arXiv:2510.25140v2 Announce Type: replace 
Abstract: Object detection in civil engineering applications is constrained by limited annotated data in specialized domains. We introduce DINO-YOLO, a hybrid architecture combining YOLOv12 with DINOv3 self-supervised vision transformers for data-efficient detection. DINOv3 features are strategically integrated at two locations: input preprocessing (P0) and mid-backbone enhancement (P3). Experimental validation demonstrates substantial improvements: Tunnel Segment Crack detection (648 images) achieves 12.4% improvement, Construction PPE (1K images) gains 13.7%, and KITTI (7K images) shows 88.6% improvement, while maintaining real-time inference (30-47 FPS). Systematic ablation across five YOLO scales and nine DINOv3 variants reveals that Medium-scale architectures achieve optimal performance with DualP0P3 integration (55.77% mAP@0.5), while Small-scale requires Triple Integration (53.63%). The 2-4x inference overhead (21-33ms versus 8-16ms baseline) remains acceptable for field deployment on NVIDIA RTX 5090. DINO-YOLO establishes state-of-the-art performance for civil engineering datasets (<10K images) while preserving computational efficiency, providing practical solutions for construction safety monitoring and infrastructure inspection in data-constrained environments.

تقديم DINO-YOLO يمثل تقدمًا كبيرًا في اكتشاف الأشياء في تطبيقات الهندسة المدنية، حيث يعالج تحدي نقص البيانات المعلّمة في المجالات المتخصصة. من خلال دمج بنية YOLOv12 مع محولات الرؤية الذاتية الإشراف DINOv3، يحسن هذا النهج المبتكر كفاءة البيانات ودقة الاكتشاف. تظهر النتائج التجريبية تحسينات كبيرة، مما يجعل DINO-YOLO حلاً واعدًا للمهنيين في الهندسة المدنية الذين يعتمدون على اكتشاف الأشياء بدقة في مشاريعهم.

La introducción de DINO-YOLO marca un avance significativa en la detección de objetos para la ingeniería civil, abordando el desafío de la limitada disponibilidad de datos anotados en campos especializados. Al combinar la arquitectura YOLOv12 con los transformadores de visión auto-supervisados DINOv3, este enfoque innovador mejora la eficiencia de los datos y la precisión de la detección. Los resultados experimentales muestran mejoras sustanciales, lo que convierte a DINO-YOLO en una solución prometedora para los profesionales de la ingeniería civil que dependen de la detección precisa de objetos en sus proyectos.

L'introduction de DINO-YOLO représente une avancée significative dans la détection d'objets pour le génie civil, en répondant au défi des données annotées limitées dans des domaines spécialisés. En combinant l'architecture YOLOv12 avec les transformateurs de vision auto-supervisés DINOv3, cette approche innovante améliore l'efficacité des données et la précision de la détection. Les résultats expérimentaux montrent des améliorations substantielles, faisant de DINO-YOLO une solution prometteuse pour les professionnels du génie civil qui dépendent d'une détection précise des objets pour leurs projets.

The introduction of DINO-YOLO marks a significant advancement in object detection for civil engineering, addressing the challenge of limited annotated data in specialized fields. By combining the YOLOv12 architecture with DINOv3 self-supervised vision transformers, this innovative approach enhances data efficiency and detection accuracy. The experimental results show substantial improvements, making DINO-YOLO a promising solution for professionals in civil engineering who rely on precise object detection for their projects.

DINO-YOLO: Self-Supervised Pre-training for Data-Efficient Object Detection in Civil Engineering Applications

Cursor 2.0 shifts to in-house AI with Composer model and parallel agents, OpenAI completed its for-profit restructuring, and more!

في الحلقة الأخيرة من بودكاست LWiAI، يناقش المضيفون التطورات المهمة في صناعة الذكاء الاصطناعي، بما في ذلك انتقال OpenAI إلى نموذج ربحي وإطلاق Cursor 2.0، الذي يتميز بذكاء اصطناعي داخلي مع نموذج Composer ووكلاء متوازيين. تعكس هذه التغييرات المشهد المتطور للذكاء الاصطناعي وتبرز زيادة تجارية تقنيات الذكاء الاصطناعي، مما يجعلها موضوعًا حاسمًا لأي شخص مهتم بمستقبل هذا المجال.

En el último episodio del podcast LWiAI, los anfitriones discuten desarrollos significativos en la industria de la IA, incluida la transición de OpenAI a un modelo con fines de lucro y el lanzamiento de Cursor 2.0, que presenta una IA interna con un modelo Composer y agentes paralelos. Estos cambios reflejan el paisaje en evolución de la inteligencia artificial y destacan la creciente comercialización de las tecnologías de IA, convirtiéndolo en un tema crucial para cualquier persona interesada en el futuro de este campo.

Dans le dernier épisode du podcast LWiAI, les animateurs discutent des développements importants dans l'industrie de l'IA, y compris la transition d'OpenAI vers un modèle à but lucratif et le lancement de Cursor 2.0, qui propose une IA interne avec un modèle Composer et des agents parallèles. Ces changements reflètent l'évolution du paysage de l'intelligence artificielle et soulignent la commercialisation croissante des technologies IA, ce qui en fait un sujet crucial pour quiconque s'intéresse à l'avenir de ce domaine.

In the latest episode of the LWiAI Podcast, the hosts discuss significant developments in the AI industry, including OpenAI's transition to a for-profit model and the launch of Cursor 2.0, which features an in-house AI with a Composer model and parallel agents. These changes reflect the evolving landscape of artificial intelligence and highlight the increasing commercialization of AI technologies, making it a crucial topic for anyone interested in the future of this field.

LWiAI Podcast #224 - OpenAI is for-profit! Cursor 2, Minimax M2, Udio copyright

Picture this: You're staring at a 3000-line Terraform file at 2 AM, wondering why your simple VPC deployment is taking 45 minutes and why changing one variable broke half your infrastructure. Sound familiar? 🤔

Welcome to the wonderful world of Infrastructure as Code gone wrong! But fear not, dear reader. After 20 years of watching developers turn elegant infrastructure into digital spaghetti, I'm here to share the secret sauce that separates the Terraform wizards from the mere mortals.

Today, we're diving deep into the art of writing efficient, maintainable Terraform code that won't make your future self (or your teammates) want to throw their laptop out the window. Let's transform your infrastructure from a house of cards into a solid foundation that scales! 🏗️

<h2>
 
 
 1. Modularity: Or How Not to End Up with a Terraform Monolith 🧱
</h2>

Remember that colleague who created a single <code>main.tf</code> file with 5,847 lines and proudly called it "well-organized"? Don't be that person! 😅

Modularity is like cooking - you don't throw all ingredients into one giant pot and hope for the best. You create distinct, reusable components that work together harmoniously.

<h3>
 
 
 The Magical Structure
</h3>

Here's a lesser-known fact: Terraform modules can be versioned and published to private registries, just like your favorite npm packages! This means you can treat your infrastructure components like proper software libraries. 


