Shubh

Here are S-tier learning advises if you're studying AI: 1. Learn by building, not just watching courses. 2. Don't chase every new framework. 3. Choose projects that genuinely interest you. 4. Read papers. 5. Implement papers yourself. 6. Study mathematics deeply. 7. Learn probability and linear algebra. 8. Build strong theoretical foundations first. 9. Develop critical thinking. 10. Understand why algorithms work.

If you're into ML, you need to study physics also. 😭

I studied production systems of Stripe, Square, Uber, and Google and made the best list of resume worthy backend projects that you can build: 1. Immutable Double-Entry Ledger Service 2. Distributed Rate Limiter (API Gateway Edge) 3. Durable Workflow Orchestrator (Saga Engine) 4. Change Data Capture (CDC) Event Bus 5. RBAC Engine (Zanzibar-Style ReBAC) 6. Raft-Replicated Distributed Key-Value Store 7. Real-Time Feature Store + Inference Gateway

Top of the line INTERNET RESOURCES for studying machine learning mathematics: ➜ Distill ➜ Setosa .io ➜ 3Blue1Brown ➜ Explained .ai ➜ CNN Explainer ➜ ImmersiveMath .com ➜ MLU-Explain by AWS ➜ TensorFlow Playground ➜ Seeing Theory by Brown ➜ StatQuest by Josh Starmer ➜ Neural Networks and Deep Learning

day 0 of vibe machine learning

JAPANESE EDITION OF FABLE JUST DROPPED Sakana Fugu is a new multi-agent orchestration system from Sakana AI. • SWE-Bench Pro: 73.7%. • TerminalBench 2.1: 82.1%. • LiveCodeBench: 93.2%. • LiveCodeBench Pro: 90.8%. • Humanity’s Last Exam: 50.0%. • CharXiv Reasoning: 86.6%. • GPQA-Diamond: 95.5%. • SciCode: 58.7%. • τ³ Banking: 20.6%. • Long-Context Reasoning: 73.3%. • MRCRv2: 93.6%. Currently their API supports: ➜ OpenAI-compatible endpoints ➜ Tool calling ➜ Web search ➜ Multimodal inputs ➜ Structured outputs ➜ Adjustable reasoning depth Fugu Ultra costs: • $5/M input tokens • $30/M output tokens • $0.50/M cached tokens

If you don't know her channel. You missed out on life.

POLYMARKET "ALLEGEDLY" PAID CREATORS TO SHOW FAKE VIRAL WINS TO ATTRACT NEW USERS. Report of Wall Street Journal: • More than 1,100 deceptive promotional videos • Creators placed fake bets on replica sites like “poiymarket .com” • The trades and balances shown were not real • 118 videos claimed nearly $900,000 in winnings • Those bets would have actually lost around $166,000 • Creators were reportedly told not to disclose Polymarket’s payments • The campaign generated over 140 million views After the WSJ started asking questions, several videos were deleted, and the replica sites went offline. Polymarket says it will audit its promotional content. Yeah, sure dude. And I'm Trump's son. The funny thing is they marketed themselves as a source of real world truth. It now appears to have grown through manufactured social proof. teapaw.org/neilmhta/status/206850…

These are all the Docker commands you need.

My machine learning index so far. Next up: - Decision Trees - Ensemble - Bagging boosting - Xgboost - Lightgbm (I'll write a bit about it, not included in the tutorial I'm referring to)

Your system design knowledge is incomplete if you don't know these latency numbers.

If you’re a programmer. YOU NEED TO STUDY MATH.

Daily reminder to deploy all your side projects. Recruiter isn't going to clone your project and run npm run dev.

No programmer can ever beat someone who codes this. credit: @xxkaixi

Wrote 80+ pages of notes for machine learning in 3 days. About 100 more pages to go.

SOMEONE BUILT OPEN SOURCE PALANTIR. - Real-Time Tracking -10,000+ commercial, military and private aircraft live on a 3D globe - 2,000+ satellites, including ISS - 1,400+ worldwide CCTV camera feeds - Earthquakes, wildfires, nuclear facilities and severe weather - DNS record lookup and enumeration - WHOIS domain intelligence - SSL/TLS certificate transparency - BGP routing and ASN lookup - Threat intelligence and IP reputation All running on a 3D interactive globe with day/night cycle, 20+ live API feeds, and a SIGINT news aggregator.