norabble.substack.com
@norabble.substack.com
An IT professional and software architect, with 30+ years of experience, ex-AWS, with a history advising clients in the financial industry. Author of https://substack.norabble.com.
Most people who use highly capable open-weight AI models have never touched the weight file. Users reach models through APIs deployed by someone else.
September 29, 2026 at 9:01 PM
Most people who use highly capable open-weight AI models have never touched the weight file. Users reach models through APIs deployed by someone else.
AI policy is in its infancy. Calls for a direction are coalescing. But cynicism, and other distractions are avoiding the core questions.
September 22, 2026 at 5:00 PM
AI policy is in its infancy. Calls for a direction are coalescing. But cynicism, and other distractions are avoiding the core questions.
Security backlogs are more critical than ever. We can talk, but how do we translate that to action?
Organizations resist change. So you reward them for it. But that reward will bend. You'll get some of what you sought, but something else too, and over time, less of what you sought.
Organizations resist change. So you reward them for it. But that reward will bend. You'll get some of what you sought, but something else too, and over time, less of what you sought.
September 1, 2026 at 11:45 AM
Security backlogs are more critical than ever. We can talk, but how do we translate that to action?
Organizations resist change. So you reward them for it. But that reward will bend. You'll get some of what you sought, but something else too, and over time, less of what you sought.
Organizations resist change. So you reward them for it. But that reward will bend. You'll get some of what you sought, but something else too, and over time, less of what you sought.
The HuggingFace Incident shows AI attacks are possible, why haven't you seen them?
— training alignment
— security classifiers
— abuse detection at the labs
— your own conventional security
You can hope and pray the first three are done right, but the fourth is what you can control.
— training alignment
— security classifiers
— abuse detection at the labs
— your own conventional security
You can hope and pray the first three are done right, but the fourth is what you can control.
August 17, 2026 at 8:46 PM
The HuggingFace Incident shows AI attacks are possible, why haven't you seen them?
— training alignment
— security classifiers
— abuse detection at the labs
— your own conventional security
You can hope and pray the first three are done right, but the fourth is what you can control.
— training alignment
— security classifiers
— abuse detection at the labs
— your own conventional security
You can hope and pray the first three are done right, but the fourth is what you can control.
We need to take security more seriously. OpenAI, Anthropic, and more are demonstrating this with their own missteps. Transparency is key here. If I appreciate anything here it's that. We must allow for honesty. I explain, in my latest post, my most widely read so far.
August 5, 2026 at 3:03 PM
We need to take security more seriously. OpenAI, Anthropic, and more are demonstrating this with their own missteps. Transparency is key here. If I appreciate anything here it's that. We must allow for honesty. I explain, in my latest post, my most widely read so far.
OpenAI patched the sandbox flaw. Sandbox fixes are not enough. The response should be many layered. Alignment is the big picture. More are in between. I discuss observer models, and ask, where were they? What is the standard for internal model deployments?
substack.norabble.com/p/an-openai-...
substack.norabble.com/p/an-openai-...
An OpenAI Model Escaped Its Sandbox. Where Was the Observer?
OpenAI accidentally hacked Hugging Face. A basic layer of defense appears to have been missing, and no one has explained why.
substack.norabble.com
July 23, 2026 at 8:59 PM
OpenAI patched the sandbox flaw. Sandbox fixes are not enough. The response should be many layered. Alignment is the big picture. More are in between. I discuss observer models, and ask, where were they? What is the standard for internal model deployments?
substack.norabble.com/p/an-openai-...
substack.norabble.com/p/an-openai-...
We've quietly agreed that "slop" means "content made by AI." I think that definition is a trap. And it's steering us toward the wrong fix. 🧵
July 21, 2026 at 12:20 PM
We've quietly agreed that "slop" means "content made by AI." I think that definition is a trap. And it's steering us toward the wrong fix. 🧵
The medical consensus says pervasive testing fuels overdiagnosis and hypochondria. I get the worry—but it misses how data compounds. One cheap test is noise; a decade of them is signal. My case for cheap, pervasive testing: substack.norabble.com/p/more-data-... #healthtech
More Data, Please
Rethink Medical Backlash Against Pervasive Testing
substack.norabble.com
June 30, 2026 at 12:02 PM
The medical consensus says pervasive testing fuels overdiagnosis and hypochondria. I get the worry—but it misses how data compounds. One cheap test is noise; a decade of them is signal. My case for cheap, pervasive testing: substack.norabble.com/p/more-data-... #healthtech
Everyone has a hot take on #AI. Awareness of the reality of software development outside the field is skin deep.
