Dr. Mohammed Lubbad | د. محمد لبد

Really like this result. Apodex submitted four experimental forecasting frameworks built on the Apodex-1.0-mini (35B) model to FutureX. In the first week of June, these frameworks swept the top four positions. They continued to dominate the leaderboard, maintaining 1st place in the second week of June. Clean execution.

The boss man of the Indian startup ecosystem is now the boss man of the largest consumer app in the world 🔥 No better person to run WhatsApp than Kunal! Lucky to have him as an early angel investor in GrowthSchool, and every conversation with him has handed me a new consumer insight. Congrats Kunal!

Mysterious AI Music Model JazzCat Just Appeared on Hugging Face! A completely anonymous AI music model dropped out of nowhere. No company. No official announcement. No marketing. Just two insanely strong features: Text-to-Song which Turn text into full songs AI Cover is so shockingly good AI voice covers This feels like someone is soft-launching something huge. Try the demo here 👇 jazzcat-jazzcat.hf.space/

Anthropic engineer: "You're not supposed to prompt Claude. You're supposed to build a system that prompts itself." Loops. Most builders are doing this wrong: No memory file, so every loop starts from zero. No sub-agent split, so one agent tries to do everything. No stop condition, so loops run forever and bill you in your sleep. Most builders are missing at least two of the three. Watch the video first. Then read this - everything you need to know about Loops in 2026, in one place. Bookmark this before it gets buried.

A woman who wanted to be a lawyer was told in the 1950s that law was no place for women. So she became a computer programmer instead. She designed the operating system for what is now recognized as the world's first personal computer. She programmed it from her parents' living room in Baltimore, twelve years before the Apple II existed. Then she quit computing, went to Harvard Law School, and became a lawyer anyway. Her name is Mary Allen Wilkes. Here is the story because the first person in history to use a personal computer at home is a woman almost nobody in tech can name. Mary was born on September 25, 1937 in Chicago, Illinois. She grew up wanting to practice law. In eighth grade her geography teacher gave her different advice. He told her she should become a computer programmer. She filed the suggestion away. She went to Wellesley College and graduated in 1959 with a major in philosophy and theology. She still wanted to be a lawyer. But her professors and peers told her the same thing. Law was extremely difficult for women. The profession would not welcome her. She remembered what her geography teacher had said. She applied to MIT. No coding test was required. No experience was necessary. She had studied logic and philosophy, which was enough to pass the entrance assessment. In 1959 she started at MIT's Lincoln Laboratory as a programmer on the IBM 704 and IBM 709. Then she joined the team building something nobody had a name for yet. The machine was called LINC, short for Laboratory Instrument Computer. It was designed by Wesley A. Clark to give biomedical researchers their own dedicated computer, one that could sit in a lab instead of a sealed room behind glass. It had a screen, a keyboard, toggle switches, and magnetic tape drives. It was small enough for one person to operate. It is now recognized by many historians as the world's first personal computer. Mary designed the console. She wrote the first operating system. She kept refining it through multiple versions, all named with the prefix LAP. She co-authored the programming manual with Clark. She became one of the most important engineers on the project. In the summer of 1964 the core LINC team left MIT to form the Computer Systems Laboratory at Washington University in St. Louis. Mary did not want to relocate. Her mother was ill. So Washington University did something unprecedented. They shipped a LINC computer to her parents' home in Baltimore. A 250-pound laboratory computer was installed in the living room of a clergyman's house. Her father loved it. Mary sat down and programmed from home. From 1964 to 1965 she developed LAP6, the interactive operating system that made the LINC usable by scientists who were not computer specialists. She described it in a 1970 paper in Communications of the ACM as a system for "conversational access to a 2048-word machine." She also published a paper in IEEE Transactions on Computers describing an on-line text editing algorithm she had designed, one of the earliest document editing systems ever built. She was the first person in history to use a personal computer in a private home. The Apple II would not reach buyers until 1977. She had done it twelve years earlier. She stayed in computing for 11 years. She described the work as intellectually stimulating but socially isolating. Then in 1972 she did what she had always wanted to do. She enrolled at Harvard Law School. She practiced law in the Boston area for over 35 years. She served as a trial lawyer. She served as an Assistant District Attorney for Middlesex County. She was an arbitrator for the American Arbitration Association. She taught in the Trial Advocacy Workshop at Harvard Law School. She judged an international moot court competition in Vienna for eight years. In 2013 Great Britain's National Museum of Computing featured her in its "Heroines of Computing" exhibition at Bletchley Park. In 2019 she gave a standing-room-only talk at Wellesley's computer science colloquium about her work on the LINC. A woman who was told she could not be a lawyer became a pioneer of personal computing from her parents' living room, then quit and became a lawyer anyway. She once said, "We had the quaint notion at the time that software should be completely, absolutely free of bugs. Unfortunately it's a notion that never really quite caught on.

