@ahmedbax.bsky.social
💭 Large language models sound like they reason — but what if that logic is a mirage?
A new Dispatch post explores why Chain-of-Thought breaks the moment data shifts.
🔗 The Mirage of Reasoning
open.substack.com/pub/ahmedkha...
#AIReasoning #ChainOfThought #LLMs #TheAKDispatch
A new Dispatch post explores why Chain-of-Thought breaks the moment data shifts.
🔗 The Mirage of Reasoning
open.substack.com/pub/ahmedkha...
#AIReasoning #ChainOfThought #LLMs #TheAKDispatch
External Redirection | LinkedIn
www.linkedin.com
November 11, 2025 at 7:23 PM
💭 Large language models sound like they reason — but what if that logic is a mirage?
A new Dispatch post explores why Chain-of-Thought breaks the moment data shifts.
🔗 The Mirage of Reasoning
open.substack.com/pub/ahmedkha...
#AIReasoning #ChainOfThought #LLMs #TheAKDispatch
A new Dispatch post explores why Chain-of-Thought breaks the moment data shifts.
🔗 The Mirage of Reasoning
open.substack.com/pub/ahmedkha...
#AIReasoning #ChainOfThought #LLMs #TheAKDispatch
Between 2025-2030, four constraints converge to force AI's biggest architectural pivot:
• Data exhaustion
• Energy costs
• Semiconductor limits
• ROI demands
Why the scaling hypothesis collapses—and what comes next.
Part 4: buff.ly/tSLE7gr
#AI #MachineLearning #AIResearch #TechStrategy
• Data exhaustion
• Energy costs
• Semiconductor limits
• ROI demands
Why the scaling hypothesis collapses—and what comes next.
Part 4: buff.ly/tSLE7gr
#AI #MachineLearning #AIResearch #TechStrategy
Part 4 – Why Compositional AI Will Actually Happen
Economic reality, not philosophical argument, will force the transition to compositional AI
open.substack.com
October 30, 2025 at 8:32 PM
Between 2025-2030, four constraints converge to force AI's biggest architectural pivot:
• Data exhaustion
• Energy costs
• Semiconductor limits
• ROI demands
Why the scaling hypothesis collapses—and what comes next.
Part 4: buff.ly/tSLE7gr
#AI #MachineLearning #AIResearch #TechStrategy
• Data exhaustion
• Energy costs
• Semiconductor limits
• ROI demands
Why the scaling hypothesis collapses—and what comes next.
Part 4: buff.ly/tSLE7gr
#AI #MachineLearning #AIResearch #TechStrategy
I just published an essay on why the current approach of scaling LLMs is failing to create reasoning agents.
#ArtificialIntelligence #MachineLearning #LargeLanguageModels #AIResearch #AIScaling #EmergentAbilities #AGI #AILimitations #DeepLearning
open.substack.com/pub/ahmedkha...
#ArtificialIntelligence #MachineLearning #LargeLanguageModels #AIResearch #AIScaling #EmergentAbilities #AGI #AILimitations #DeepLearning
open.substack.com/pub/ahmedkha...
Part 2: Sophisticated Statistics, Not Understanding
The technical reality of LLMs and why the scaling hypothesis is collapsing
open.substack.com
October 22, 2025 at 1:22 PM
I just published an essay on why the current approach of scaling LLMs is failing to create reasoning agents.
#ArtificialIntelligence #MachineLearning #LargeLanguageModels #AIResearch #AIScaling #EmergentAbilities #AGI #AILimitations #DeepLearning
open.substack.com/pub/ahmedkha...
#ArtificialIntelligence #MachineLearning #LargeLanguageModels #AIResearch #AIScaling #EmergentAbilities #AGI #AILimitations #DeepLearning
open.substack.com/pub/ahmedkha...
Reposted
Have you encountered definitions of AGI that avoid these philosophical problems? What criteria would you use to determine if a system possessed “general intelligence”?
open.substack.com/pub/ahmedkha...
