Mustafa 🦄 (@_mustafa30_) header
Mustafa 🦄 (@_mustafa30_) profile picture

Founding AI Eng | 24k+ on Instagram | Agents, MCP, LLMs, Evals and everything in between 🏆

London, United Kingdom
Joined March 2019
Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)Photo by Mustafa 🦄 (@_mustafa30_)
Trying to build a better network on here. If you’re into tech, startups or the internet, let’s connect.
2
1
16
new system prompt rules when interacting when building your own agent with frontier models: 1. give it judgement not rules 2. design interfaces and think about your tool designs instead of just giving examples more in the thread below
1
23
3. models have gotten quite competent at progressive disclosure. no need to inflate sys prompt. 4. No need to repeat instructions in sys prompt. simple tool descriptions do the work.
2
13
Mustafa 🦄 retweeted
When I'm rich, there will be signs
183
771
56
12,891
469,833
Mustafa 🦄 retweeted
I had so much fun creating the Self-Improving AI Agents course with @achowdhery, and teaching it twice in one year at Stanford! We also collaborated with Stanford Online to make the course available online: YouTube: youtu.be/6YnLB0XbTnI?si=MVwR… The field is moving incredibly fast, but we tried to focus on the core concepts that help us build better AI systems. I hope you enjoy it!
41
119
6
1,139
107,636
Baidu just open-sourced an OCR model that reads entire 40-page documents in one shot. It's called Unlimited-OCR. 3 billion parameters but only 500 million active during inference. Runs 100% locally on your machine. Why this matters: traditional OCR tools chop documents page by page. Tables that span two pages break. Reading order gets lost. Cross-page context disappears. Unlimited-OCR processes the whole document at once. 32K context window. Text, formulas, tables, reading order all preserved across pages. Output comes out as clean structured Markdown. → 93% accuracy on the standard benchmark. +6 points over the baseline. → Error rate stays below 0.11 even past 40 pages. → Multilingual out of the box. → 2.12 million downloads on Hugging Face last month. 14,600 GitHub stars. For context: Amazon Textract, Google Cloud Vision, and Azure Document Intelligence all charge per page. This runs locally for free.
81
446
31
3,971
417,439
Why has Vercel become the default hosting platform for vibe coders? There are plenty of alternatives. What made Vercel win?
1
25
🔗
Why did Cursor become the default AI IDE? There are plenty of alternatives. What made Cursor win?
1
17
Highly recommended read Want an interesting use of LLM and deep learning
We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2. Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication. We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating. 🧵👇
29
if you're into ai agents, robotics, open source, solopreneurship, startup ideas, product design, human psychology, cognitive science, systems thinking, cybersecurity, machine learning, computer vision, hardware hacking let's connect 🤝
2
1
32
Someone just open sourced a complete AI agency and it hit 50K GitHub stars in under two weeks. It's called The Agency. And it's not a prompt template. It's 147 specialized AI agents across 12 divisions - engineering, design, marketing, product, QA, support, spatial computing
1
32
Okay, let me get this straight... 1. You have 1-click n8n, Zapier, and Lindy AI agent templates to automate your content, outreach, research, reporting, anything. 2. You can clone your voice with 5 seconds of audio. 3. You can run billion-parameter AI models on your laptop for free. 4. You have agents that write, debug, and ship code while you sleep. 5. You can turn scripts into talking head videos with your face and voice. 6. You can spin up AI agents to negotiate bills and cancel subscriptions. 7. You can summarize 300-page PDFs in seconds. 8. You can turn rough Looms into polished blog posts instantly. 9. You can generate personalized landing pages for every visitor. 10. You can talk to your data like it's ChatGPT. 11. You can train a custom GPT on your company docs and have it answer support tickets 24/7. 12. You can A/B test 100 ad creatives before you spend a dollar... all AI-generated. 13. You can scrape the internet, analyze competitors, and generate go-to-market strategies overnight. 14. You can deepfake yourself into any language and accent to localize content globally. 15. You can run AI to audit your finances and suggest tax optimizations (for a pretty much free second opinion) 16. You can generate an entire week of social content with tone, brand, and format baked in. 17. You can create full e-commerce stores from just a CSV and a brand vibe. 18. You can deploy voice agents that call leads, qualify them, and update your CRM... in your voice. 19. TLDR; you can do a lot, my friend This is the greatest unlock for solo builders since the App Store. Enjoy it.
196
666
53
7,433
700,596
Mustafa 🦄 retweeted
Do you know some coders in India who don’t have access to a laptop?
23
30
75
17,781
AsapSCIENCE: This channel uses science to explain everyday phenomena and answer common questions.
1
53
Vsauce: This channel explores a wide range of topics, including science, philosophy, and mathematics, in a fun and engaging way.
153