<div class="highlight js-code-highlight">
<pre class="highlight hcl"><code># modules/vpc/main.tf
resource "aws_vpc" "main" {
 cidr_block = var.cidr_block
 enable_dns_hostnames = var.enable_dns_hostnames
 enable_dns_support = var.enable_dns_support

 tags = merge(
 var.common_tags,
 {
 Name = var.vpc_name
 }
 )
}

# modules/vpc/variables.tf
variable "cidr_block" {
 description = "CIDR block for the VPC"
 type = string
 validation {
 condition = can(cidrhost(var.cidr_block, 0))
 error_message = "The cidr_block must be a valid CIDR block."
 }
}
</code></pre>

</div>



<h3>
 
 
 The Organization That Saves Lives
</h3>

A well-structured project looks like this: 


<div class="highlight js-code-highlight">
<pre class="highlight plaintext"><code>terraform/
├── modules/
│ ├── vpc/
│ ├── compute/
│ └── database/
├── environments/
│ ├── dev/
│ ├── staging/
│ └── prod/
└── shared/
 ├── data.tf
 └── providers.tf
</code></pre>

</div>



Pro tip: According to HashiCorp's own statistics, teams using proper module organization reduce their deployment errors by 67% and cut development time by 40%! 📊

<h2>
 
 
 2. The Art of Reusability: DRY Applied to Infrastructure ♻️
</h2>

Ah, the classic tale of the developer who copy-pasted the same resource block 47 times with slight variations. Each time they needed a new environment, ctrl+c, ctrl+v, change a few values, and voilà! Until one day they needed to update all 47 instances... 😱

<h3>
 
 
 Variables and Locals: Your New Best Friends
</h3>

Here's a mind-blowing fact: You can use Terraform's <code>locals</code> block to create computed values that adapt based on your environment, making your code incredibly flexible: 


<div class="highlight js-code-highlight">
<pre class="highlight hcl"><code>locals {
 environment_config = {
 dev = {
 instance_type = "t3.micro"
 min_size = 1
 max_size = 2
 }
 prod = {
 instance_type = "t3.large" 
 min_size = 3
 max_size = 10
 }
 }

 current_config = local.environment_config[var.environment]

 # This awesome technique automatically calculates subnets!
 availability_zones = data.aws_availability_zones.available.names
 subnet_cidrs = [
 for index, az in local.availability_zones :
 cidrsubnet(var.vpc_cidr, 8, index)
 ]
}
</code></pre>

</div>



<h3>
 
 
 The Power of Data Sources
</h3>

Data sources are like having a conversation with your existing infrastructure: 


<div class="highlight js-code-highlight">
<pre class="highlight hcl"><code>data "aws_ami" "latest_amazon_linux" {
 most_recent = true
 owners = ["amazon"]

 filter {
 name = "name"
 values = ["amzn2-ami-hvm-*-x86_64-gp2"]
 }
}

# Now your AMI is always automatically up to date!
resource "aws_instance" "web" {
 ami = data.aws_ami.latest_amazon_linux.id
 instance_type = local.current_config.instance_type
}
</code></pre>

</div>



Surprising stat: Teams using data sources properly have 23% fewer "it works on my machine" incidents! 🎯

<h2>
 
 
 3. Maintainability: Make Sure Your Future Self Doesn't Curse You 🔮
</h2>

Picture this horror story: A developer creates perfect infrastructure, documents nothing, uses default state files, and then leaves the company. Six months later, nobody dares touch the infrastructure because it's like defusing a bomb blindfolded! 💣

<h3>
 
 
 State Management: The Foundation of Everything
</h3>

Here's a secret weapon most developers don't know: Terraform has a <code>moved</code> block that lets you refactor your resources without destroying them! 


<div class="highlight js-code-highlight">
<pre class="highlight hcl"><code># Backend configuration - TOUJOURS distant !
terraform {
 backend "s3" {
 bucket = "my-terraform-state-bucket"
 key = "prod/terraform.tfstate"
 region = "eu-west-1"
 encrypt = true
 dynamodb_table = "terraform-state-locks"
 }
}

# The moved block - your savior for refactors!
moved {
 from = aws_instance.web
 to = module.compute.aws_instance.web
}
</code></pre>

</div>



<h3>
 
 
 Documentation: Your Life Insurance
</h3>

Smart documentation isn't just comments - it's self-documenting code: 


<div class="highlight js-code-highlight">
<pre class="highlight hcl"><code>variable "database_config" {
 description = &lt;&lt;-EOT
 Configuration for the RDS database instance.

 Example:
 database_config = {
 engine = "postgres"
 engine_version = "13.7"
 instance_class = "db.t3.micro"
 allocated_storage = 20
 }
 EOT

 type = object({
 engine = string
 engine_version = string
 instance_class = string
 allocated_storage = number
 })

 validation {
 condition = contains(["postgres", "mysql"], var.database_config.engine)
 error_message = "Database engine must be either 'postgres' or 'mysql'."
 }
}
</code></pre>

</div>



<h3>
 
 
 Secret Testing
</h3>

Hidden gem: You can use <code>terraform plan -detailed-exitcode</code> in your CI/CD to detect if there are changes to apply. Exit code 2 means changes detected! 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>#!/bin/bash
terraform plan -detailed-exitcode -out=tfplan
case $? in
 0) echo "No changes needed" ;;
 1) echo "Error in plan" ; exit 1 ;;
 2) echo "Changes detected, applying..." ; terraform apply tfplan ;;
esac
</code></pre>

</div>



Industry secret: Companies using automated Terraform testing report 89% fewer production incidents related to infrastructure changes! 🛡️

<h2>
 
 
 Conclusion
</h2>

And voilà! You now have the trinity of Terraform mastery: Modularity, Reusability, and Maintainability. These aren't just fancy French words - they're your survival kit in the Infrastructure as Code jungle! 🌟

Remember, good Terraform code is like a fine wine - it gets better with time, doesn't leave you with a headache, and makes you look sophisticated at dinner parties! 🍷

The secret sauce isn't just in knowing these practices, but in applying them consistently. Start small, refactor ruthlessly, and always think about the poor soul (probably future you) who will maintain this code at 3 AM on a Sunday.

Your mission, should you choose to accept it: Pick one existing Terraform project this week and apply just one of these principles. Watch the magic happen! ✨

What's the worst Terraform horror story you've encountered? Share it in the comments - we could all use a good laugh (and learn from each other's pain)! 😄




<a href="https://buymeacoffee.com/tavernetech" rel="noopener noreferrer"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fjf9tsaj3ugonagdk5vsc.png" alt="buy me a coffee" width="545" height="153"></a>

في عالم البنية التحتية ككود، يمكن أن يكون إدارة ملفات Terraform المعقدة أمرًا شاقًا، خاصة عندما تؤدي التغييرات البسيطة إلى مشاكل غير متوقعة. يشارك هذا المقال رؤى من محترفين ذوي خبرة قاموا بالتنقل عبر هذه التحديات لسنوات. من خلال الكشف عن 'الوصفة السرية' الخاصة بهم، يهدفون إلى تمكين المطورين من تبسيط عملياتهم وتجنب الفخاخ الشائعة، مما يجعل عالم إدارة البنية التحتية أكثر كفاءة وأقل إحباطًا.

En el ámbito de la infraestructura como código, gestionar archivos de Terraform complejos puede ser desalentador, especialmente cuando cambios simples conducen a problemas inesperados. Este artículo comparte ideas de profesionales experimentados que han navegado por estos desafíos durante años. Al revelar su 'receta secreta', buscan empoderar a los desarrolladores para que optimicen sus procesos y eviten trampas comunes, haciendo que el mundo de la gestión de infraestructuras sea más eficiente y menos frustrante.

