Gemma 4 on Old GPUs: Why a $700 Used Card Beats $20,000 of Professional Hardware I benchmarked Google’s Gemma 4 across five GPU configurations spanning four hardware generations — from a decade-old Titan X to dual RTX…

Just published a benchmark of Google's Gemma 4 across 5 GPU configurations, from a 2016 Titan X to dual RTX A6000s (€20k of pro hardware).

For single-user local inference, a used RTX 3090 wins. 🧵

medium.com/ai-advances/...

Pretty much, yes — except the backlog now fits on a napkin and the sprint planning takes a coffee break instead of half a day.

The AI-Accelerated Spiral: What Replaces Agile When Building Gets Fast Methodologies don’t die. They lose the constraints that justified them. Here’s a practical framework for when AI compresses your…

Methodologies don't die — they lose their excuses. New piece on what replaces Agile when AI compresses your dev cycle to days.

medium.com/operations-r...

#AI #Develop

The Socratic Model: Why the Wisest AI is the One That Knows When to Ask. Moving beyond monolithic LLMs: A 3B-parameter hierarchical system that outperforms generalists through selective delegation.

🚀 New on Towards AI: The Socratic Model: Why the Wisest AI is the One That Knows When to Ask.

Moving beyond monolithic LLMs: A 3B-parameter hierarchical system that outperforms generalists through selective delegation.

medium.com/towards-arti...

#AI #LLM #Research #Agents #GenerativeAI

When AI Models Learn to Learn: Continuous Knowledge Without Catastrophic Forgetting Dream pruning enables multi-domain learning while preserving existing capabilities

When AI Models Learn to Learn

I taught a 1.5B model logic after math. It didn't forget. +18 points logic, math core intact, zero catastrophic errors.

Dream pruning as continual learning.

#MachineLearning #LLM #AI

medium.com/towards-arti...

Dream Pruning: What Happens When AI Models Sleep Biologically-inspired consolidation produces balanced intelligence in LLMs.

"Dream Pruning: What Happens When AI Models Sleep"

SVD compression inspired by deep sleep. The compressed 1.5B model beats the full one: 69.4% vs 56.9%. Zero catastrophic errors.

#MachineLearning #AI

medium.com/towards-arti...

The Death of Coding is an Illusion: A Field Guide to the AI Orchestration Era Stop treating LLMs like magic oracles. Start engineering them like distributed systems.

Coding isn't dying. The "code scribe" is.

Senior devs accept MORE AI code, not less. The real skill now is orchestration: breaking problems, delegating to AI agents, evaluating output.

Field guide from the R&D trenches 👇
medium.com/ai-advances/...

#AI #SoftwareEngineering #LLM #FutureOfWork

Good point, but the model doesn't just "know it's wrong" — it delegates to a deterministic calculator. So effective production accuracy = 33% solved internally + 100% tool accuracy on the rest ≈ 100%.

The 61% model never delegates. It hallucinates on the other 39% with full confidence.

What if AI Models Learned Like Humans Do? What happens when you stop asking an AI to be a calculator and start asking it to have intuition? We built and tested a four-phase…

What if AI Models Learned Like Humans Do?

Standard fine-tuning: 61% accuracy, 0% metacognition.
Progressive training: 33% accuracy, 100% metacognition.

Same model, same data.

pub.towardsai.net/what-if-ai-models-learned-like-humans-do-c69c19f29d0c

#MachineLearning #LLM #AI

#nature #photography #macro #wildlife #entomology #pollinators #bumblebee #naturephotography #macrography #shallowdepthoffield #bokeh #nature_brilliance #shotwithlove #wildlifeonearth

I asked Claude Opus 4.6 to invent a Commodore 64 game from scratch. It designed METEOR STORM, wrote 1,654 lines of C, debugged its own memory crash, and shipped in 90 minutes. One human hint.

github.com/dexmac221/C64AIToolChain

#AI #RetroComputing #Commodore64 #GameDev

Claude 4.6 and the Commodore 64: When an LLM Writes, Builds, and Playtests Its Own Game I gave an AI agent a 1982 computer, a cross-compiler, and an emulator. It read the docs, picked its own project, wrote an Arkanoid clone in…

AI writes C64 code → compiles → injects into VICE emulator → game plays itself → VLM screenshots and analyzes the screen → AI fixes bugs → repeat.