#AI #Philosophy #Technology
open.substack.com/pub/ahmedkha...
#AI #Philosophy #Technology
Part 1: The Unfathomable Pursuing the Indefinable
Beyond the AGI Mirage | A Four-Part Series
open.substack.com
October 18, 2025 at 10:32 AM
Have you encountered definitions of AGI that avoid these philosophical problems? What criteria would you use to determine if a system possessed “general intelligence”?
open.substack.com/pub/ahmedkha...
#AI #Philosophy #Technology
open.substack.com/pub/ahmedkha...
#AI #Philosophy #Technology
Ahmed Khamassi
ahmedkhamassi.substack.com?r=f2siz&utm_...
ahmedkhamassi.substack.com?r=f2siz&utm_...
Ahmed Khamassi | Substack
PhD engineer turned PE partner. I architect AI transformations and write about technology, strategy, and the human side of change—bridging academic rigour with commercial impact across industries and ...
ahmedkhamassi.substack.com
October 12, 2025 at 10:25 AM
Ahmed Khamassi
ahmedkhamassi.substack.com?r=f2siz&utm_...
ahmedkhamassi.substack.com?r=f2siz&utm_...
In the US, politicians, institutions, and the people submitted to fascism.
Let’s hope Europeans show more courage against this wave of hatred and extremism.
#democracy #fascism
Let’s hope Europeans show more courage against this wave of hatred and extremism.
#democracy #fascism
October 8, 2025 at 10:00 PM
In the US, politicians, institutions, and the people submitted to fascism.
Let’s hope Europeans show more courage against this wave of hatred and extremism.
#democracy #fascism
Let’s hope Europeans show more courage against this wave of hatred and extremism.
#democracy #fascism
2022–24 was the era of scale. 2025 begins the era of control.
LLMs exploded, optimisation returned, RL’s recovering.
AI research is shifting from bigger to better.
📊 Read the full data-driven story:
From Scale to Control (open.substack.com/pub/ahmedkha...)
#AI #ML #LLMs #Research
LLMs exploded, optimisation returned, RL’s recovering.
AI research is shifting from bigger to better.
📊 Read the full data-driven story:
From Scale to Control (open.substack.com/pub/ahmedkha...)
#AI #ML #LLMs #Research
October 8, 2025 at 9:46 PM
2022–24 was the era of scale. 2025 begins the era of control.
LLMs exploded, optimisation returned, RL’s recovering.
AI research is shifting from bigger to better.
📊 Read the full data-driven story:
From Scale to Control (open.substack.com/pub/ahmedkha...)
#AI #ML #LLMs #Research
LLMs exploded, optimisation returned, RL’s recovering.
AI research is shifting from bigger to better.
📊 Read the full data-driven story:
From Scale to Control (open.substack.com/pub/ahmedkha...)
#AI #ML #LLMs #Research
Reposted
Wind farms using AI to coordinate turbines like a flock of birds—upstream turbines steering wakes away from neighbours. Results: 8-33% more power, decisions in <30 secs, adaptive to changing winds. Still challenges with rapid shifts, but DRL is reshaping renewable energy. 🌬️
#WindEnergy #DRL #AI
#WindEnergy #DRL #AI
September 25, 2025 at 11:15 AM
Wind farms using AI to coordinate turbines like a flock of birds—upstream turbines steering wakes away from neighbours. Results: 8-33% more power, decisions in <30 secs, adaptive to changing winds. Still challenges with rapid shifts, but DRL is reshaping renewable energy. 🌬️
#WindEnergy #DRL #AI
#WindEnergy #DRL #AI
Your C-XOs think agentic AI is just ChatGPT
They're wrong.
Real production AI needs:
Orchestrated reasoning chains
MCPs + tools
Proprietary ML
Hallucination controls
Enterprise arch
It's accessible, but it's engineering.