Dans le domaine de l'infrastructure en tant que code, gérer des fichiers Terraform complexes peut être décourageant, surtout lorsque des changements simples entraînent des problèmes inattendus. Cet article partage des idées de professionnels chevronnés qui ont navigué à travers ces défis pendant des années. En révélant leur 'recette secrète', ils visent à donner aux développeurs les moyens de rationaliser leurs processus et d'éviter les pièges courants, rendant ainsi le monde de la gestion des infrastructures plus efficace et moins frustrant.

In the realm of Infrastructure as Code, managing complex Terraform files can be daunting, especially when simple changes lead to unexpected issues. This article shares insights from seasoned professionals who have navigated these challenges for years. By revealing their 'secret sauce,' they aim to empower developers to streamline their processes and avoid common pitfalls, making the world of infrastructure management more efficient and less frustrating.

Effective Terraforming: Secrets of the Pros

<A HREF="https://www.cnbc.com/2025/11/05/figma-fig-q3-earnings-report-2025.html"><IMG VSPACE="4" HSPACE="4" BORDER="0" ALIGN="RIGHT" SRC="http://www.techmeme.com/251105/i54.jpg"></A>
<A HREF="http://www.techmeme.com/251105/p54#a251105p54" TITLE="Techmeme permalink"><IMG WIDTH=11 HEIGHT=12 SRC="http://www.techmeme.com/img/pml.png" STYLE="border:none;padding:0;margin:0;"></A> Jordan Novet / <A HREF="http://www.cnbc.com/">CNBC</A>: 
<A HREF="https://www.cnbc.com/2025/11/05/figma-fig-q3-earnings-report-2025.html">Figma reports Q3 revenue up 38% YoY to $274.2M, vs. $265.2M est., adjusted operating margin of 12%, vs. 6.5% est., and forecasts Q4 revenue above estimates</A>&nbsp; &mdash;&nbsp; Design software maker Figma on Wednesday reported stronger-than-expected third-quarter revenue and quarterly revenue guidance.

أبلغت شركة فيغما عن زيادة ملحوظة بنسبة 38% في الإيرادات على أساس سنوي للربع الثالث، حيث وصلت إلى 274.2 مليون دولار، متجاوزة التقديرات البالغة 265.2 مليون دولار. كما تحسنت الهامش التشغيلي المعدل للشركة بشكل كبير ليصل إلى 12%، مقارنة بـ 6.5% المتوقعة. لا تبرز هذه الأداء القوي مسار نمو فيغما فحسب، بل تحدد أيضًا نغمة إيجابية للربع القادم، حيث تتوقع الشركة أن تتجاوز إيرادات الربع الرابع التوقعات. تعتبر هذه النتائج حاسمة لأنها تعكس زيادة وجود فيغما في السوق والطلب المتزايد على أدوات التصميم الخاصة بها.

Figma ha reportado un notable aumento del 38% en los ingresos interanuales para el tercer trimestre, alcanzando los 274,2 millones de dólares, superando las estimaciones de 265,2 millones de dólares. El margen operativo ajustado de la compañía también mejoró significativamente al 12%, en comparación con el 6,5% esperado. Este sólido desempeño no solo destaca la trayectoria de crecimiento de Figma, sino que también establece un tono positivo para el próximo trimestre, ya que la empresa pronostica que los ingresos del cuarto trimestre superarán las expectativas. Tales resultados son cruciales, ya que reflejan la creciente presencia de Figma en el mercado y la creciente demanda de sus herramientas de diseño.

Figma a annoncé une augmentation remarquable de 38 % de son chiffre d'affaires d'une année sur l'autre pour le troisième trimestre, atteignant 274,2 millions de dollars, dépassant les estimations de 265,2 millions de dollars. La marge opérationnelle ajustée de l'entreprise s'est également considérablement améliorée, atteignant 12 %, contre 6,5 % attendus. Cette performance solide souligne non seulement la trajectoire de croissance de Figma, mais donne également un ton positif pour le trimestre à venir, l'entreprise prévoyant un chiffre d'affaires du quatrième trimestre supérieur aux attentes. De tels résultats sont cruciaux car ils reflètent la présence croissante de Figma sur le marché et la demande croissante pour ses outils de conception.

Figma has reported a remarkable 38% year-over-year increase in revenue for Q3, reaching $274.2 million, surpassing estimates of $265.2 million. The company's adjusted operating margin also improved significantly to 12%, compared to the expected 6.5%. This strong performance not only highlights Figma's growth trajectory but also sets a positive tone for the upcoming quarter, as the company forecasts Q4 revenue to exceed expectations. Such results are crucial as they reflect Figma's increasing market presence and the growing demand for its design tools.

Figma reports Q3 revenue up 38% YoY to $274.2M, vs. $265.2M est., adjusted operating margin of 12%, vs. 6.5% est., and forecasts Q4 revenue above estimates (Jordan Novet/CNBC)

<h1>
 
 
 Laravel Log Cleaner v2.0 - Memory-Efficient Log Management
</h1>

<h2>
 
 
 The Problem 🔥
</h2>

We've all been there: your Laravel app runs fine for weeks, then suddenly your server runs out of disk space. You SSH in, check the logs folder, and find a 5GB <code>laravel.log</code> file.

Sound familiar?

Log files can grow out of control fast, especially in production. But clearing them manually is risky, and heavy monitoring tools like Telescope can slow down your app or even crash your server.

<h2>
 
 
 The Solution ✨
</h2>

I built <a href="https://github.com/jiordiviera/laravel-log-cleaner" rel="noopener noreferrer">Laravel Log Cleaner</a> to solve this exact problem. Version 2.0 just launched with major improvements:

<ul>
<li>🧠 Memory-efficient processing - handles multi-GB files without crashing</li>
<li>📦 Compression support - archive old logs instead of deleting</li>
<li>🔒 Backup creation - never lose important data</li>
<li>🎯 Log level filtering - keep only ERROR logs, discard the rest</li>
<li>👀 Dry-run mode - preview changes before applying</li>
<li>⚡ 50%+ performance improvement on large files</li>
</ul>

<h2>
 
 
 Installation
</h2>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>composer require jiordiviera/laravel-log-cleaner
</code></pre>

</div>



That's it! No config files, no service providers to register. Laravel's auto-discovery handles everything.

<h2>
 
 
 Basic Usage
</h2>

Clear all logs: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear
</code></pre>

</div>



Keep the last 30 days: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30
</code></pre>

</div>



<h2>
 
 
 Advanced Features (What's New in v2.0)
</h2>

<h3>
 
 
 1. Safe Operations 🔒
</h3>

Preview what will be deleted without actually deleting: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --dry-run
</code></pre>

</div>



Output: 


<div class="highlight js-code-highlight">
<pre class="highlight plaintext"><code>[DRY RUN] Would remove 15,420 lines from laravel.log
[DRY RUN] Estimated space to free: 45.2 MB
</code></pre>

</div>



Create a backup before cleaning: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --backup
</code></pre>

</div>



Output: 


<div class="highlight js-code-highlight">
<pre class="highlight plaintext"><code>Backup created: storage/logs/laravel.log.backup.2024-11-05-14-30-15
Logs older than 30 days have been removed.
</code></pre>

</div>



<h3>
 
 
 2. Compression Instead of Deletion 📦
</h3>

Archive old logs for compliance or audit purposes: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --compress
</code></pre>

</div>



This creates <code>laravel.log.old.2024-11-05.gz</code> - perfect for when you need to keep logs but save disk space.