On screen, programmers appear ~300x less than doctors. That gap could be smaller, if we tried.
substack.norabble.com/p/the-invisi...
On screen, programmers appear ~300x less than doctors. That gap could be smaller, if we tried.
substack.norabble.com/p/the-invisi...
The Invisible Profession
Opinions about AI are outrunning our understanding of the software beneath it
substack.norabble.com
June 23, 2026 at 12:35 PM
Everyone has a hot take on #AI. Awareness of the reality of software development outside the field is skin deep.
On screen, programmers appear ~300x less than doctors. That gap could be smaller, if we tried.
substack.norabble.com/p/the-invisi...
On screen, programmers appear ~300x less than doctors. That gap could be smaller, if we tried.
substack.norabble.com/p/the-invisi...
Six months ago I mapped the AI build-out as a four-layer gamble. Six months later significant hurdles are passed, but there's more to come. Capex is racing $410B → $700B → ~$1T. Today's revenue justifies last year's bet, not next year's.
substack.norabble.com/p/the-ai-gam...
#AI #AIbubble
substack.norabble.com/p/the-ai-gam...
#AI #AIbubble
The AI Gamble, Six Months On
A layer-by-layer look at the AI value chain — chips to applications — finds firmer footing on capex, but a mountain only half climbed.
substack.norabble.com
June 16, 2026 at 12:14 PM
Six months ago I mapped the AI build-out as a four-layer gamble. Six months later significant hurdles are passed, but there's more to come. Capex is racing $410B → $700B → ~$1T. Today's revenue justifies last year's bet, not next year's.
substack.norabble.com/p/the-ai-gam...
#AI #AIbubble
substack.norabble.com/p/the-ai-gam...
#AI #AIbubble
The AI bubble debate has a new form: the 'token bubble.' Coming from software with a long interest in economics, I think the framing is off before we get the evidence. A thread on what's actually going on. 🧵 1/
Are We in a Token Bubble?
The wrong question — and a better one for reading the AI boom.
substack.norabble.com
June 9, 2026 at 12:19 PM
The AI bubble debate has a new form: the 'token bubble.' Coming from software with a long interest in economics, I think the framing is off before we get the evidence. A thread on what's actually going on. 🧵 1/
Hiring became a war nobody declared. Employers filtered to cut costs. Applicants gamed the filters. AI removed the last natural brake. Now it's a prisoner's dilemma, one-sided fixes only escalate it. The exit is collaborative.
substack.norabble.com/p/hirings-ac...
#hiring
substack.norabble.com/p/hirings-ac...
#hiring
Hiring's Accidental War
One-sided fixes become weapons. The exit is collaborative.
substack.norabble.com
June 2, 2026 at 11:48 AM
Hiring became a war nobody declared. Employers filtered to cut costs. Applicants gamed the filters. AI removed the last natural brake. Now it's a prisoner's dilemma, one-sided fixes only escalate it. The exit is collaborative.
substack.norabble.com/p/hirings-ac...
#hiring
substack.norabble.com/p/hirings-ac...
#hiring
'Bachelor's degree or equivalent experience' is logically redundant AND AI hiring tools exploit the ambiguity against candidates. New piece on why lazy hiring is getting worse — and why the recent grad unemployment panic is based on misread data: substack.norabble.com/p/or-equival...
#hiring #skills
#hiring #skills
Or Equivalent Experience
Lazy Mistakes in Hiring and the Truth Behind Jobs Data
substack.norabble.com
May 26, 2026 at 11:43 AM
'Bachelor's degree or equivalent experience' is logically redundant AND AI hiring tools exploit the ambiguity against candidates. New piece on why lazy hiring is getting worse — and why the recent grad unemployment panic is based on misread data: substack.norabble.com/p/or-equival...
#hiring #skills
#hiring #skills
AI safety underfunding isn't negligence — it's what liability caps produce. The incentive gap is quantifiable, and quantifying it points toward what corrective policy should look like. substack.norabble.com/p/ai-safety-... #AISafety #AIPolicy
AI Safety Is Underfunded by Design
A Model for Incentive-Aligned AI Safety Policy
substack.norabble.com
May 19, 2026 at 1:58 PM
AI safety underfunding isn't negligence — it's what liability caps produce. The incentive gap is quantifiable, and quantifying it points toward what corrective policy should look like. substack.norabble.com/p/ai-safety-... #AISafety #AIPolicy
The window to govern AI misuse is open—but won't stay that way. Three compounding problems: rogue jurisdictions, unmonitorable open models, internet anonymity. Hard tradeoffs required.
#AIPolicy #AISafety
substack.norabble.com/p/challenges...
#AIPolicy #AISafety
substack.norabble.com/p/challenges...
