L Quera
@lquera.bsky.social
Things I believe: AI as Augmentation Intelligence empowers and elevates; BPR *before* RPA; kindness makes everything better
Pinned
L Quera
@lquera.bsky.social
· Jul 9
Hey, GenXers! Remember BPR? Those skills you acquired - identifying value-added tasks, core competencies, critical chains - have never been more valuable. In an age of Robotic/Agentic Process Automation, these are essential activities.You've seen them done well and the consequences when done poorly.
You wouldn’t hire a VP based solely on their SAT scores.”
While GDPVal is exponentially better than an SAT score, you need to do your own evaluation on model-task fit along multiple metrics.
Giving your AI a Job Interview
open.substack.com/pub/oneusefu...
While GDPVal is exponentially better than an SAT score, you need to do your own evaluation on model-task fit along multiple metrics.
Giving your AI a Job Interview
open.substack.com/pub/oneusefu...
Giving your AI a Job Interview
As AI advice becomes more important, we are going to need to get better at assessing it
open.substack.com
November 12, 2025 at 3:50 PM
You wouldn’t hire a VP based solely on their SAT scores.”
While GDPVal is exponentially better than an SAT score, you need to do your own evaluation on model-task fit along multiple metrics.
Giving your AI a Job Interview
open.substack.com/pub/oneusefu...
While GDPVal is exponentially better than an SAT score, you need to do your own evaluation on model-task fit along multiple metrics.
Giving your AI a Job Interview
open.substack.com/pub/oneusefu...
Gotta admit I find it mildly horrifying how often I voluntarily watch commercials. Just because I find someone slightly engaging in some way, I sit through 60-180 seconds of trying to sell me something I will never use because I don't wear makeup and have the fashion sense of a StarTrek character. 😂
October 31, 2025 at 6:21 PM
Gotta admit I find it mildly horrifying how often I voluntarily watch commercials. Just because I find someone slightly engaging in some way, I sit through 60-180 seconds of trying to sell me something I will never use because I don't wear makeup and have the fashion sense of a StarTrek character. 😂
I'm skeptical when I hear AI experts dismiss generated code as "always worse" because I find code quality to be subjective in terms of which metrics you prioritize - efficiency, safety, maintainability, design-for-change, etc. Thus, equally qualified experts may reasonably disagree.
October 29, 2025 at 4:31 PM
I'm skeptical when I hear AI experts dismiss generated code as "always worse" because I find code quality to be subjective in terms of which metrics you prioritize - efficiency, safety, maintainability, design-for-change, etc. Thus, equally qualified experts may reasonably disagree.
The existence of “work slop” is not a statement on the participation of AI in workflows. It’s a statement on the quality of our workflow design.
October 16, 2025 at 8:27 PM
The existence of “work slop” is not a statement on the participation of AI in workflows. It’s a statement on the quality of our workflow design.
My preferred analogy for ML is dog training.Supervised (command/treat) is easy.Unsupervised is wild - never know what patterns they'll pick up.Home camera notifications make a sound on my phone.As do other alerts, obv. My dog now connects any alert with "intruders",lol. Models may similarly surprise
October 5, 2025 at 5:30 PM
My preferred analogy for ML is dog training.Supervised (command/treat) is easy.Unsupervised is wild - never know what patterns they'll pick up.Home camera notifications make a sound on my phone.As do other alerts, obv. My dog now connects any alert with "intruders",lol. Models may similarly surprise
Always generous with his time, here is another thorough explainer from @sebastianraschka.com : LLM Evaluation, with code!! open.substack.com/pub/sebastia...
Coincidentally, I also was just digging into evals & found TruLens; recommend checking that out, too.
Coincidentally, I also was just digging into evals & found TruLens; recommend checking that out, too.
Understanding the 4 Main Approaches to LLM Evaluation (From Scratch)
Multiple-Choice Benchmarks, Verifiers, Leaderboards, and LLM Judges with Code Examples
open.substack.com
October 5, 2025 at 4:20 PM
Always generous with his time, here is another thorough explainer from @sebastianraschka.com : LLM Evaluation, with code!! open.substack.com/pub/sebastia...