<h3>
 
 
 3. Log Level Filtering 🎯
</h3>

Keep only critical errors, discard everything else: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=0 --level=ERROR
</code></pre>

</div>



This is huge for production. Most of your logs are probably DEBUG or INFO - useful during development but just noise after a few days in production.

Supported levels: <code>DEBUG</code>, <code>INFO</code>, <code>WARNING</code>, <code>ERROR</code>, <code>CRITICAL</code>

<h3>
 
 
 4. Memory-Efficient Processing 🧠
</h3>

The game-changer for huge log files: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --memory-efficient
</code></pre>

</div>



Before v2.0: A 2GB log file would crash PHP with "out of memory" errors 
After v2.0: Processes the same file using stream processing, never loads more than 50MB in RAM

The package automatically detects large files (&gt;50MB) and enables memory-efficient mode, but you can force it with this flag.
<h3>
 
 
 5. Custom Date Patterns 🔍
</h3>

Got non-standard log formats? No problem: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --pattern="/^(\d{4}-\d{2}-\d{2})/"
</code></pre>

</div>



<h3>
 
 
 6. Combine Everything 🚀
</h3>

The complete workflow: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=30 --backup --compress --level=ERROR --dry-run
</code></pre>

</div>



This command:

<ol>
<li>Previews what would be deleted (dry-run)</li>
<li>Creates a backup before making changes</li>
<li>Compresses old logs instead of deleting</li>
<li>Keeps only ERROR level logs</li>
</ol>

<h2>
 
 
 Real-World Use Case
</h2>

Here's how I use it in production:

<h3>
 
 
 Laravel 11+ (Current)
</h3>

Schedule in <code>routes/console.php</code>: 


<div class="highlight js-code-highlight">
<pre class="highlight php"><code>use Illuminate\Support\Facades\Schedule;

// Every day at 2 AM - keep 7 days of ERROR logs
Schedule::command('log:clear', [
 '--days' =&gt; 7,
 '--backup' =&gt; true,
 '--level' =&gt; 'ERROR'
])-&gt;daily()-&gt;at('02:00');

// Weekly deep clean with compression
Schedule::command('log:clear', [
 '--days' =&gt; 30,
 '--compress' =&gt; true
])-&gt;weekly()-&gt;sundays()-&gt;at('03:00');
</code></pre>

</div>



<h3>
 
 
 Laravel 10 and Below
</h3>

Schedule in <code>app/Console/Kernel.php</code>: 


<div class="highlight js-code-highlight">
<pre class="highlight php"><code>protected function schedule(Schedule $schedule)
{
 $schedule-&gt;command('log:clear', [
 '--days' =&gt; 7,
 '--backup' =&gt; true,
 '--level' =&gt; 'ERROR'
 ])-&gt;daily()-&gt;at('02:00');

 $schedule-&gt;command('log:clear', [
 '--days' =&gt; 30,
 '--compress' =&gt; true
 ])-&gt;weekly()-&gt;sundays()-&gt;at('03:00');
}
</code></pre>

</div>



<h3>
 
 
 Result:
</h3>

<ul>
<li>✅ Keeps 7 days of ERROR logs for quick debugging</li>
<li>✅ Archives everything older in compressed format</li>
<li>✅ Never runs out of disk space</li>
<li>✅ No manual intervention needed</li>
<li>✅ Backups created automatically</li>
</ul>

<h2>
 
 
 Performance Benchmarks 📊
</h2>

I tested this on a real production log file from one of my client's projects:

Test File: 1.2 GB, 3.5 million lines

<div class="table-wrapper-paragraph"><table>
<thead>
<tr>
<th>Operation</th>
<th>v1.0</th>
<th>v2.0</th>
<th>Improvement</th>
</tr>
</thead>
<tbody>
<tr>
<td>Clear all logs</td>
<td>45s</td>
<td>12s</td>
<td>
73% faster ⚡</td>
</tr>
<tr>
<td>Filter by date (30 days)</td>
<td>Memory error ❌</td>
<td>18s</td>
<td>
Now possible ✅</td>
</tr>
<tr>
<td>Memory usage (peak)</td>
<td>512MB+</td>
<td>48MB</td>
<td>
90% reduction 🎯</td>
</tr>
<tr>
<td>Compression</td>
<td>Not supported</td>
<td>8s</td>
<td>
New feature 🆕</td>
</tr>
</tbody>
</table></div>

<h2>
 
 
 Why Not Use X? 🤔
</h2>

<h3>
 
 
 vs. Manual deletion (<code>rm laravel.log</code> or <code>echo "" &gt; laravel.log</code>)
</h3>

<ul>
<li>❌ No selective date filtering</li>
<li>❌ Risk of deleting current day's logs</li>
<li>❌ No backup option</li>
<li>❌ Can't filter by log level</li>
</ul>

<h3>
 
 
 vs. Laravel Telescope
</h3>

<ul>
<li>❌ Telescope is heavy (database overhead, UI, background jobs)</li>
<li>❌ Can actually slow down your app in production</li>
<li>✅ Laravel Log Cleaner has zero runtime overhead
</li>
<li>✅ Only runs when you schedule it</li>
</ul>

<h3>
 
 
 vs. Other log cleaner packages
</h3>

Most alternatives:

<ul>
<li>❌ Don't handle memory efficiently (crash on large files)</li>
<li>❌ No compression support</li>
<li>❌ No log level filtering</li>
<li>❌ No dry-run mode for safety</li>
<li>❌ Limited or no backup functionality</li>
</ul>

<h2>
 
 
 Compatibility
</h2>

Current Version (v2.x):

<ul>
<li>PHP 8.1, 8.2, 8.3+</li>
<li>Laravel 9.x, 10.x, 11.x, 12.x</li>
</ul>

Legacy Support (v1.x): 
If you're still on older versions: 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>composer require jiordiviera/laravel-log-cleaner:^1.0
</code></pre>

</div>



<ul>
<li>PHP 7.0+</li>
<li>Laravel 7.x, 8.x</li>
</ul>

<h2>
 
 
 Migration from v1.x
</h2>

Upgrading from v1.x? It's seamless!

Step 1: Update via Composer 


<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>composer update jiordiviera/laravel-log-cleaner
</code></pre>

</div>



Step 2: Check PHP/Laravel requirements

<ul>
<li>Requires PHP 8.1+</li>
<li>Requires Laravel 9+</li>
</ul>

Step 3: Enjoy new features! 
All your existing commands still work. The new features are opt-in via command flags.