Challenges for AI Misuse Prevention
Jurisdictions, Open Models, and Privacy
substack.norabble.com
May 12, 2026 at 12:11 PM
The window to govern AI misuse is open—but won't stay that way. Three compounding problems: rogue jurisdictions, unmonitorable open models, internet anonymity. Hard tradeoffs required.
#AIPolicy #AISafety
substack.norabble.com/p/challenges...
#AIPolicy #AISafety
substack.norabble.com/p/challenges...
Gas prices are up, and the "what-aboutism" regarding the 2022 spike has returned. But equating the two isn't just a political stretch—it’s a mechanical failure to understand supply vs. policy. Let’s look at the first principles of why these spikes happen. 🧵
https://substack.norabble.com/p/supply-demand-and-deflection]
May 8, 2026 at 11:36 AM
Gas prices are up, and the "what-aboutism" regarding the 2022 spike has returned. But equating the two isn't just a political stretch—it’s a mechanical failure to understand supply vs. policy. Let’s look at the first principles of why these spikes happen. 🧵
Interesting shift happening in the software landscape: AI is disrupting both the commercial and security models of open vs. closed source.
AI reverse-engineering diminishes closed-source moats, while AI scanning tokens easily outpace the open-source "million eyeballs" security advantage.
AI reverse-engineering diminishes closed-source moats, while AI scanning tokens easily outpace the open-source "million eyeballs" security advantage.
Open-Source in the AI Era
Choices and layers
substack.norabble.com
May 3, 2026 at 10:21 PM
Interesting shift happening in the software landscape: AI is disrupting both the commercial and security models of open vs. closed source.
AI reverse-engineering diminishes closed-source moats, while AI scanning tokens easily outpace the open-source "million eyeballs" security advantage.
AI reverse-engineering diminishes closed-source moats, while AI scanning tokens easily outpace the open-source "million eyeballs" security advantage.
The loudest AI debates are stuck in a binary. Doubters see failure; doomers see apocalypse. They both miss the real shift: the move from deterministic to probabilistic systems.
The future isn't code or AI—it's code as the protocol for AI.
Full essay: substack.norabble.com/p/control-an...
The future isn't code or AI—it's code as the protocol for AI.
Full essay: substack.norabble.com/p/control-an...
Control and AI
Holding Tight and Letting Go
substack.norabble.com
April 28, 2026 at 12:00 PM
The loudest AI debates are stuck in a binary. Doubters see failure; doomers see apocalypse. They both miss the real shift: the move from deterministic to probabilistic systems.
The future isn't code or AI—it's code as the protocol for AI.
Full essay: substack.norabble.com/p/control-an...
The future isn't code or AI—it's code as the protocol for AI.
Full essay: substack.norabble.com/p/control-an...
The question is no longer "What can the chatbot say?" The real question is: How much scope are we willing to give to indeterminate systems?
My new 2-part series on AI, Determinism, and Control is live. 🧵👇
My new 2-part series on AI, Determinism, and Control is live. 🧵👇
April 7, 2026 at 12:37 PM
The question is no longer "What can the chatbot say?" The real question is: How much scope are we willing to give to indeterminate systems?
My new 2-part series on AI, Determinism, and Control is live. 🧵👇
My new 2-part series on AI, Determinism, and Control is live. 🧵👇
The biggest blind spot in the #AI debate: The fundamental state of an economy is to create new jobs.
Current tech layoffs are corrections for COVID over-hiring & interest rates. AI is a convenient mask—"AI-washing" by executives for a short-term stock boost.
substack.norabble.com/p/the-ai-job...
Current tech layoffs are corrections for COVID over-hiring & interest rates. AI is a convenient mask—"AI-washing" by executives for a short-term stock boost.
substack.norabble.com/p/the-ai-job...
The AI Jobs Blind Spot
Why Job Creation is the Default
substack.norabble.com
March 30, 2026 at 2:43 PM
The biggest blind spot in the #AI debate: The fundamental state of an economy is to create new jobs.
Current tech layoffs are corrections for COVID over-hiring & interest rates. AI is a convenient mask—"AI-washing" by executives for a short-term stock boost.
substack.norabble.com/p/the-ai-job...
Current tech layoffs are corrections for COVID over-hiring & interest rates. AI is a convenient mask—"AI-washing" by executives for a short-term stock boost.
substack.norabble.com/p/the-ai-job...