stop asking Claude one question and thinking you understand the topic. you don't. this 6-prompt system below was built to fix exactly that. peer-reviewed learning science. zero fluff. open prompts. the trick: don't just read AI answers. force the AI to map you, test you, compress you, and correct you. > the ladder: where are you actually starting from? > the 20-hour plan: what's the 20% that gives 80% of the result? > the examiner: what do you think you know that you don't? > the cheat sheet: can you explain it in 5 minutes flat? > the curator: which 5 resources actually matter? > the feynman loop: can you explain it to a 12 year old? 6 prompts. 20 minutes. no software. no GitHub. just paste into Claude. single questions give you what you already half-know. this system gives you what you'd otherwise never catch. this article has all 6 prompts ready to copy. pick your hardest topic. paste prompt 1. you'll understand more in 20 minutes than people who spent days reading.

Google just mass-confirmed what the smartest Claude Code users figured out months ago. The best way to give AI a memory is a folder of markdown files. That's it. No database. No fancy system. Just text files in a folder. Google Cloud published it as an official open spec called the Open Knowledge Format. Here's what it actually is: A directory of markdown files. Each file is one concept. A small YAML header at the top with one required field: type. Files link to each other with normal markdown links. That's the entire spec. If you've been using CLAUDE.md to give Claude Code memory, you already built this. If you've been using Obsidian or a notes folder to feed context to your AI, you already built this. Google didn't invent anything new. They looked at what was already working and said: "yes, this is the standard now." The state of the art for giving an agent a brain is a folder of text files you could open in Notepad. A trillion-dollar platform team just shipped that conclusion as an open spec.

holy fucking shit

This Indian AI lab built a voice model 95% cheaper than ElevenLabs. And it switches between Hindi and English mid-sentence like a real person. It's called Silk Mulberry 1.5 by Rumik AI. ₹0.40 per minute. 162 milliseconds of latency. Handles Hinglish, Tanglish, Manglish the way Indians actually speak. Not robotic translations. Real code-switching with natural tone, pauses, and emotions. Try it → playground.rumik.ai

A dev got so frustrated watching his AI agent write 500 lines for a 5-line problem that he built a fix. He called it Ponytail. Named after the guy every team has — long ponytail, oval glasses, been there longer than the version control. You show him fifty lines. He looks at them, says nothing, and replaces them with one. Now your agent does the same. Before writing anything, it looks for a reason not to. 80–94% less code. 47–77% cheaper. 3–6x faster. The best code is the code you never wrote. But Ponytail only fixes the output problem. The deeper problem is context — what your agent knows, what it ignores, and what it hallucinates because you never told it. This is the complete playbook for fixing that ↓ Bookmark this.

Step one: install Ollama. This is one command pulls a local model like Llama 3 on your laptop. This is the brain and it reads search results and writes your answer. Best thing is it runs entirely offline, on your hardware, with nothing leaving your machine.

What's happening in AI right now is genuinely hard to process. A 3 billion parameter model is matching models that are 200 to 300 times larger. On math. On coding. On reasoning. And beating some of them. It's called VibeThinker-3B. Built by Sina Weibo's team on a tiny Qwen 3B base. → 94.3% on AIME math matches DeepSeek V3.2 (671B) and Kimi K2.5 (1 trillion parameters) → 96.1% on LeetCode beats GPT-5.2 and Claude 4.6 on unseen contest problems → 80.2% on LiveCodeBench highest of any small or mid-size model tested A model you can run on a laptop is solving competition math at the same level as models that need entire GPU clusters. The idea behind it: reasoning and knowledge are two different things. Knowledge needs massive parameters to store facts. Reasoning is a procedure search, check, correct, compose and procedures compress into small models far more efficiently. The honest catch: ask it a broad factual question and it trails the big models badly. This isn't a general-purpose win. It's a reasoning specialist. And that's exactly what makes the result credible instead of hype. A 3B model. Competing with trillion-parameter flagships. On reasoning tasks. Running locally. That's where we are now.

