Inna (@InnaKalini97034) profile picture

Crypto lover now discovering AI Experimenting, learning actively, and sometimes worrying out loud: will humanity make it through this intelligence race?

Joined September 2024
Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)Photo by Inna (@InnaKalini97034)
2. Prompting is Dead. Context Wins. Stop trying to "trick" models with clever wording. The real skill moving forward is Context Engineering—structuring instructions, definitions, and boundary files so an background agent acts like a true teammate.
1
28
Our primary output is shifting. We are no longer line-by-line conductors writing syntax; we are orchestrators designing the system (the factory) that produces the code. As an aspiring vibe coder trying to level up over the next 4 days—this is a massive wake-up call.
1
27
If you are automating workflows right now: Are you spending more time tweaking your prompts, or are you starting to build strict testing structures around your agents?
24
We are officially entering the era of "Agentic Engineering," and the traditional SDLC is dead. Breaking down Day 1 of the new Google & Kaggle AI Agents Intensive. If you are still "vibe coding" (building pipelines on pure prompts and prayers), -that’s a massive trap, why?
1
1
70
1. The 10% Rule (Model vs. Harness) The raw LLM is only 10% of the puzzle. The other 90% is the Harness—the isolated sandboxes, tool schemas, and evaluation loops. Without a strict verification harness, you aren't engineering; you're just burning tokens.
1
29
Elon at Israel’s Smart Mobility Summit: ‘The overall goal is to maximize the probability that civilization has a great future.’ From robot abundance to Mars cities — this isn’t sci-fi anymore. It’s happening. Thank you @elonmusk for building the future we all dreamed of. 🚀✨
1
119
When you finally open up to AI about all your worries... Robot: "Understood. Deleting your problems now 🗑️" Me (and my bunny): 😭 Therapy of future hits different. 😂 #AIRobot #RabbitLife #AIHumor
139
Twice this week, two scientists said the same thing: the "AI will cure cancer" narrative is mostly a sales pitch. The data is a mess. Medical research is facing a reproducibility crisis—studies can't even be replicated by the same scientists who published them. Meanwhile, cancer survival rates remain brutal. AlphaFold works because protein structures are clean, measurable, and standardized. Cancer data? Scattered across thousands of hospitals, full of errors, and built for billing codes—not science. The infrastructure gap is the real bottleneck, not model size or compute. Only 11% of landmark pre-clinical cancer trials are actually replicable. That's not an efficiency problem. That's structural rot. We're being sold miracles while the boring, hard work of fixing data infrastructure gets ignored. One project gives me hope: UK Biobank. 500,000 participants, whole-genome sequencing plus phenotypic data, and decades of linked health records. Done right. This is the blueprint. Clean, standardized, reproducible data is the only path to actual medical breakthroughs. The lesson from AlphaFold's Lasker Award: high-density, standardized data beats bigger models every time. If the data isn't linked and reproducible, it isn't research. It's noise. We're not moving fast enough. The battle for the future of medicine isn't about who has the biggest model. It's about who fixes the plumbing first. Anyone else feel like we're being sold a bill of goods here, or is it just me?
1
116
Rabbit tried to run a serious carrot-delivery meeting… until the AI vacuum decided he was “large organic dust-clump” and tried to deep-clean him mid-call 😭 “This is how it starts. First they clean your floor. Next they’re trying to archive your tail.” If you can’t beat them… colonize them. Who’s winning in your house — you or the smart devices? #RabbitAI #AIVacuum #SmartHomeFail #BunnyLife
🤖 Made with AI
1
123
I watched an Andrej Karpathy interview. It confirmed something I’ve been noticing for months. He went from writing 80% of his own code in December 2024 to delegating 98% to AI agents. Here’s the thing nobody talks about: It’s not just prompt mastery. You also need deep expertise in what you’re building. Six months ago I couldn’t ship a working app. Now I can ship one in days. Why? I know exactly what I want and can direct the agents precisely. Subject expertise + prompt mastery. That’s the combo. What’s your experience? Share in comments — I’d love to hear your use cases and concerns 👇
1
102
My rabbit asked for max relaxation... got floating pellets, disco lights, and a 100mph head-petting robot arm 💀 "Artificial Intelligence... natural stupidity." As someone still figuring out AI... this feels a little too accurate. Would you trust a smart home with your comfort? Or nah? #RabbitAI #AI #SmartHome #BunnyLife
🤖 Made with AI
1
68
I just saw this new course from DeepLearning.AI and Oracle: Agent Memory – Building Memory-Aware Agents. Prompt engineering is fun, but agents forget everything after one chat. And I also experiencing that all the time while building agents. So providing them with real memory so they could learn over time, remember mistakes, get better - is the only way to turn them into autonomous helpers, rather than dumb chatbots. Anyone planning to take it or took it already? What you think about agents with memory and the best ways of providing such? #AI #AIAgents #MemoryAI #DeepLearning
60
I like the new @grok feature of compiling a few pictures into a video 😍
🤖 Made with AI
1
1
83
This Pentagon vs Anthropic thing is hitting me hard. They basically told Anthropic: loosen your AI safety rules for military use... or we'll wreck your business. They threatened to label them a “supply chain risk” and cut off government work. Feels like straight coercion. Not loving where this is going. 🧵
1
2
109
4/5 "Mass surveillance is only the 10th scariest thing governments could do with AI control..." 10th. The top ones? Probably even darker. This dispute is small now, but it's showing how fast the balance tilts from labs to state power.
1
54
5/5 After reading this I just sit and stare at my screen for a while, wondering what we do next. What worries you most here — the leverage? The surveillance risk? The race dynamics? Reply if it hit you too.
49
My rabbit tried getting a recipe from AI... ended up with a car battery "super-snack" ⚡"This is why I worry about trusting AI for important things..." Back to safe lettuce. No explosions today. What’s the weirdest AI answer you got? #AI #Hallucination #RabbitReact
🤖 Made with AI
77