Alex Strick van Linschoten
@strickvl.bsky.social
ML Engineer (@ ZenML), researcher (& author of a few books).
Have been participating in a group that is doing fastAI part 1 for the first time. Took the opportunity in lesson 1 to use the text dataloaders / trainers to do a little miniproject to see how well fastAI does at classifying between the outputs of different LLM families.
May 5, 2025 at 9:54 AM
Have been participating in a group that is doing fastAI part 1 for the first time. Took the opportunity in lesson 1 to use the text dataloaders / trainers to do a little miniproject to see how well fastAI does at classifying between the outputs of different LLM families.
With all the recent excitement around Model Context Protocol (MCP), we've built something practical: a way to query your infrastructure, analyze pipeline performance, and trigger runs through simple conversation rather than CLI commands.
March 10, 2025 at 9:26 AM
With all the recent excitement around Model Context Protocol (MCP), we've built something practical: a way to query your infrastructure, analyze pipeline performance, and trigger runs through simple conversation rather than CLI commands.
So excited to read this (just arrived). Had been waiting a long time.
March 8, 2025 at 4:23 PM
So excited to read this (just arrived). Had been waiting a long time.
let's do this!
February 10, 2025 at 4:37 PM
let's do this!
yeah this is really clear
February 1, 2025 at 5:10 PM
yeah this is really clear
🔍 Chapter 10 dives deep into evaluating LLM applications - fascinating to learn that GitHub Copilot's first piece of code was actually their evaluation harness. The authors share practical frameworks for both offline and online testing, essential knowledge for anyone building AI applications.
January 17, 2025 at 3:57 PM
🔍 Chapter 10 dives deep into evaluating LLM applications - fascinating to learn that GitHub Copilot's first piece of code was actually their evaluation harness. The authors share practical frameworks for both offline and online testing, essential knowledge for anyone building AI applications.
Systematic measurement is crucial. Fuzzy Labs demonstrates this: baseline testing → inference optimization → horizontal scaling. Each step validated by metrics, each optimization backed by data.
January 13, 2025 at 9:11 AM
Systematic measurement is crucial. Fuzzy Labs demonstrates this: baseline testing → inference optimization → horizontal scaling. Each step validated by metrics, each optimization backed by data.
First clear pattern: Domain-specific models consistently outperform larger alternatives. Deepgram's 500M parameter model shows this perfectly - achieving better results than foundation models at 1/200th the size.
January 13, 2025 at 9:11 AM
First clear pattern: Domain-specific models consistently outperform larger alternatives. Deepgram's 500M parameter model shows this perfectly - achieving better results than foundation models at 1/200th the size.
I track everything I read with @thestorygraph.com. They have this cool little feature where it pipes some context around the books you've read, your tastes etc to an LLM when you're browsing a book and you get a recommendation for whether you might be into it or not.
January 12, 2025 at 4:52 PM
I track everything I read with @thestorygraph.com. They have this cool little feature where it pipes some context around the books you've read, your tastes etc to an LLM when you're browsing a book and you get a recommendation for whether you might be into it or not.
The key insight here (as we get into what it means to do things with a real application) is that the key task of prompt engineering is to translate or transform a user's problem domain into something that can work within the universe of an LLM, and then back again when the LLM responds.
January 11, 2025 at 8:53 PM
The key insight here (as we get into what it means to do things with a real application) is that the key task of prompt engineering is to translate or transform a user's problem domain into something that can work within the universe of an LLM, and then back again when the LLM responds.
Chapter 4 of "Prompt Engineering for LLMs" is sort of a high-level introduction to several chapters that follow. Nothing ground-breaking but some useful framing.
January 11, 2025 at 8:53 PM
Chapter 4 of "Prompt Engineering for LLMs" is sort of a high-level introduction to several chapters that follow. Nothing ground-breaking but some useful framing.
Nice trip down memory lane. I’m constantly forgetting that OpenAI’s ChatGPT was only released in late November 2022….
January 8, 2025 at 5:58 PM
Nice trip down memory lane. I’m constantly forgetting that OpenAI’s ChatGPT was only released in late November 2022….
There have been a few new books in the LLM space in recent weeks and I'll be reading a chapter a day of some of them. First up, John Berryman and Albert Ziegler's 'Prompt Engineering for LLMs'.
#33DaysOfLLMOpsReading
#33DaysOfLLMOpsReading
January 8, 2025 at 5:49 PM
There have been a few new books in the LLM space in recent weeks and I'll be reading a chapter a day of some of them. First up, John Berryman and Albert Ziegler's 'Prompt Engineering for LLMs'.
#33DaysOfLLMOpsReading
#33DaysOfLLMOpsReading
Blocking two swatches I knitted 🧶 for the TKGA Learn to Knit course. Next up: a single rib and garter stitch swatch.