Coincidentally, I also was just digging into evals & found TruLens; recommend checking that out, too.
Coincidentally, I also was just digging into evals & found TruLens; recommend checking that out, too.
An under-appreciated aspect of Gen AI apps is the process of testing them. That step exposes weaknesses in your business foundations - sloppy data, poor governance, ill-designed security, incomplete documentation, unknowledgeable knowledge workers, etc. AI isn't magic so get your house in order.
September 30, 2025 at 5:41 PM
An under-appreciated aspect of Gen AI apps is the process of testing them. That step exposes weaknesses in your business foundations - sloppy data, poor governance, ill-designed security, incomplete documentation, unknowledgeable knowledge workers, etc. AI isn't magic so get your house in order.
Reposted by L Quera
You have four autonomy dials you can tune in your agentic AI system. Are you using them?
medium.com/google-cloud...
medium.com/google-cloud...
The Agency Spectrum: An AI Risk Management Framework
We can now build autonomous systems that pursue meaningful, high-level goals. Yet, for every inspiring success story, there is a…
medium.com
September 26, 2025 at 4:25 PM
You have four autonomy dials you can tune in your agentic AI system. Are you using them?
medium.com/google-cloud...
medium.com/google-cloud...
Wow, have you seen this new benchmark from OpenAI, GDPval? Anthropic Claude Opus 4.1 came in close to human parity on an eval of 44 professional occupations across the top 9 industries in US GDP. Gemini 2.5Pro scored about half as well and I'm so curious why. openai.com/index/gdpval/
Measuring the performance of our models on real-world tasks
We’re introducing GDPval, a new evaluation that measures model performance on economically valuable, real-world tasks across 44 occupations.
openai.com
September 29, 2025 at 5:10 PM
Wow, have you seen this new benchmark from OpenAI, GDPval? Anthropic Claude Opus 4.1 came in close to human parity on an eval of 44 professional occupations across the top 9 industries in US GDP. Gemini 2.5Pro scored about half as well and I'm so curious why. openai.com/index/gdpval/
"What Orwell feared were those who would ban books. What Huxley feared was that there would be no reason to ban a book because there would be no one who wanted to read one."
Neil Postman
I've had so many Orwell v Huxley debates*...never imagined we'd wind up with both at the same time.
Neil Postman
I've had so many Orwell v Huxley debates*...never imagined we'd wind up with both at the same time.
September 28, 2025 at 4:58 PM
"What Orwell feared were those who would ban books. What Huxley feared was that there would be no reason to ban a book because there would be no one who wanted to read one."
Neil Postman
I've had so many Orwell v Huxley debates*...never imagined we'd wind up with both at the same time.
Neil Postman
I've had so many Orwell v Huxley debates*...never imagined we'd wind up with both at the same time.
"We are so fixated on how technology will out-skill us that we miss the many ways that technology can de-skill us." Derek Thompson
www.derekthompson.org/p/the-end-of...
www.derekthompson.org/p/the-end-of...
The End of Thinking
The rise of AI's "thinking" machines is not the problem. The decline of thinking people is.
www.derekthompson.org
September 28, 2025 at 4:54 PM
"We are so fixated on how technology will out-skill us that we miss the many ways that technology can de-skill us." Derek Thompson
www.derekthompson.org/p/the-end-of...
www.derekthompson.org/p/the-end-of...
Interesting data on the use of LLMs - topics, geographies, user education levels, etc. - from both OpenAI
openai.com/index/how-pe...
and Anthropic:
www.anthropic.com/research/ant...
Growth acceleration coming from low-mid-income countries so will be fascinating to see impacts of AI dispersion.
openai.com/index/how-pe...
and Anthropic:
www.anthropic.com/research/ant...
Growth acceleration coming from low-mid-income countries so will be fascinating to see impacts of AI dispersion.
Anthropic Economic Index report: Uneven geographic and enterprise AI adoption
To study such patterns of early AI adoption, we extend the Anthropic Economic Index along two important dimensions, introducing a geographic analysis of Claude.ai conversations and a first-of-its-kind...
www.anthropic.com
September 23, 2025 at 6:17 PM
Interesting data on the use of LLMs - topics, geographies, user education levels, etc. - from both OpenAI
openai.com/index/how-pe...
and Anthropic:
www.anthropic.com/research/ant...