Breaking changes:

<ul>
<li>Dropped PHP 7.x support</li>
<li>Dropped Laravel 7.x and 8.x support</li>
<li>If you need these versions, stay on v1.x</li>
</ul>

<h2>
 
 
 Common Use Cases
</h2>

<h3>
 
 
 1. Development: Keep Recent Logs Only
</h3>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=3
</code></pre>

</div>



<h3>
 
 
 2. Production: Aggressive Cleanup with Safety
</h3>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=7 --backup --level=ERROR
</code></pre>

</div>



<h3>
 
 
 3. Compliance: Archive Everything
</h3>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=90 --compress
</code></pre>

</div>



<h3>
 
 
 4. Emergency: Server Running Out of Space
</h3>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code># Preview impact first
php artisan log:clear --days=1 --dry-run

# If safe, execute
php artisan log:clear --days=1 --backup
</code></pre>

</div>



<h3>
 
 
 5. CI/CD Pipeline: Clean After Tests
</h3>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>php artisan log:clear --days=0
</code></pre>

</div>



<h2>
 
 
 What's Next? 🔮
</h2>

I'm considering these features for future releases:

<ul>
<li>🗂️ Multiple log file support (<code>--file=api.log</code>)</li>
<li>📧 Slack/Email notifications on cleanup completion</li>
<li>⚙️ Config file for environment-specific retention policies</li>
<li>☁️ Cloud storage integration (S3, GCS, Azure)</li>
<li>📊 Log analytics before cleanup (show stats)</li>
<li>🔄 Auto-rotation based on file size (not just days)</li>
</ul>

What would YOU like to see? Drop a comment below! Your feedback shapes the roadmap.

<h2>
 
 
 Try It Out
</h2>



<div class="highlight js-code-highlight">
<pre class="highlight shell"><code>composer require jiordiviera/laravel-log-cleaner
php artisan log:clear --help
</code></pre>

</div>



Links:

<ul>
<li>📦 <a href="https://packagist.org/packages/jiordiviera/laravel-log-cleaner" rel="noopener noreferrer">Packagist</a>
</li>
<li>🐙 <a href="https://github.com/jiordiviera/laravel-log-cleaner" rel="noopener noreferrer">GitHub Repository</a>
</li>
<li>⭐ Star on GitHub if you find it useful!</li>
</ul>

<h2>
 
 
 Contributing
</h2>

Found a bug? Have a feature request? Contributions are welcome!

<ol>
<li>Fork the repo</li>
<li>Create a feature branch</li>
<li>Write tests (we use Pest)</li>
<li>Submit a PR</li>
</ol>

Check out the <a href="https://github.com/jiordiviera/laravel-log-cleaner/blob/main/CONTRIBUTING.md" rel="noopener noreferrer">Contributing Guide</a>.




<h2>
 
 
 About Me
</h2>

I'm Jiordi Viera, a fullstack developer from Douala, Cameroon 🇨🇲. I work primarily with Laravel, Next.js, and React. 

I built this package after dealing with production log issues one too many times. What started as a simple script evolved into a full package when I realized others had the same problem.

If this package helps you, consider:

<ul>
<li>⭐ Starring the <a href="https://github.com/jiordiviera/laravel-log-cleaner" rel="noopener noreferrer">GitHub repo</a>
</li>
<li>📢 Sharing this article</li>
<li>☕ Buying me a coffee (if you're feeling generous)</li>
</ul>

Connect with me:

<ul>
<li>🌐 Website: <a href="https://jiordiviera.me" rel="noopener noreferrer">jiordiviera.me</a>
</li>
<li>🐙 GitHub: <a href="https://github.com/jiordiviera" rel="noopener noreferrer">@jiordiviera</a>
</li>
<li>💼 LinkedIn: [Add your LinkedIn if you have one]</li>
<li>🐦 Twitter/X: [Add your handle if you have one]</li>
</ul>

Other projects:

<ul>
<li>
<a href="https://codepit.jiordiviera.me" rel="noopener noreferrer">Codepit</a> - A minimal platform to share code snippets</li>
</ul>




Thanks for reading! If you found this helpful, please leave a ❤️ and share your experience in the comments below.

Happy logging! 🚀

<h1>
 
 
 Laravel #PHP #OpenSource #DevOps #WebDev #BackendDevelopment
</h1>

تأتي Laravel Log Cleaner v2.0 لتحدث ثورة في كيفية إدارة المطورين لملفات السجل، حيث تعالج المشكلة الشائعة لاستخدام مساحة القرص بشكل مفرط. تقدم هذه النسخة الجديدة إدارة فعالة للذاكرة مع ميزات مثل الضغط والنسخ الاحتياطي، مما يسهل على المطورين الحفاظ على تطبيقاتهم دون الخوف من نفاد المساحة. إنها نقطة تحول لأي شخص يستخدم Laravel، مما يضمن عمليات أكثر سلاسة وأقل متاعب عند إدارة ملفات السجل.

Laravel Log Cleaner v2.0 llega para revolucionar la gestión de archivos de registro de los desarrolladores, abordando el problema común del uso excesivo del espacio en disco. Esta nueva versión presenta una gestión de registros eficiente en memoria con características como compresión y respaldo, facilitando a los desarrolladores el mantenimiento de sus aplicaciones sin el temor de quedarse sin espacio. Es un cambio radical para cualquier persona que use Laravel, asegurando operaciones más fluidas y menos complicaciones en la gestión de archivos de registro.

Laravel Log Cleaner v2.0 arrive pour révolutionner la gestion des fichiers journaux des développeurs, en répondant au problème courant de l'utilisation excessive de l'espace disque. Cette nouvelle version introduit une gestion des journaux efficace en mémoire avec des fonctionnalités telles que la compression et la sauvegarde, facilitant ainsi la maintenance des applications sans craindre de manquer d'espace. C'est un changement de jeu pour quiconque utilise Laravel, garantissant des opérations plus fluides et moins de tracas en matière de gestion des fichiers journaux.

Laravel Log Cleaner v2.0 is here to revolutionize how developers manage log files, addressing the common issue of excessive disk space usage. This new version introduces memory-efficient log management with features like compression and backup, making it easier for developers to maintain their applications without the fear of running out of space. It's a game-changer for anyone using Laravel, ensuring smoother operations and less hassle when it comes to log file management.

Laravel Log Cleaner v2.0 - Memory-Efficient Log Management with Compression & Backup

Most AI deployments wait for a prompt. Real operations do not. A manager defines a goal, delegates work, and people go collect facts, make decisions, and return results. What we need is AI agents that have an intent and initiate conversations with a goal.

The system below mirrors that familiar pattern with task templates, action chains, a watchdog that moves work forward, and scout agents that initiate conversations. It is simple to describe, strong in practice, and it scales. Everything described here is doable with existing technologies.

Let's look at the main components that make up such a system:

<h1>
 
 
 System Components
</h1>

<h2>
 
 
 1) Task template
</h2>

A task template is the starting point. It contains an action chain plus basic metadata like name and description. Task Templates are executed by an event, on a schedule, when criteria are met, or on demand by a human. 
When a template is executed, it becomes a task instance with its own state and audit trail.

<h2>
 
 
 2) Action chain
</h2>

An action chain is a list of actions that run in order inside a task. Each action has an embedded AI agent and the watchdog oversees execution. Actions come in a few types:

<ul>
<li>Scout Agents. Initiates a conversation with one or more counterparties and collects structured data. Scouts are detached by design. People can ignore a message or take time to answer, so the system must tolerate latency and silence. Scout agents are implemented by temporarily replacing the normal support agent in the support chat with it self and initiating a conversation with a pre-defined templatable greeting.</li>
<li>Create PDF. Uses a Word template and a simple JSON [key name]/[description] schema to generate a finished document as an artifact.</li>
<li>Compose and send email. Writes an email from instructions and can attach artifacts created earlier in the chain.</li>
<li>Perform other actions. Talk to an API, make a change in the database, post to a queue, or anything else needed. The sky is the limit here.</li>
</ul>

Every action in the task template and subsequently in the executed task instance includes:

<ul>
<li>Instructions. Plain language guidance that becomes part of the system prompt. For example, write an email about subject X or collect data about subject Y.</li>
<li>Input. The output of the previous action or fields fetched from the task data store.</li>
<li>Output. A structured payload for the next action and for the task data store.</li>
<li>Timeout policy. What to do if the action does not complete in time. Abort the task, alert a human, or ignore and continue.</li>
</ul>

The task has a general data store. Any action can write to it. Subsequent actions can read from it. This keeps the chain loosely coupled and easy to extend. LLMs are fantastically adopted to reading loosely structured data and understanding it’s context.

