know I just posted one 😅 but maintaining the timeline!
Another paper accepted at IEEE COMPUTINGCON 2026 🎉
Presenting on Sep 7–8, likely online due to work constraints.
3 in a row this year. More to go! 🚀📚
#Research#IEEE#AI
A little late, but worth sharing! 🚀
Presented my paper “A Multi-Agent Graph RAG Framework with React and A2A Protocol” at IEEE INSTcon 2026, NIT Rourkela.
Exploring Multi-Agent Systems, Graph RAG & A2A communication. 🤖
#AI#GraphRAG#AgenticAI#IEEE
Reading this fascinating 2024 paper on LLMs and the human brain 🧠🤖
The interesting part: better-performing LLMs show more brain-like language processing, reach similar neural representations in fewer layers, and seem to benefit strongly from context.
@Steph_Bickel
Just went through “Dense Passage Retrieval for Open-Domain Question Answering” — loved how a simple dual-encoder setup, trained with in-batch negatives, can beat traditional BM25 so convincingly. A great paper on making dense retrieval actually work in practice.
Ever wondered how apps like Instagram load profile pictures so fast for millions of users?
In this video, I explain the complete pipeline:
📤 S3 Uploads
⚙️ Image Processing
🌍 CloudFront CDN
🏷️ URL Versioning
✨ BlurHash
System Design made simple.
#AWS#CloudFront#S3#System
Most people think HuggingGPT is another AI model.
It isn't.
It's an AI manager that plans tasks, picks the best expert models from Hugging Face, combines their outputs, and returns one final answer.
#AI#HuggingGPT#LLM#AIAgents
@ClementDelangue As a student researcher, I couldn't agree more. Open datasets and open-weight models have shaped much of my learning. Open ecosystems don't just build better AI—they build better researchers.
@ylecun@ClementDelangue Interesting perspective. The recent success of open-weight models suggests that openness is still one of the strongest catalysts for research progress, even as deployment becomes more controlled.
Most people think 1 token = 1 word.
In this video I explain:
• What AI tokens actually are
• Tokenization
• Token IDs
• Image & Audio Tokens
• Input vs Output Tokens
• Context Window
• API Pricing
If you use ChatGPT or build AI apps, you'll finally understand how LLMs work
What happens when AI agents face hundreds of tasks at once?
Today's agents struggle with concurrency—maintaining context, prioritizing, and preserving state under pressure.
Next-gen systems need better coordination, memory, and real-time decision making.
@hwchase17@swyx#ai
What if AI agents didn't have to re-reason through every routine task?
A procedural memory layer could transform past reasoning into reusable execution patterns—making agents faster, more efficient, and less computationally wasteful.
@fchollet@hwchase17 Thoughts 👀
#ai#agent
What if the next in AI isn’t bigger models, but smarter attention?
The prefrontal cortex filters information before it overloads the brain. Maybe LLMs need the same biologically inspired resource management instead of endless scaling.
@GaryMarcus@bengoertzel Thoughts? 👀
#ai
GloVe proves meaning is in the ratio: Ice/Steam co-occurrence ratios perfectly capture thermodynamics phases.
Is language just a statistical matrix? #nlp#Glove#Ai#stats
📄 "Distributed Representations of Words and Phrases and their Compositionality" (2013) extended Word2Vec with phrase learning. 🚀
But if it still couldn't truly represent entire sentences or long documents, did Transformers finally solve it—or are there still gaps? 🤔
Can we encode thoughts as vectors & math?
Word2Vec (Mikolov et al. 2013) made king - man + woman ≈ queen real by turning words into distributed vectors.Simple log-linear model, Huffman softmax, CBOW & Skip-gram.Can we do the same for full thoughts? Or already?#Word2Vec