#AgenticAI #EnterpriseAI #AITransformation #DigitalTransformation #MLOps
They're wrong.
Real production AI needs:
Orchestrated reasoning chains
MCPs + tools
Proprietary ML
Hallucination controls
Enterprise arch
It's accessible, but it's engineering.
#AgenticAI #EnterpriseAI #AITransformation #DigitalTransformation #MLOps
September 24, 2025 at 6:10 PM
Your C-XOs think agentic AI is just ChatGPT
They're wrong.
Real production AI needs:
Orchestrated reasoning chains
MCPs + tools
Proprietary ML
Hallucination controls
Enterprise arch
It's accessible, but it's engineering.
#AgenticAI #EnterpriseAI #AITransformation #DigitalTransformation #MLOps
They're wrong.
Real production AI needs:
Orchestrated reasoning chains
MCPs + tools
Proprietary ML
Hallucination controls
Enterprise arch
It's accessible, but it's engineering.
#AgenticAI #EnterpriseAI #AITransformation #DigitalTransformation #MLOps
Good humans dedicate their lives to exploring a small but vital part of what can make our lives better and healthier. They read, experiment, develop theories, analyse, face challenges with vast knowledge, publish their findings, undergo peer review, teach others, face failures, yet never give up.
"We were told thalidomide was a safe drug and it wasn't..."
Nigel Farage says he has 'no idea' if Donald Trump is right about paracetamol being linked to autism.
Nigel Farage says he has 'no idea' if Donald Trump is right about paracetamol being linked to autism.
September 24, 2025 at 4:48 PM
Good humans dedicate their lives to exploring a small but vital part of what can make our lives better and healthier. They read, experiment, develop theories, analyse, face challenges with vast knowledge, publish their findings, undergo peer review, teach others, face failures, yet never give up.
Reposted
LLMs dominate public attention 📈, but the fundamentals making them possible (reinforcement learning, deep RL) remain niche.
Are we over-indexing on LLM hype and missing the next breakthrough in the “quiet” methods?
#AI #LLM #ReinforcementLearning #DeepLearning #AITrends #TechLeadership #FutureOfAI
Are we over-indexing on LLM hype and missing the next breakthrough in the “quiet” methods?
#AI #LLM #ReinforcementLearning #DeepLearning #AITrends #TechLeadership #FutureOfAI
September 23, 2025 at 12:45 PM
LLMs dominate public attention 📈, but the fundamentals making them possible (reinforcement learning, deep RL) remain niche.
Are we over-indexing on LLM hype and missing the next breakthrough in the “quiet” methods?
#AI #LLM #ReinforcementLearning #DeepLearning #AITrends #TechLeadership #FutureOfAI
Are we over-indexing on LLM hype and missing the next breakthrough in the “quiet” methods?
#AI #LLM #ReinforcementLearning #DeepLearning #AITrends #TechLeadership #FutureOfAI
The #UK government showered Trump with honours, pomp and ceremony last week making him look respectable. Now, they’re telling us not to listen to his madness and nonsense about paracetamol. Make your minds up! @mrjamesob.bsky.social
September 23, 2025 at 7:12 AM
The #UK government showered Trump with honours, pomp and ceremony last week making him look respectable. Now, they’re telling us not to listen to his madness and nonsense about paracetamol. Make your minds up! @mrjamesob.bsky.social
Don’t you wish they challenged Trump on the same grounds back then?
There should be a high bar for politicians attacking the media, but maybe Ed Davey has half a point here?
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
September 22, 2025 at 7:24 PM
Don’t you wish they challenged Trump on the same grounds back then?
Reposted
There should be a high bar for politicians attacking the media, but maybe Ed Davey has half a point here?
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
September 22, 2025 at 4:46 PM
There should be a high bar for politicians attacking the media, but maybe Ed Davey has half a point here?
