025 at 3:47 PM
Was driving the kids to school today (not my typical) and we talked about Jumanji (from the 90s) which led about big bugs in the Carboniferous, then to how bees didn't exist, and then how flowering plants are relatively new
I love this era where these are the conversations we have w/ my 5yr olds
I love this era where these are the conversations we have w/ my 5yr olds
sooo the roof started leaking *bad* in my office
But no worries, we have a siphon going so maybe this will relieve some of the puddle forming above me?
(thanks physics for the assist)
But no worries, we have a siphon going so maybe this will relieve some of the puddle forming above me?
(thanks physics for the assist)
October 14, 2025 at 2:05 AM
sooo the roof started leaking *bad* in my office
But no worries, we have a siphon going so maybe this will relieve some of the puddle forming above me?
(thanks physics for the assist)
But no worries, we have a siphon going so maybe this will relieve some of the puddle forming above me?
(thanks physics for the assist)
I'm w/ my kids to get a small toy after our family vaccination trip, and after witnessing a complete freakout until ~5s after getting their shots & them being perfectly fine, seeing how it really is just anticipation of the needle
Going to pretend it's metaphor for something but don't know what yet
Going to pretend it's metaphor for something but don't know what yet
October 14, 2025 at 12:19 AM
I'm w/ my kids to get a small toy after our family vaccination trip, and after witnessing a complete freakout until ~5s after getting their shots & them being perfectly fine, seeing how it really is just anticipation of the needle
Going to pretend it's metaphor for something but don't know what yet
Going to pretend it's metaphor for something but don't know what yet
I've been on this earth for over 3 decades & I still don't understand why the option for hand dryers to blow into your face exists
October 14, 2025 at 12:04 AM
I've been on this earth for over 3 decades & I still don't understand why the option for hand dryers to blow into your face exists
Good reminder to check your local library to see if they have have free access to @oreilly.bsky.social's digital catalog
(you won't be surprised to hear that every Bay Area library I've checked has the O'Reilly catalog available for free)
(you won't be surprised to hear that every Bay Area library I've checked has the O'Reilly catalog available for free)
The Illustrated Guide to AI Agents
New book announcement!
Thrilled that together with @maartengr.bsky.social , we're writing a new book titled “An Illustrated Guide to AI Agents” and published by @oreilly.bsky.social.
New book announcement!
Thrilled that together with @maartengr.bsky.social , we're writing a new book titled “An Illustrated Guide to AI Agents” and published by @oreilly.bsky.social.
October 13, 2025 at 3:28 PM
Good reminder to check your local library to see if they have have free access to @oreilly.bsky.social's digital catalog
(you won't be surprised to hear that every Bay Area library I've checked has the O'Reilly catalog available for free)
(you won't be surprised to hear that every Bay Area library I've checked has the O'Reilly catalog available for free)
OpenAI's 'ai video social media' primary purpose is to get human feedback on generation, right?
I relate it back to what TikTok did first w/ short-form video in getting fast feedback for recommendations. But instead of training a rec system, sora is looking for what makes a 'good generated video'
I relate it back to what TikTok did first w/ short-form video in getting fast feedback for recommendations. But instead of training a rec system, sora is looking for what makes a 'good generated video'
October 13, 2025 at 2:14 PM
OpenAI's 'ai video social media' primary purpose is to get human feedback on generation, right?
I relate it back to what TikTok did first w/ short-form video in getting fast feedback for recommendations. But instead of training a rec system, sora is looking for what makes a 'good generated video'
I relate it back to what TikTok did first w/ short-form video in getting fast feedback for recommendations. But instead of training a rec system, sora is looking for what makes a 'good generated video'
Reposted by Victor Geislinger
Happy Indigenous Peoples Day
October 13, 2025 at 1:02 PM
Happy Indigenous Peoples Day