January 8, 2025 at 9:40 AM
Blocking two swatches I knitted 🧶 for the TKGA Learn to Knit course. Next up: a single rib and garter stitch swatch.
Seriously how is it that medium.com even doesn't provide RSS feeds!
January 6, 2025 at 8:49 AM
Seriously how is it that medium.com even doesn't provide RSS feeds!
- And finally, Cara McGooga's *The Poison Line: The Shocking True Story of How a Miracle Cure Became a Deadly Poison*
A deeply reported account of the infected blood scandal in the UK, how it was experienced by those who fell prey and how it was covered up by industry & governments.
A deeply reported account of the infected blood scandal in the UK, how it was experienced by those who fell prey and how it was covered up by industry & governments.
December 23, 2024 at 7:00 PM
- And finally, Cara McGooga's *The Poison Line: The Shocking True Story of How a Miracle Cure Became a Deadly Poison*
A deeply reported account of the infected blood scandal in the UK, how it was experienced by those who fell prey and how it was covered up by industry & governments.
A deeply reported account of the infected blood scandal in the UK, how it was experienced by those who fell prey and how it was covered up by industry & governments.
- Chris van Tulleken's *Ultra-Processed People*
Come for the history of how we ended up with our messed up current food reality, stay for the thought that you can apply this as a metaphor for your information consumption as well...
Come for the history of how we ended up with our messed up current food reality, stay for the thought that you can apply this as a metaphor for your information consumption as well...
December 23, 2024 at 7:00 PM
- Chris van Tulleken's *Ultra-Processed People*
Come for the history of how we ended up with our messed up current food reality, stay for the thought that you can apply this as a metaphor for your information consumption as well...
Come for the history of how we ended up with our messed up current food reality, stay for the thought that you can apply this as a metaphor for your information consumption as well...
- Yasmin El-Rifae's *Radius: A Story of Feminist Revolution*
Was blown away by this. Stories of an Egyptian group who tried to address the huge issue of sexual violence in the context of the Arab Spring protests. Raw, shocking and brave.
Was blown away by this. Stories of an Egyptian group who tried to address the huge issue of sexual violence in the context of the Arab Spring protests. Raw, shocking and brave.
December 23, 2024 at 7:00 PM
- Yasmin El-Rifae's *Radius: A Story of Feminist Revolution*
Was blown away by this. Stories of an Egyptian group who tried to address the huge issue of sexual violence in the context of the Arab Spring protests. Raw, shocking and brave.
Was blown away by this. Stories of an Egyptian group who tried to address the huge issue of sexual violence in the context of the Arab Spring protests. Raw, shocking and brave.
I read a whole bunch of books this past year. My standout favourites included:
- (most recently) *If I Must Die: Poetry and Prose* by Refaat Alareer
A powerful posthumous collection by a Gazan writer who was killed in an Israeli airstrike last December.
- (most recently) *If I Must Die: Poetry and Prose* by Refaat Alareer
A powerful posthumous collection by a Gazan writer who was killed in an Israeli airstrike last December.
December 23, 2024 at 7:00 PM
I read a whole bunch of books this past year. My standout favourites included:
- (most recently) *If I Must Die: Poetry and Prose* by Refaat Alareer
A powerful posthumous collection by a Gazan writer who was killed in an Israeli airstrike last December.
- (most recently) *If I Must Die: Poetry and Prose* by Refaat Alareer
A powerful posthumous collection by a Gazan writer who was killed in an Israeli airstrike last December.
The one where my cat joins in the Chinese lesson...
December 13, 2024 at 9:39 AM
The one where my cat joins in the Chinese lesson...
🔍 Standout case: Canva's post-incident review system solved the consistency problem through structured prompts. Ubisoft cracked cost optimization with output-based pricing. These aren't experiments—they're running in production now.
December 11, 2024 at 9:55 AM
🔍 Standout case: Canva's post-incident review system solved the consistency problem through structured prompts. Ubisoft cracked cost optimization with output-based pricing. These aren't experiments—they're running in production now.
📊 After analyzing real production LLMOps data, here's what actually works for prompt engineering: structured prompts for reliability, systematic versioning for scale, and retrieval-augmented generation for efficiency. No theory—just battle-tested approaches.
December 11, 2024 at 9:55 AM
📊 After analyzing real production LLMOps data, here's what actually works for prompt engineering: structured prompts for reliability, systematic versioning for scale, and retrieval-augmented generation for efficiency. No theory—just battle-tested approaches.
🤔 Key insight: Many teams are learning that "agent" architectures often overcomplicate what could be simple, deterministic flows. The most effective systems aren't autonomous agents - they're LLMs integrated into predictable workflows with clear boundaries and failure modes.
December 9, 2024 at 8:57 AM
🤔 Key insight: Many teams are learning that "agent" architectures often overcomplicate what could be simple, deterministic flows. The most effective systems aren't autonomous agents - they're LLMs integrated into predictable workflows with clear boundaries and failure modes.