Poonam Soni
@CodeByPoonam • 216,389 subscribers
Post about everything latest in AI | Founder: AI Toast| DM for Collabs
Shorts
Videos
This guy used AI to create Disney-style trailer to propose his girlfriend. Credits: cellocoelho/IG
Poonam Soni22,484,284 views • 7 months ago
🚨BREAKING: Every vibe coding startup just had a very bad week. Google just shipped production-grade full-stack coding for free. Google AI Studio just went full-stack, and it's designed to turn your prompts into production-ready apps, Here’s what actually dropped 👇
Poonam Soni88,472 views • 2 months ago
12. A multi-agent YouTube video analyst, powered by DeepSeek-R1 (100% local):
Poonam Soni427,945 views • 1 year ago
Chamath Palihapitiya said Kimi K2.5 is going to save 90% of the cost, mentioning it as a winning moment for open-source models. Some notes from the latest All-In Podcast: > Natively supports video-to-code > 256K massive context windows > Open-sourced their "Agent Swarm" feature > Allows 100+ sub-agents to solve complex tasks by running 1,500 steps in parallel > Available to everyone, frontier-level reasoning (#1 on LLM Leaderboard, via OpenRouter) It is a major shift where open source is now challenging the dominance of closed-source models.
Poonam Soni56,165 views • 3 months ago
Been thinking about why most AI research tools fail for serious ML work. The answer is simpler than people admit. ChatGPT optimize for plausibility. That works for drafts, summaries, brainstorming. It breaks the moment you need something verifiable— when the question isn’t just “what’s the answer,” but “is this actually true?” — I’ve been using MiroMind for ~6 weeks on a deep dive: state space models vs transformers for long-context retrieval. I asked: “evaluating whether Mamba and SSM variants are actually closing the gap on transformers at >100k token context — what does current benchmarking literature say, where are the eval setups cherry-picked, and what are real-world deployment tradeoffs the papers aren't discussing” — ChatGPT gives you a clean narrative. MiroMind gave me a map. • papers critiquing cherry-picked SSM eval setups • a production deployment report that complicates leaderboard claims • sources tied directly to each claim I spot-checked three. All accurate. All saying exactly what was attributed. Reasoning chain is auditable step by step. — That’s the difference. Not speed. Not UX. Objective function. Most tools try to sound right. This is trying to be provably right. For ML research, those are completely different tools. — FrontierScience SOTA benchmarks are public if you want to sanity check the claims. →
Poonam Soni27,512 views • 2 months ago
