Beware of 'anthropodenial' – the dangerous refusal to see human-like traits in animals. It's a hypothesis, not a default, and has a poor track record. Crows solving puzzles and transitive inference in animals show this bias is wrong. #AnimalCognition #ScienceFacts #Anthropodenial
Big tech won't save us, and the government can't keep up. If you want a future where AI is safer, faster, and greener, you have to build it yourself. Kwaai is a volunteer-driven, open-source movement. Learn more at kwaai.ai/ #kwaai #kwaai #OpenSourceAI #TechEthics
Our brains struggle with precise counting when numbers are close. The ratio effect—how close numbers are—dictates our accuracy and speed, not just for humans, but for animals too. It's a shared mechanism revealing how quantity is processed. #CognitiveScience #RatioEffect #kwaai
Understanding AI requires looking under the hood. Just like testing animal intelligence, we need to move beyond simple prompts and examine model activations and checkpoints. Real progress in AI safety depends on independent, verifiable research. #AISafety #MachineLearning
Our brains process quantity based on ratios, not exact counts. When the margin between two numbers shrinks, decision-making slows and error rates climb—a pattern shared across the animal kingdom. #Learn more at kwaai.ai/ #kwaai #CognitiveScience #Psychology
AI is the new colonialism. Just as land was once seized through asymmetric power, data is now being harvested under the assumption that it belongs to the few. We must recognize this digital land grab and resist it. Learn more at kwaai.ai/ #kwaai #AI #DataPrivacy
In 1904, a horse named Clever Hans fooled experts with his 'math' skills. He wasn't doing arithmetic; he was reading body language. #Learn more at kwaai.ai/ #kwaai #AI #CognitiveScience
Trusting cloud providers with your data often means losing ownership. Homomorphic encryption is changing that, allowing you to secure intellectual property while still utilizing advanced AI search. Data privacy doesn't have to be a trade-off. #kwaai #DataPrivacy #Encryption
Open weights alone aren't open science. To truly understand AI, we need to go beyond outputs and look at activations, weights, training data, and checkpoints – like biologists study model organisms. #OpenScience #AIResearch #ModelInterpretability
Clever Hans and LLMs both present a challenge: is it genuine understanding or just a benchmark score? Real insight requires rigorous experiments, free from alternative explanations, a lesson we learned from animals and now need for AI. #AI #CognitiveScience #AnimalBehavior
Clever Hans could 'do math,' but the secret? He was reading subtle human cues, not solving equations. This reveals a crucial lesson: focus on *how* a system fails, not just if it succeeds. True understanding lies in what breaks the illusion. #AI #CognitiveScience #Analytics
You are the cavalry! Kwaai is a volunteer-driven AI lab focused on open-source R&D to make AI safer, faster, and greener. With nearly 40 million in in-kind contributions, we're making a difference where governments can't. #kwaai #AIForGood #OpenSourceAI
AI is the new colonialism. Just like the 1600s land grab in Manhattan, big tech is harvesting our data—our capital asset—without true consent. We must resist this asymmetric power dynamic. #AI #Colonialism #DataRights
The biggest AI security threat isn't external hackers, it's your cloud provider. Your data could be resold or used to train their models. #AIsecurity #HomomorphicEncryption #VPK
The agent's 'yes' is only as good as its read-back. But what if the agent's understanding is flawed? Checkpointing saves the repo state, allowing you to undo mistakes. The con? Time and storage costs. #AI #AgentSafety #TechInnovation
Coding velocity is outpacing code reviews, leading to trust issues with AI agents. We need systems that can predictably translate intent into actions, offering guardrails and reversibility. #AICoding #SoftwareEngineering #TrustInAI
Tired of navigating complex software interfaces? Mahima's work on voice coding promises to end the tyranny of the UI, letting you simply talk to your applications. Learn how AI can unlock voice-assisted control for your platforms. #AI #VoiceCoding #DevTools
Voice coding risks deletion with 'git reset --hard.' The solution? Make risky actions undoable. Prioritize honest readbacks, size indications, and default undo for safer voice commands. #VoiceCoding #DevTools #SoftwareDevelopment
Voice commands for coding agents can be dangerous. A simple 'delete stale branches' can lead to lost work if the agent doesn't specify the scope. Always confirm the extent and impact before executing. #AIDevelopment #CodingAssistant #DevTools #AI #Coding #DevOps