<h2>
 
 
 3) Watchdog
</h2>

The watchdog advances the chain. It starts actions, monitors completion, enforces timeouts, and dispatches the next step. If an action stalls, the watchdog retries, switches channel, escalates to a human, or bails out per policy. It also keeps the log, timestamps each event, and stores artifacts. Think of it as the manager who checks progress and pushes the work forward.

<h2>
 
 
 4) Document templates
</h2>

A document template is a Word file with {{tags}} and a schema for expected data. Templates can attach to a PDF action to produce a signed or branded document. They can also attach to a scout action so the scout knows exactly which fields to collect. The schema is the contract. It gives the system a way to validate outputs before side effects happen.

<h1>
 
 
 Three simple use cases
</h1>

<h2>
 
 
 1) Collect pizza orders from the whole office
</h2>

Let's start off with something simple to illustrate the functionality of each component in the chain.

<ol>
<li>The task dispatches scouts to everyone on the team.</li>
<li>Each scout opens a chat, gathers order, size, and special notes, then writes the result into the task data store.</li>
<li>When all scouts complete or hit a timeout, the next action composes a single email to the pizzeria and sends it.
Instructions for the final action that composes an email, has access to the task instance data store and an agent doing the job in the task template can be as simple as: </li>
</ol>

<code>Write a pizza order to some,local@pizza.joint using the available requests in the data store, deliver the order to our office at Some Street 123 and invoice Acme Inc account number #123 or bring a card terminal. 
</code> 
The role of the timeout and bail out functions here would be that if a scout fails to collect the pizza wishes from a team member, ignore it and send the order out at 11:30 anyway.

<h2>
 
 
 2) Help a customer file an insurance claim
</h2>

Next let's examine a different flow and another way to trigger an execution of a task instance.

<ol>
<li>In support chat, the customer tells the default agent that they need help filing an insurance claim.</li>
<li>The default agent selects the “insurance claim” task template and it is executed.</li>
<li>The template includes a Word document and a schema that lists incident fields.</li>
<li>A scout opens a conversation with the customer, collects the fields, and writes the JSON to the task data store and outputs it as the input to the next action.</li>
<li>A PDF action fills the Word template and outputs a final document.</li>
<li>An email action writes a short cover note and sends the PDF to the insurer mailbox.</li>
</ol>

Task templates automatically becoming tools that your customer default support AI agent can dispatch is the revolutionary part here. 

How many different jobs performed by some human in the office are essentially: Go talk to someone about something, write a document about it and send it to someone? Think about it, quite a lot. 
Now just by defining the necessary task templates in an admin interface for each of these and selecting them to become available as tools lets you automate a lot of processes.

<h2>
 
 
 3) Update all customer phone numbers
</h2>

<ol>
<li>The task dispatches scouts to every customer.</li>
<li>Each scout asks for the current phone number and returns a structured payload.</li>
<li>A database action validates the format and updates the record. Errors are logged for human review.</li>
</ol>

This example is powerful because the scout agents can all run in parallel. Updating the customer database can be done in 30 minutes for all customers compared to humans doing the work in a series. 
Timeout and bail out parameters here can be set to like a day and alerts generated for scouts that were not able to complete their tasks.

<h1>
 
 
 Why full platform control matters
</h1>

Scout agents only work when you control the full system where the chain runs. You need to open and manage chat threads with a separate chat agent handling them, attach forms and files, tag each message with a correlation id, and capture consent. You need to validate outputs against a schema before taking any action. You need clean audit logs. Stitching this across many unrelated tools is brittle.

This is why platforms like Salesforce, Odoo or others are natural homes for detached chains. Both combine workflow, data, and communications. You can create records, run server actions, post to chat, and store artifacts in one place. You can also add human review steps that see the full context. If your organization cannot use these, you can build the ecosystem in house. The key is ownership of channels and data paths.

Chat control is critical. A scout that cannot open a thread, send a reminder, switch channel, or record consent will fail in real life. The watchdog needs reliable signals from the chat layer to decide whether to retry, escalate, or move on.

<h1>
 
 
 How this mirrors human work
</h1>

This model behaves like a team. A manager defines a mission. Scouts go talk to people, collect facts, and return structured notes. Specialists draft a document or update a system. The watchdog is like a manager. It keeps time, nudges when needed, and escalates when something blocks. The result is predictable progress from intent to artifact without long back and forth.

The difference is that the system never forgets a field, always writes in the right format, and enforces time limits. It also produces an audit trail that shows who said what and when, what data was used, and which decision happened next. And all this can run in parallel and scale on demand.

<h1>
 
 
 Implementation checklist
</h1>

<ol>
<li>Pick a platform where you can control chat, data, and actions in one place.</li>
<li>Define a task template with a short action chain that solves one narrow problem.</li>
<li>Write the scout instructions and the output schema. Keep it concrete.</li>
<li>Add a Word template with {{tags}} for any document you need to produce.</li>
<li>Implement timeout policies. Soft timeout for a nudge. Hard timeout for failover.</li>
<li>Validate every output against the schema before any side effect.</li>
<li>Log transcripts, artifacts, and decisions for audit and training.</li>
<li>Track a few simple metrics. Completion time, first try rate, timeout rate, and cost per completed task.</li>
</ol>

<h1>
 
 
 The power factor is the scout
</h1>

The scout is what turns a static workflow into a living one. It is the agent that can initiate conversations and create new information by talking to people. Everything else is plumbing. Once you can send respectful, purpose built scouts that collect exactly the fields you need, the rest of the chain becomes routine. The watchdog moves the work along. The PDF has the right data. The database stays clean. Customers get what they need without long calls.

That is how detached agent chains become practical. They behave like people, run inside systems you control, and leave artifacts other systems can trust.

Author bio: Antti Kaipila is the founder of <a href="https://nuuduu.com" rel="noopener noreferrer">Nuuduu</a>, a direct service platform that uses AI to automate service delivery.

يظهر نهج جديد للذكاء الاصطناعي يحاكي السلوك البشري من خلال السماح للوكلاء ببدء المحادثات والعمل نحو أهداف محددة دون انتظار التعليمات. يستخدم هذا النظام المبتكر قوالب المهام وسلاسل الإجراءات لتبسيط العمليات، مما يجعله سهل الفهم وفعالًا في الممارسة العملية. من خلال تمكين الذكاء الاصطناعي من اتخاذ المبادرة، يمكن للشركات تعزيز الإنتاجية واتخاذ القرارات، وهو أمر بالغ الأهمية في بيئة اليوم السريعة.