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
For example, the first three paragraphs of this BBC story repeat Reform's £243bn figure without saying that the thinktank behind the estimate now disowns it. (This is mentioned in paragraph 35)
Reposted
🧠 New AI architecture mimics brain's hierarchical processing: HRM achieves 40% on AGI benchmarks with just 27M params & 1K examples. Solves extreme Sudoku at 74.5% (GPT-4: 0%). Dual-module design enables deep reasoning whilst training 100x more efficiently.
📄 arxiv.org/html/2506.21734v3
#ai
📄 arxiv.org/html/2506.21734v3
#ai
September 22, 2025 at 11:20 AM
🧠 New AI architecture mimics brain's hierarchical processing: HRM achieves 40% on AGI benchmarks with just 27M params & 1K examples. Solves extreme Sudoku at 74.5% (GPT-4: 0%). Dual-module design enables deep reasoning whilst training 100x more efficiently.
📄 arxiv.org/html/2506.21734v3
#ai
📄 arxiv.org/html/2506.21734v3
#ai
Reposted
Last night at the dock-of-the-bay ANTIFA meeting after we watched the tide roll away, us terrorists were all in agreement we want to see ALL Americans enjoy basic human rights.
It got REALLY wild.
It got REALLY wild.
September 21, 2025 at 3:57 PM
Last night at the dock-of-the-bay ANTIFA meeting after we watched the tide roll away, us terrorists were all in agreement we want to see ALL Americans enjoy basic human rights.
It got REALLY wild.
It got REALLY wild.
“Anti-terror tsar ‘vibrating with worry’ over wave of political violence - The Times and The Sunday Times.”
Excuse me @thetimes.com , what about #elonmusk who literally called for it live on screen? Is the @thetimes.com distracting from the real far right problem?
Excuse me @thetimes.com , what about #elonmusk who literally called for it live on screen? Is the @thetimes.com distracting from the real far right problem?
September 21, 2025 at 5:40 PM
“Anti-terror tsar ‘vibrating with worry’ over wave of political violence - The Times and The Sunday Times.”
Excuse me @thetimes.com , what about #elonmusk who literally called for it live on screen? Is the @thetimes.com distracting from the real far right problem?
Excuse me @thetimes.com , what about #elonmusk who literally called for it live on screen? Is the @thetimes.com distracting from the real far right problem?
WILD: DeepSeek just published the first LLM paper in Nature magazine. Not a preprint. Not a blog post. Actual peer-reviewed Nature.
The breakthrough? Pure reinforcement learning without human training data. The model taught itself to reason through trial and error alone.
The breakthrough? Pure reinforcement learning without human training data. The model taught itself to reason through trial and error alone.
DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning - Nature
A new artificial intelligence model, DeepSeek-R1, is introduced, demonstrating that the reasoning abilities of large language models can be incentivized through pure reinforcement learning, removing…
www.nature.com
September 21, 2025 at 8:05 AM
WILD: DeepSeek just published the first LLM paper in Nature magazine. Not a preprint. Not a blog post. Actual peer-reviewed Nature.
The breakthrough? Pure reinforcement learning without human training data. The model taught itself to reason through trial and error alone.
The breakthrough? Pure reinforcement learning without human training data. The model taught itself to reason through trial and error alone.
Reposted
Everyone, from big business CEOs to small business owners to highly paid professionals and lowly paid service staff, has urged the Chancellor to tax someone else.
From the new Private Eye, out now.
From the new Private Eye, out now.
September 18, 2025 at 3:59 PM
Everyone, from big business CEOs to small business owners to highly paid professionals and lowly paid service staff, has urged the Chancellor to tax someone else.
From the new Private Eye, out now.
From the new Private Eye, out now.
Reposted
I cancelled #disneyplus for kowtowing to a mobster administration. #canceldisneyplus.
September 19, 2025 at 7:07 AM
I cancelled #disneyplus for kowtowing to a mobster administration. #canceldisneyplus.