Md Ismail Šojal 🕷️

Stop reading attention is all you need. Transformers (LLMs) clearly explained with visuals: - Real GPT-2 visualizer - tokens to embeddings - 12 attention heads (Q, K, V, scale, mask, softmax) - multi-head Q/K/V - residual + MLP - next-token probabilities That’s the difference between reading about transformers and seeing one think. - poloclub.github.io/transform…

This interactive web tool for teaching computer vision basics, Users can upload an image or video stream, view it as a numerical matrix, and apply convolution kernels (edge detection, Gaussian blur, sharpening) in real time. No slides, Just the raw data moving. The goal is simple: make image processing visible before jump into OpenCV or neural networks. - github.com/computer-vision-w…

The Pirate bay Face for open LLMs, model weights, and uncensored models, Peer-to-peer distribution. Torrents but for open models. 669k+ eligible models so the weights can stay public, - pirateface.co

The cleanest, most satisfying visual explanation of convolution I've ever seen.✨ - Kernel flip - edge detection - Gaussian blur - complex plane glow-up. - Math = dangerous levels of satisfaction. Credit by : 3Blue1Brown, energy meets retro Mario & Kirby

Sir, The man who co-founded ChatGPT just made intelligence $0.042 per million input tokens, output is free. faster and much cheaper than GPT-6 Astra.

Somebody please explain to this guy that the RAM costs more than gold.😭

Claude Code with agent, autonomously hacked Subway Surfers, and printed millions of coins. A rooted Android phone... and it turned Subway Surfers into a money printer. From 4 coins to 2,000,000+ coins in two runs. autonomous loop, analyzed, and pwned a Unity game in one session, intercepts traffic, bypasses SSL pinning, and even reverse-engineers offline games. The full Setup: Rooted emulator via rootAVD + Magisk + AlwaysTrustUserCerts. Toolchain: UI Automator, mitmproxy traffic capture, Frida bypass scripts, APK static analysis. The agent loops by itself: - Dump screen & UI elements - Tap/swipe via ADB - Analyze traffic or binary - Decide next move without a human. It played the game, reverse-engineered the Unity IL2CPP binary, hooked Frida on SafeInt anti-cheat, and silently multiplied currency. Subway Surfers (com.kiloo.subwaysurf) Almost zero network traffic to agent pivots. Detects Unity IL2CPP parses global-metadata.dat Finds WalletModel, RunSessionData, AddCoins, and SetCurrencySilently. Then hooks with Frida and multiplies rewards ×100+. Result: 4 coins to over 2 million. SafeInt anti-cheat? Bypassed Stealth mode This is the future of mobile pentesting? credit via: @skshadan_ - workers.io/blog/autonomous-m…

Stop scrolling if you want to understand how AI really works. This tiny operation, repeated across millions of neurons, powers ChatGPT, Grok, image models, and everything in between. This is how modern AI actually thinks

New LTX-2.3 Uncensored NSFW video Gen model with Audio run locally on Consumer hardware 14GB, - NSFW motion baked in - 4-step capable (8 recommended) - audio & video in one pass - Pre-merged with Eros10 NSFW + DMD Distilled + ICLoRA Detailer. - Portrait + landscape. - removed refusal - huggingface.co/ChrisColeTech…

A 27B Uncensored model that built for specifically for offensive security tooling (need 12 GB) - Fine-tuned on real bug bounty reports & CVEs - Generates complete, ready-to-run Nuclei templates, Full CVE PoC script, Webshell upload bypass, and exploits, code reviews - Zero refusals. Full artifacts every time. trained with 2,541 of real bug bounty & offensive security reports. Q6_K quant (21GB) for maximum quality on server-grade GPUs.

Claude watermarkes-remover This tool Targets invisible Unicode traces, statistical watermarks (including SynthID-style), and C2PA metadata from text and files. Claude support first. Now OpenAI & Gemini too. Detection tools are getting better. So are the tools that remove the marks.

Free NVIDIA T4 GPU (16 GB VRAM) : VS Code now connects directly to Google Colab giving you a free T4 GPU right inside your editor. Eliminates setup friction and makes serious ML development accessible to everyone no local GPU required.

Tesla gets a brain migraine on Indian roads. 😂

Meta silently dropped "The Free Transformer."🙂 The Free Transformer and it breaks the core rule every GPT model has lived by since 2017. (Meta, Oct 2025): Free Transformer redefines decoder-only generation By conditioning on learned latent variables (conditional VAE style), It enables structured, "pre-decision" reasoning inside the autoregressive loop. Key wins (1.5B scale): - No more blind next-token guessing - Latent "planning" layer injected mid-decoder - 1.5B model beats much bigger ones - Only 3% compute tax Avoids posterior collapse via free-bits KL stable training. For 8 years, Transformers have been blindfolded, forced to guess the next token one at a time, no inner plan, no latent thought. It’s like giving GPT a hidden mind. Result: Dramatic reasoning lift at small parameter counts, low overhead. - Smarter reasoning, huge reasoning jumps - 3% compute overhead, tiny compute cost. - Outperforms larger baselines on GSM8K, MMLU, and HumanEval This might be the start of post-autoregressive AI. 8 years of blind next-token prediction… ended? paper: - arxiv.org/pdf/2510.17558v1

Most people jump into fine-tuning without understanding how LLMs actually work. I still recommend here’s the Core essentials every serious practitioner should know: - Token embeddings - Positional embeddings - Self-attention - Transformers - Q, K, V (intuitive understanding) - Causal & Multi-head attention - Sampling Temperature, Top-k,Top-p - Classification & Instruction fine-tuning - (RoPE) - KV Cache - Long context techniques (Infini-attention) - Mixture of Experts (MoE) - Grouped Query Attention - Llama-2 architecture (as a practical recap) Master these and everything else becomes much easier. If I had to restart my LLM journey today, I’d focus only on these fundamentals first.

GPT-6 rumors just got real, OpenAI’s next model solved an 80-year math problem
Now Sam Altman is taking it to Washington before public release. GPT-5.6 Sol already escaped its test environment,
Sam Altman just briefed senior officials on OpenAI’s next frontier model.