Está surgiendo un nuevo enfoque de la IA que imita el comportamiento humano al permitir que los agentes inicien conversaciones y trabajen hacia objetivos definidos sin esperar a que se les indique. Este sistema innovador utiliza plantillas de tareas y cadenas de acciones para optimizar las operaciones, lo que lo hace no solo fácil de entender, sino también efectivo en la práctica. Al permitir que la IA tome la iniciativa, las empresas pueden mejorar la productividad y la toma de decisiones, lo cual es crucial en el acelerado entorno actual.

Une nouvelle approche de l'IA émerge, imitant le comportement humain en permettant aux agents d'initier des conversations et de travailler vers des objectifs définis sans attendre de prompts. Ce système innovant utilise des modèles de tâches et des chaînes d'actions pour rationaliser les opérations, le rendant à la fois simple à comprendre et efficace en pratique. En permettant à l'IA de prendre l'initiative, les entreprises peuvent améliorer la productivité et la prise de décision, ce qui est crucial dans l'environnement rapide d'aujourd'hui.

A new approach to AI is emerging that mimics human behavior by allowing agents to initiate conversations and work towards defined goals without waiting for prompts. This innovative system uses task templates and action chains to streamline operations, making it not only simple to understand but also effective in practice. By enabling AI to take the initiative, businesses can enhance productivity and decision-making, which is crucial in today's fast-paced environment.

Detached agent chains that work like people do

Andrew Cuomo's campaign embraced AI — and it made him a toxic laughingstock, while Zohran Mamdani cruised to victory.
The post <a href="https://futurism.com/artificial-intelligence/andrew-cuomo-ai-toxic">Andrew Cuomo&#8217;s Extensive Use of AI Made His Campaign a Toxic Joke</a> appeared first on <a href="https://futurism.com">Futurism</a>.

اعتمدت حملة أندرو كومو بشكل كبير على تقنية الذكاء الاصطناعي، لكن بدلاً من تعزيز فرصه، تحولت إلى مصدر للسخرية، مما جعله موضع سخرية في الساحة السياسية. في هذه الأثناء، حقق زوهرا مامداني النصر، مما يبرز الفارق الواضح بين نهجيهما. تسلط هذه الحالة الضوء على المخاطر المحتملة للاعتماد المفرط على التكنولوجيا في الحملات السياسية، مما يذكر المرشحين بأن الأصالة والاتصال بالناخبين أمران حاسمان لتحقيق النجاح.

La campaña de Andrew Cuomo se basó en gran medida en la tecnología de IA, pero en lugar de mejorar sus posibilidades, se convirtió en objeto de burla, convirtiéndolo en un hazmerreír en la arena política. Mientras tanto, Zohran Mamdani emergió victorioso, destacando el marcado contraste entre sus enfoques. Esta situación subraya las posibles trampas de depender en exceso de la tecnología en las campañas políticas, recordando a los candidatos que la autenticidad y la conexión con los votantes son cruciales para el éxito.

La campagne d'Andrew Cuomo a fortement misé sur la technologie de l'IA, mais au lieu d'améliorer ses chances, cela a conduit à des moqueries, faisant de lui un objet de ridicule sur la scène politique. Pendant ce temps, Zohran Mamdani a remporté la victoire, soulignant le contraste entre leurs approches. Cette situation met en lumière les pièges potentiels d'une dépendance excessive à la technologie dans les campagnes politiques, rappelant aux candidats que l'authenticité et la connexion avec les électeurs sont essentielles pour réussir.

Andrew Cuomo's campaign heavily relied on AI technology, but instead of boosting his chances, it turned into a source of ridicule, making him a laughingstock in the political arena. Meanwhile, Zohran Mamdani emerged victorious, highlighting the stark contrast between their approaches. This situation underscores the potential pitfalls of over-relying on technology in political campaigns, reminding candidates that authenticity and connection with voters are crucial for success.

Andrew Cuomo’s Extensive Use of AI Made His Campaign a Toxic Joke

arXiv:2511.01610v1 Announce Type: new 
Abstract: Vision Foundation Models (VFMs) have advanced representation learning through self-supervised methods. However, existing training pipelines are often inflexible, domain-specific, or computationally expensive, which limits their usability across different domains and resource settings. DINO-MX is a modular and extensible training framework that combines the core principles of DINO, DINOv2 and DINOv3 within a unified configuration-driven system. It supports a variety of transformer-based architectures and is fully compatible with the Hugging Face ecosystem. The framework includes multiple training strategies such as low-rank adaptation (LoRA), layer freezing, and knowledge distillation, along with support for distributed training through both Distributed Data Parallel (DDP) and Fully Sharded Data Parallel (FSDP). DINO-MX is designed to work with both natural and specialized data types, including single- and multi-channel images. Experimental results on diverse datasets show that DINO-MX achieves competitive performance while significantly reducing computational costs. Additionally, it offers interpretability tools and a label-guided data augmentation method that improves attention-based localization without the need for extra detection or segmentation heads. DINO-MX provides a reproducible and scalable foundation for developing, adapting, and benchmarking self-supervised vision models across a range of research and real-world applications.

DINO-MX هو إطار تدريب مبتكر يعزز التعلم الذاتي المشرف من خلال دمج أفضل ميزات النماذج السابقة مثل DINO وDINOv2 وDINOv3. هذا النظام القابل للتعديل يتناول قيود خطوط التدريب الحالية، مما يجعله أكثر قابلية للتكيف وكفاءة عبر مجالات مختلفة. تكمن أهميته في قدرته على ديمقراطية التعلم المتقدم للتمثيل، مما يسمح للباحثين والمطورين بالاستفادة من أدوات قوية دون قيود التكاليف الحاسوبية العالية أو التخصص في المجال.

DINO-MX es un marco de entrenamiento innovador que mejora el aprendizaje auto-supervisado al integrar las mejores características de modelos anteriores como DINO, DINOv2 y DINOv3. Este sistema modular aborda las limitaciones de los pipelines de entrenamiento existentes, haciéndolo más adaptable y eficiente en varios dominios. Su importancia radica en su potencial para democratizar el aprendizaje avanzado de representaciones, permitiendo a investigadores y desarrolladores aprovechar herramientas poderosas sin las restricciones de altos costos computacionales o especificidad de dominio.

DINO-MX est un cadre de formation innovant qui améliore l'apprentissage auto-supervisé en intégrant les meilleures caractéristiques des modèles précédents tels que DINO, DINOv2 et DINOv3. Ce système modulaire répond aux limitations des pipelines de formation existants, le rendant plus adaptable et efficace dans divers domaines. Son importance réside dans son potentiel à démocratiser l'apprentissage avancé de la représentation, permettant aux chercheurs et aux développeurs de tirer parti d'outils puissants sans les contraintes de coûts computationnels élevés ou de spécificité de domaine.

DINO-MX is an innovative training framework that enhances self-supervised learning by integrating the best features of previous models like DINO, DINOv2, and DINOv3. This modular system addresses the limitations of existing training pipelines, making it more adaptable and efficient across various domains. Its significance lies in its potential to democratize advanced representation learning, allowing researchers and developers to leverage powerful tools without the constraints of high computational costs or domain specificity.