No human in the loop!

Tested with Claude 4.6 building Arkanoid from scratch.

🔗 ai.gopubby.com/claude-4-6-a...

#AI #Commodore64 #Claude #AIAgents

Social Gardens: Rethinking Digital Spaces for Human Connection Why the Architecture of Social Media Shapes Who We Become

Can AI help us be more human? 🤖

In my new article, I explore "Social Gardens": digital spaces where AI acts as a neutral facilitator (a gardener) to distribute voice and prevent hierarchies, rather than an algorithm optimizing for outrage.

#AI #SocialMedia

medium.com/@gianlucabai...

#genova #nervi #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

The Orchestrator's Workflow: How to Actually Work with AI Agents There is a heated debate currently circulating in the software engineering world. On one side,...

AI amplifies what you already are.

Chaotic? Faster chaos.
Disciplined? Powerful copilot.

How I actually work with AI Agents

dev.to/dexmac221/the-orchestrators-workflow-how-to-actually-work-with-ai-agents-2akk

#AIAgents #SoftwareDevelopment #CodingWithAI #DeveloperLife #Programming

#genova #nervi #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

#genova #nervi #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

Social Gardens: Rethinking Digital Spaces for Human Connection Why the Architecture of Social Media Shapes Who We Become

Can AI help us be more human? 🤖

In my new article, I explore "Social Gardens": digital spaces where AI acts as a neutral facilitator (a gardener) to distribute voice and prevent hierarchies, rather than an algorithm optimizing for outrage.

#AI #SocialMedia

medium.com/@gianlucabai...

Code as Compression: The End of the Implementation Bottleneck A recent article about Claude Code 2.1 made waves in the engineering community, though perhaps for...

Is coding dead? No, it's just being compressed.

In the age of AI agents, source code is becoming an ephemeral artifact—decompressed from high-entropy prompts.

Read my analysis on the Stochastic Compiler era:

dev.to/dexmac221/co...

#SoftwareEngineering #GenAI #LLM #DevCommunity

#luna #moon #moonlight #fullmoon #gorgeous #halfmoon #sky #sleep #moonrise #nighttime #nature #photooftheday #luna #nightsky #goodnightmoon #moonphases #night #lunas #moonlovers #light #photo

skyline from #recco #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

🎭 I fired the Director. The AI started writing itself.

"In Search of an Author" — multi-agent system where story emerges from character interactions, not central control.
Pirandello meets LLMs

👉 github.com/dexmac221/In...

#AI #MultiAgentSystems #OpenSource #GenerativeAI #EmergentBehavior

#recco #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

#recco #liguria #italy #art #sky #photography #colour #light #sunset #clouds #photooftheday

thanks, I'm not the author, i tested it and installed on a commodore 128 with CP/M

hi, yes, you can run vice with a CP/M disk, I tested here in warp mode.

Finally running native CP/M 3.0 on the Commodore 128! 💾

The Hardware way, a real benchmark for the system!

The Stack: 🛠️ C128 + 1571 Drive ⚡ Kung Fu Flash 💾 SD2IEC 💿 DraCopy (Sector Cloning)

That A> prompt is pure satisfaction. 🚀

#Commodore128 #CPM #RetroComputing #Z80 #C128 #8bit

Converting an AI-Generated image (nanobanana in this case) into a 3D-printable model is easy using:

sam 3D : aidemos.meta.com/segment-anyt...

Blender, Cura3D, and a good 3D Printer

#gemini #nanobanana #3dprinting #ai #artificialintelligence #sam3d #blender

AI Should Be "Blind" (And That's a Good Thing) Stop trying to give agents "eyes" to look at human interfaces. Give them terminals, deterministic...

Stop giving AI agents "eyes" to click buttons. Give them terminals.

Visual agents = 20 years of RPA fragility + hallucinations.
Blind agents = 100% success rate in my tests.

The future is headless.

dev.to/dexmac221/ai...

Happy New Year! #happynewyear2026 #happynewyears #luna #moon #moonlight #fullmoon #gorgeous #halfmoon #sky #sleep #moonrise #nighttime #nature #photooftheday #luna #nightsky #goodnightmoon #moonphases #night #lunas #moonlovers #light #photo