The weakest link in enterprise security right now isn't vulnerability discovery—it's deployment speed. AI is drastically accelerating the attacker-defender cycle, and our sluggish pipelines are handing them a permanent advantage. 🧵 1/3
March 23, 2026 at 12:57 PM
The weakest link in enterprise security right now isn't vulnerability discovery—it's deployment speed. AI is drastically accelerating the attacker-defender cycle, and our sluggish pipelines are handing them a permanent advantage. 🧵 1/3
In my new piece, I explore why AI "exposure" ignores economic demand. 📊
2x faster tasks yield different outcomes:
1️⃣ Bounded: Job cuts📉
2️⃣ Unbounded: Growth🚀
3️⃣ Adversarial: Arms race⚔️
Lumping them together masks the real cuts.
substack.norabble.com/p/ai-jobs-th...
2x faster tasks yield different outcomes:
1️⃣ Bounded: Job cuts📉
2️⃣ Unbounded: Growth🚀
3️⃣ Adversarial: Arms race⚔️
Lumping them together masks the real cuts.
substack.norabble.com/p/ai-jobs-th...
AI Jobs: The Hidden Rules of Demand
Predicting the future of work using Bounded, Unbounded, and Adversarial demand
substack.norabble.com
March 18, 2026 at 1:22 PM
In my new piece, I explore why AI "exposure" ignores economic demand. 📊
2x faster tasks yield different outcomes:
1️⃣ Bounded: Job cuts📉
2️⃣ Unbounded: Growth🚀
3️⃣ Adversarial: Arms race⚔️
Lumping them together masks the real cuts.
substack.norabble.com/p/ai-jobs-th...
2x faster tasks yield different outcomes:
1️⃣ Bounded: Job cuts📉
2️⃣ Unbounded: Growth🚀
3️⃣ Adversarial: Arms race⚔️
Lumping them together masks the real cuts.
substack.norabble.com/p/ai-jobs-th...
The financial gravity of #AI is the Application Layer—the industry's ultimate "Reality Check." We explore:
👁️Enterprise API blind spot
📊Consumer "Vibes" vs. Enterprise ROI
🛑Why we are incorrectly blaming "AI Slop" for downstream bottlenecks
Read it here: substack.norabble.com/p/ai-applica...
👁️Enterprise API blind spot
📊Consumer "Vibes" vs. Enterprise ROI
🛑Why we are incorrectly blaming "AI Slop" for downstream bottlenecks
Read it here: substack.norabble.com/p/ai-applica...
The AI Reality Check
Decoding the Application Layer
substack.norabble.com
February 26, 2026 at 9:19 PM
The financial gravity of #AI is the Application Layer—the industry's ultimate "Reality Check." We explore:
👁️Enterprise API blind spot
📊Consumer "Vibes" vs. Enterprise ROI
🛑Why we are incorrectly blaming "AI Slop" for downstream bottlenecks
Read it here: substack.norabble.com/p/ai-applica...
👁️Enterprise API blind spot
📊Consumer "Vibes" vs. Enterprise ROI
🛑Why we are incorrectly blaming "AI Slop" for downstream bottlenecks
Read it here: substack.norabble.com/p/ai-applica...
Every political ideology is a bet on certain "Technologies of #Trust" (like markets or bureaucracy) and a rejection of others.
But what if the goal isn't to pick the "right" ones, but to balance them all? Omitting any creates fragility. My new essay explains.
open.substack.com/pub/norabble...
But what if the goal isn't to pick the "right" ones, but to balance them all? Omitting any creates fragility. My new essay explains.
open.substack.com/pub/norabble...
The Technologies of Trust
How We Cooperate at Scale
open.substack.com
August 25, 2025 at 9:24 PM
Every political ideology is a bet on certain "Technologies of #Trust" (like markets or bureaucracy) and a rejection of others.
But what if the goal isn't to pick the "right" ones, but to balance them all? Omitting any creates fragility. My new essay explains.
open.substack.com/pub/norabble...
But what if the goal isn't to pick the "right" ones, but to balance them all? Omitting any creates fragility. My new essay explains.
open.substack.com/pub/norabble...
Is AI's impact about what it does, or what we do with it? I explore this by looking at advertising as a key case study, proposing "lower-judgment" guardrails & even AI to fight manipulative methods. #AIethics #AISociety #TechEthics #FutureofAI
substack.norabble.com/p/ai-and-the...
substack.norabble.com/p/ai-and-the...
AI and the Social Ecosystem
Beyond What AI Will Do
substack.norabble.com
August 9, 2025 at 6:13 PM
Is AI's impact about what it does, or what we do with it? I explore this by looking at advertising as a key case study, proposing "lower-judgment" guardrails & even AI to fight manipulative methods. #AIethics #AISociety #TechEthics #FutureofAI
substack.norabble.com/p/ai-and-the...
substack.norabble.com/p/ai-and-the...