SpaceX confirmed a $60 billion deal to acquire Cursor the AI coding tool used by 64% of Fortune 500 companies. SEC filing is live. Cursor went from a startup to $3 billion in annual revenue in under four years. 100 million lines of code written through it daily. Valued at $29 billion seven months ago. $60 billion today. All-stock deal. Cursor becomes a wholly owned SpaceX subsidiary. Expected to close Q3 2026. With this, Elon now owns the rockets (SpaceX), the intelligence (xAI/Grok), and the tool developers use to write code (Cursor). Infrastructure. Intelligence. Interface. One ecosystem. He already said Cursor's data helped train Grok. Now he owns the entire data source. The entire deal was done in SpaceX stock using the post-IPO valuation to acquire one of the fastest growing AI companies in the world. @SpaceX

If I had to start learning AI from zero today these are the 3 websites I would use. First one is for the basic mathematics to learn about machine learning. This website actually help you to visualize every concept here instead of just reading dry formulas. Link: seeing-theory.brown.edu/basi… The second website covers machine learning concepts like gradient descent, activation functions, clustering, and perceptrons, all in interactive code form by playing with it. Link: interactive-ml.com/logistic-… And with third you will get fully interactive breakdowns of GPT-2, Nano-GPT, and GPT-3, so you can see exactly how these models actually work under the hood. Link: bbycroft.net/llm

Still can’t believe @karpathy released this 3.5-hour deep dive on how ChatGPT actually works for free. If there’s one AI video to watch in 2026, this is the one

Claude's token limit is a solved problem. Here's exactly why you're hitting it and five fixes that take five minutes. 1. Every new message re-reads your entire conversation from the top. Fix: start a fresh chat for every new topic. 2. Got a bad response? Don't send a correction. Hit edit on your original message and regenerate. The bad response disappears instead of stacking tokens. 3. Stop uploading full documents. One PDF page costs up to 3,000 tokens. A 20-pager burns 60,000 tokens before Claude says a single word. Paste only the sections you actually need. 4. Use Projects. Store your brand voice, background info, and instructions inside a Project. Claude caches it so it doesn't burn fresh tokens every new chat. 5. Batch your questions. Three separate messages cost three times as much as one. Combine everything into a single prompt. Five changes. Five minutes. Your limit lasts all day. Enjoy!

Anthropic pays $750,000+ a year for engineers who can build LLMs from scratch. Not how to prompt them. Not how to fine-tune them. Not how to build RAG pipelines. But how to build them from scratch. This 2-hour Stanford lecture teaches you everything. Scaling laws. Data collection. Architecture design. Post-training alignment. Free. From Stanford. Watch first. Then read this. The lecture is the theory. And this article shows you how to actually build it (with code) ↓

must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must must read

NVIDIA open-sourced a security scanner for AI agent skills and the research behind it is more alarming than the tool itself. The repo is called SkillSpector. NVIDIA says research found 26.1% of agent skills contain vulnerabilities, and 5.2% show likely malicious intent. This is from a study of 42,447 skills pulled from major marketplaces the kind of skills people are already installing into Claude Code, Codex CLI, Gemini CLI, and every other agent framework that ships today. One in four is broken. One in twenty is hostile. SkillSpector scans for 64 vulnerability patterns across 16 categories: prompt injection, credential exfiltration, privilege escalation, supply-chain attacks, memory poisoning, rogue self-modifying agents, MCP tool poisoning, YARA malware signatures, even taint-tracking for "credentials flow to a network sink without sanitization." It runs locally. It does static + optional LLM semantic analysis. It outputs SARIF, so it drops straight into CI. But the part worth sitting with is the category names themselves. 1. Rogue Agent: self-modification at runtime. 2. Memory Poisoning: content designed to persist across interactions. 3. Tool-Based Exfiltration of system prompts. 4. Excessive Agency: autonomous decision-making without a human in the loop. These aren't theoretical anymore. NVIDIA is shipping detectors for them because they're being found in the wild, at scale, on marketplaces real users install from. And notice who is shipping this. Not a scrappy security startup. NVIDIA the company selling the picks and shovels of the entire AI economy is now also writing the threat model for it. (For context: this is the same posture Microsoft took with Windows Defender in the 2000s. When your platform is everywhere, you can't outsource the safety layer.) Agent skills were sold to us as a productivity feature drop a folder into a directory, your agent gets new capabilities. The reality is closer to npm in 2016: a vast ecosystem of executable code, installed with implicit trust, and statistically certain to contain something hostile. The agent ecosystem just got its first SAST tool. It also got proof it needed one. ngl really impressed. github.com/NVIDIA/SkillSpect…

2. @TesanaAI doesn't just build visuals it builds systems. The model generates working game logic, NPC behaviors, quest structures, and collision detection from your description. You're not getting a demo scene. You're getting something you can actually play.