Growth acceleration coming from low-mid-income countries so will be fascinating to see impacts of AI dispersion.
openai.com/index/how-pe...
and Anthropic:
www.anthropic.com/research/ant...
Growth acceleration coming from low-mid-income countries so will be fascinating to see impacts of AI dispersion.
TIL there's another player in privacy-preserving ML! JAX-Privacy from Google-DeepMind enables differentially-private training. Add this to your toolbox, along with OpenMined and Flower. jax-privacy.readthedocs.io/en/latest/ Haven't tried it yet but docs include a LoRA fine-tuning example.
JAX Privacy documentation — JAX Privacy 0.4.0 documentation
jax-privacy.readthedocs.io
September 20, 2025 at 2:38 PM
TIL there's another player in privacy-preserving ML! JAX-Privacy from Google-DeepMind enables differentially-private training. Add this to your toolbox, along with OpenMined and Flower. jax-privacy.readthedocs.io/en/latest/ Haven't tried it yet but docs include a LoRA fine-tuning example.
Agent2Agent looks super useful! Thanks, @developers.google.com !
"Build with ADK (or any framework), equip with MCP (or any tool), and communicate with A2A to remote agents, local agents, and humans. "
a2a-protocol.org/latest/
"Build with ADK (or any framework), equip with MCP (or any tool), and communicate with A2A to remote agents, local agents, and humans. "
a2a-protocol.org/latest/
A2A Protocol
The Agent2Agent protocol is an open standard that allows different AI agents to securely communicate, collaborate, and solve complex problems together.
a2a-protocol.org
September 18, 2025 at 4:33 PM
Agent2Agent looks super useful! Thanks, @developers.google.com !
"Build with ADK (or any framework), equip with MCP (or any tool), and communicate with A2A to remote agents, local agents, and humans. "
a2a-protocol.org/latest/
"Build with ADK (or any framework), equip with MCP (or any tool), and communicate with A2A to remote agents, local agents, and humans. "
a2a-protocol.org/latest/
Simple yet hard wisdom from Ryan Holiday's blog: "Don’t become implicated in the ugliness. Don’t let it infect you. Don’t become cynical or bitter... Don’t let bad times make you a bad person." ryanholiday.net/you-must-avo...
You Must Avoid Getting Corrupted By This - RyanHoliday.net
I’m giving a talk in Austin next week (only a few tickets left) and San Diego in February. Grab seats and come see me! My study of history has led me to believe that there is a kind of dark matter ins...
ryanholiday.net
September 15, 2025 at 3:53 PM
Simple yet hard wisdom from Ryan Holiday's blog: "Don’t become implicated in the ugliness. Don’t let it infect you. Don’t become cynical or bitter... Don’t let bad times make you a bad person." ryanholiday.net/you-must-avo...
"The paradox of working with AI wizards is that competence and opacity rise together...embrace provisional trust...working with “good enough”...because perfect verification is becoming impossible." New from @emollick.bsky.social on the changing relationship with agents: substack.com/app-link/pos...
On Working with Wizards
Verifying magic on the jagged frontier
substack.com
September 14, 2025 at 4:14 PM
"The paradox of working with AI wizards is that competence and opacity rise together...embrace provisional trust...working with “good enough”...because perfect verification is becoming impossible." New from @emollick.bsky.social on the changing relationship with agents: substack.com/app-link/pos...
If you're struggling to get your agents to factor complex business processes into their tool-calling, Knowledge Graphs are a great idea. See SAP's course on DeepLearningAI to learn more from the business process experts: learn.deeplearning.ai/courses/know...
Knowledge Graphs for AI Agent: API Discovery - DeepLearning.AI
Construct a knowledge graph and use it to enable your AI agent to find and call the right APIs in the right order.
learn.deeplearning.ai
September 13, 2025 at 8:50 PM
If you're struggling to get your agents to factor complex business processes into their tool-calling, Knowledge Graphs are a great idea. See SAP's course on DeepLearningAI to learn more from the business process experts: learn.deeplearning.ai/courses/know...