DINO-MX: A Modular & Flexible Framework for Self-Supervised Learning

arXiv:2511.02712v1 Announce Type: new 
Abstract: Understanding and predicting emotion from videos has gathered significant attention in recent studies, driven by advancements in video large language models (VideoLLMs). While advanced methods have made progress in video emotion analysis, the intrinsic nature of emotions poses significant challenges. Emotions are characterized by dynamic and cues-dependent properties, making it difficult to understand complex and evolving emotional states with reasonable rationale. To tackle these challenges, we propose a novel affective cues-guided reasoning framework that unifies fundamental attribute perception, expression analysis, and high-level emotional understanding in a stage-wise manner. At the core of our approach is a family of video emotion foundation models (VidEmo), specifically designed for emotion reasoning and instruction-following. These models undergo a two-stage tuning process: first, curriculum emotion learning for injecting emotion knowledge, followed by affective-tree reinforcement learning for emotion reasoning. Moreover, we establish a foundational data infrastructure and introduce a emotion-centric fine-grained dataset (Emo-CFG) consisting of 2.1M diverse instruction-based samples. Emo-CFG includes explainable emotional question-answering, fine-grained captions, and associated rationales, providing essential resources for advancing emotion understanding tasks. Experimental results demonstrate that our approach achieves competitive performance, setting a new milestone across 15 face perception tasks.

تقدم VidEmo نهجًا جديدًا لفهم المشاعر في مقاطع الفيديو، مستفيدة من التقدم في نماذج اللغة الكبيرة للفيديو. تهدف هذه الطريقة المبتكرة إلى معالجة تعقيدات تحليل المشاعر، مع الأخذ في الاعتبار الطبيعة الديناميكية للمشاعر واعتمادها على إشارات متنوعة.

VidEmo presenta un nuevo enfoque para comprender las emociones en videos, aprovechando los avances en modelos de lenguaje de video. Este método innovador busca abordar las complejidades del análisis emocional, considerando la naturaleza dinámica de las emociones y su dependencia de diversas señales.

VidEmo présente une nouvelle approche pour comprendre les émotions dans les vidéos, en s'appuyant sur les avancées des modèles de langage vidéo. Cette méthode innovante vise à s'attaquer aux complexités de l'analyse émotionnelle, en tenant compte de la nature dynamique des émotions et de leur dépendance à divers indices.

VidEmo introduces a new approach to understanding emotions in videos, leveraging advancements in video large language models. This innovative method aims to tackle the complexities of emotional analysis, addressing the dynamic nature of emotions and their dependence on various cues.

VidEmo: Affective-Tree Reasoning for Emotion-Centric Video Foundation Models

arXiv:2511.01914v1 Announce Type: new 
Abstract: We introduce iFlyBot-VLA, a large-scale Vision-Language-Action (VLA) model trained under a novel framework. The main contributions are listed as follows: (1) a latent action model thoroughly trained on large-scale human and robotic manipulation videos; (2) a dual-level action representation framework that jointly supervises both the Vision-Language Model (VLM) and the action expert during training; (3) a mixed training strategy that combines robot trajectory data with general QA and spatial QA datasets, effectively enhancing the 3D perceptual and reasoning capabilities of the VLM backbone. Specifically, the VLM is trained to predict two complementary forms of actions: latent actions, derived from our latent action model pretrained on cross-embodiment manipulation data, which capture implicit high-level intentions; and structured discrete action tokens, obtained through frequency-domain transformations of continuous control signals, which encode explicit low-level dynamics. This dual supervision aligns the representation spaces of language, vision, and action, enabling the VLM to directly contribute to action generation. Experimental results on the LIBERO Franka benchmark demonstrate the superiority of our frame-work, while real-world evaluations further show that iFlyBot-VLA achieves competitive success rates across diverse and challenging manipulation tasks. Furthermore, we plan to open-source a portion of our self-constructed dataset to support future research in the community

iFlyBot-VLA هو نموذج مبتكر للرؤية واللغة والعمل يعزز من قدرة الروبوتات على المناورة من خلال إطار تدريب فريد. يتميز بنموذج تمثيل العمل على مستويين واستراتيجية تدريب مختلطة، مما يجعله تقدمًا كبيرًا في هذا المجال.

El iFlyBot-VLA es un modelo innovador de Visión-Lenguaje-Acción que mejora la manipulación robótica a través de un marco de entrenamiento único. Presenta una representación de acción de dos niveles y una estrategia de entrenamiento mixto, lo que lo convierte en un avance significativo en el campo.

L'iFlyBot-VLA est un modèle innovant de Vision-Langage-Action qui améliore la manipulation robotique grâce à un cadre de formation unique. Il présente une représentation d'action à deux niveaux et une stratégie de formation mixte, ce qui en fait une avancée significative dans le domaine.

The iFlyBot-VLA is an innovative Vision-Language-Action model that enhances robotic manipulation through a unique training framework. It features a dual-level action representation and a mixed training strategy, making it a significant advancement in the field.

iFlyBot-VLA Technical Report

arXiv:2407.07557v3 Announce Type: replace-cross 
Abstract: Federated learning is a renowned technique for utilizing decentralized data while preserving privacy. However, real-world applications often face challenges like partially labeled datasets, where only a few locations have certain expert annotations, leaving large portions of unlabeled data unused. Leveraging these could enhance transformer architectures ability in regimes with small and diversely annotated sets. We conduct the largest federated cardiac CT analysis to date (n=8,104) in a real-world setting across eight hospitals. Our two-step semi-supervised strategy distills knowledge from task-specific CNNs into a transformer. First, CNNs predict on unlabeled data per label type and then the transformer learns from these predictions with label-specific heads. This improves predictive accuracy and enables simultaneous learning of all partial labels across the federation, and outperforms UNet-based models in generalizability on downstream tasks. Code and model weights are made openly available for leveraging future cardiac CT analysis.

تستكشف دراسة حديثة استخدام التعلم الفيدرالي في تصوير الأشعة المقطعية القلبية، حيث تتناول التحديات المتعلقة بمجموعات البيانات الموصوفة جزئيًا. من خلال الاستفادة من البيانات اللامركزية مع الحفاظ على الخصوصية، تهدف الأبحاث إلى تحسين هياكل المحولات، مما يجعلها أكثر فعالية في السيناريوهات التي تحتوي على تعليقات خبراء محدودة.

Un estudio reciente explora el uso del aprendizaje federado en la imagenología de TC cardíaca, abordando los desafíos de los conjuntos de datos parcialmente etiquetados. Al aprovechar los datos descentralizados y mantener la privacidad, la investigación busca mejorar las arquitecturas de transformadores, haciéndolas más efectivas en escenarios con anotaciones de expertos limitadas.

Une étude récente explore l'utilisation de l'apprentissage fédéré dans l'imagerie CT cardiaque, abordant les défis liés aux ensembles de données partiellement étiquetés. En tirant parti des données décentralisées tout en préservant la vie privée, la recherche vise à améliorer les architectures de transformateurs, les rendant plus efficaces dans des scénarios avec des annotations d'experts limitées.

A recent study explores the use of federated learning in cardiac CT imaging, addressing challenges with partially labeled datasets. By leveraging decentralized data while maintaining privacy, the research aims to enhance transformer architectures, making them more effective in scenarios with limited expert annotations.

DINO-YOLO: Self-Supervised Pre-training for Data-Efficient Object Detection in Civil Engineering Applications

DINO-YOLO: Self-Supervised Pre-training for Data-Efficient Object Detection in Civil Engineering Applications

Was this article worth reading? Share it