“All the manifestations of the world of measurement, the winning & losing..., all are based on a single assumption that is hidden from our awareness.The assumption is that life is about staying alive & making it through, surviving in a world of scarcity and peril.”The Art of Possibility h/t tferriss
September 13, 2025 at 8:41 PM
“All the manifestations of the world of measurement, the winning & losing..., all are based on a single assumption that is hidden from our awareness.The assumption is that life is about staying alive & making it through, surviving in a world of scarcity and peril.”The Art of Possibility h/t tferriss
Excellent post by @sebastianraschka.com on the Qwen architecture, with code! magazine.sebastianraschka.com/p/qwen3-from...
Understanding and Implementing Qwen3 From Scratch
A Detailed Look at One of the Leading Open-Source LLMs
magazine.sebastianraschka.com
September 7, 2025 at 8:24 PM
Excellent post by @sebastianraschka.com on the Qwen architecture, with code! magazine.sebastianraschka.com/p/qwen3-from...
The more we see examples of bot misuse, some with tragic consequences, the more we need to focus on security rather than pure functionality. I found this as an option to explore: ai.meta.com/research/pub...
LlamaFirewall: An open source guardrail system for building secure AI agents | Research - AI at Meta
Large language models (LLMs) have evolved from simple chatbots into autonomous agents capable of performing complex tasks such as editing production...
ai.meta.com
September 4, 2025 at 6:15 PM
The more we see examples of bot misuse, some with tragic consequences, the more we need to focus on security rather than pure functionality. I found this as an option to explore: ai.meta.com/research/pub...
One thing you notice immediately with Gemini-2.5-pro is that the context management is insanely good. That's going to become even more of a competitive differentiator as we move our app design mindset to context engineering. Latency may be weaker but personally, I'll take quality over speed.
August 18, 2025 at 4:40 PM
One thing you notice immediately with Gemini-2.5-pro is that the context management is insanely good. That's going to become even more of a competitive differentiator as we move our app design mindset to context engineering. Latency may be weaker but personally, I'll take quality over speed.
I'm continually delighted by Gemini-2.5-pro. It handles whatever I throw at it like a pro (aptly named 😀) and is just plain pleasant to work with. I haven't played with GPT5 much yet simply because I'm happy where I am. I want to compare to ensure I leverage the best but how much better could it be?
August 18, 2025 at 4:37 PM
I'm continually delighted by Gemini-2.5-pro. It handles whatever I throw at it like a pro (aptly named 😀) and is just plain pleasant to work with. I haven't played with GPT5 much yet simply because I'm happy where I am. I want to compare to ensure I leverage the best but how much better could it be?
As someone with many combat badges from the BPR wars, this post from @emollick.bsky.social resonated. open.substack.com/pub/oneusefu... Leverage the power of AI as alien intelligence, a fresh perspective on old problems. Oh, the possibilities!
The Bitter Lesson versus The Garbage Can
Does process matter? We are about to find out.
open.substack.com
July 28, 2025 at 2:34 PM
As someone with many combat badges from the BPR wars, this post from @emollick.bsky.social resonated. open.substack.com/pub/oneusefu... Leverage the power of AI as alien intelligence, a fresh perspective on old problems. Oh, the possibilities!
I know little of sports but I'd love to know how big league coaches do it.How do they create a team with psychological safety when star talent is rewarded so differently?One of my greatest joys is building happy, performant teams but they've been teams of equals.Been too long since I've read Wooden.
July 23, 2025 at 4:53 PM
I know little of sports but I'd love to know how big league coaches do it.How do they create a team with psychological safety when star talent is rewarded so differently?One of my greatest joys is building happy, performant teams but they've been teams of equals.Been too long since I've read Wooden.
The recent "hacquihires" are so depressing. The impact on teamwork will be immediate. If only star talent wins, everyone needs to jockey for position with elbows out, fighting for every ounce of credit. Of course, everyone has always been incentivized to build a brand but this feels different.
July 23, 2025 at 4:46 PM
The recent "hacquihires" are so depressing. The impact on teamwork will be immediate. If only star talent wins, everyone needs to jockey for position with elbows out, fighting for every ounce of credit. Of course, everyone has always been incentivized to build a brand but this feels different.