Gabriel WC Cheung’s Post

Ever wondered how the next generation of AI agents will make decisions, learn, and interact with the world more intelligently? 🤖💡 The paper "Cognitive Architectures for Language Agents" introduces CoALA, a groundbreaking framework that combines the reasoning power of Large Language Models (LLMs) with structured cognitive processes inspired by cognitive science. CoALA positions LLMs at the core of modular systems, enabling language agents to: 🧠 Manage memory (working, episodic, semantic, procedural). 🔄 Plan and execute actions in structured loops. 🌐 Interact with complex environments, from robotics to digital simulations. What’s exciting is how this framework draws from decades of AI research to organize and enhance modern language agents, setting the stage for more adaptable, general-purpose intelligence. The future of AI agents isn’t just about solving tasks, it’s about building systems that think, learn, and grow like humans. This paper offers both a theoretical foundation and actionable insights to design smarter agents. Read more here: https://lnkd.in/egWj5RCf #ArtificialIntelligence #LanguageModels #CognitiveScience #MachineLearning #Innovation #AIResearch #LLMs #TechForGood #AIAgents #AgenticAI #AgenticAgents #AgenticWorkflows

𝐋𝐋𝐌𝐬 𝐀𝐫𝐞 𝐧𝐨𝐭 𝐒𝐭𝐮𝐩𝐢𝐝, 𝐎𝐮𝐫 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐌𝐞𝐭𝐡𝐨𝐝𝐬 𝐀𝐫𝐞 The authors of the paper "Anthropocentric bias and the possibility of artificial cognition" Raphaël Millière and Charles Rathkopf argue that Cognitive capacities of Large Language Models (LLMs) are often evaluated with biases: - 𝐓𝐲𝐩𝐞-𝐈 𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐨𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐁𝐢𝐚𝐬: Overlooking how auxiliary factors might impede LLM performance despite competence. - 𝐓𝐲𝐩𝐞-𝐈𝐈 𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐨𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐁𝐢𝐚𝐬: Dismissing LLM mechanistic strategies different from humans as non-genuine competence. The solution mentioned in the paper is basically conducting an empirically-driven, iterative approach combining behavioral experiments and mechanistic studies to map cognitive tasks to LLM capacities. Key concepts mentioned in the paper were: - 𝐂𝐨𝐦𝐩𝐞𝐭𝐞𝐧𝐜𝐞, which is the system’s internal knowledge or computational capacity. - 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞, which is observable behavior exercising the capacity. These concepts are important because misjudgments about LLMs arise when performance failures are interpreted as competence failures due to auxiliary factors. The paper also explained Anthropocentric Bias as: - 𝐓𝐲𝐩𝐞-𝐈, which is the misinterpretation of LLM performance failures due to auxiliary task demands, computational limitations, or mechanistic interference. - 𝐓𝐲𝐩𝐞-𝐈𝐈, which is the assumption that non-human strategies indicate a lack of genuine competence. #LLM #LLMs #ai #cognition #machinelearning #AGI #deeplearning

Curious about how AI could learn #language and #emotion just like humans? The #Xzistor #Mathematical #Model of #Mind presents a revolutionary approach, integrating cognitive architecture and reinforcement learning to push beyond traditional AI. Explore how this model could shape the future of #intelligent #agents. https://lnkd.in/e7Zat8fx #AI #CognitiveScience #ReinforcementLearning #ArtificialIntelligence #FutureOfAI

New Blog! Check it out! I explore the grasp of human logic LLMs exhibit through 'Waiting for Godot'" In this blog, I delve into the unconventional yet profoundly insightful experiment of evaluating Large Language Models (LLMs) using Samuel Beckett's iconic absurdist play, Waiting for Godot. Why this play, you ask? Its essence of existential questioning and the infinite wait for meaning provides a unique rubric for assessing the capabilities of AI in grasping human logic and the nuances of literary themes. 🤖 Among various models tested, GPT-4 emerged as a fascinating case study, showcasing an uncanny ability to understand and replicate the thematic essence of Beckett's work. This journey not only highlights the capabilities of current AI technology but also prompts us to think about the future of AI in literary analysis and philosophical discourse. I invite you to read about the intriguing findings around the intersection of AI, literature, and philosophy. Let's discuss the potential of AI to transcend its computational boundaries and venture into the realm of existential inquiry and artistic interpretation. https://lnkd.in/exY2NS7K #AI #Literature #Philosophy #WaitingForGodot #GPT4 #ArtificialIntelligence #Innovation #TechInsights

This article will brief about the mathematical intuition behind LoRA and QLoRA #llm #genai #finetuning #lora #ai

The concept of “emergent abilities” in large language models (LLMs) stands as a focal point of intrigue and debate. The journey through this landscape reveals two contrasting narratives. On one hand, researchers have observed what seemed like abrupt, unpredictable improvements in LLM capabilities as they scaled, a phenomenon initially labeled as emergent, drawing parallels with phase transitions in physics. This notion painted a picture of AI development as a series of sudden leaps into new realms of possibility, a perspective that has colored our understanding and expectations of AI’s potential and risks.

If you haven’t I highly recommend reading this commentary paper. There’s so much talk about AI augmenting human skills, but recent research points to the opposite. Why? How could we design AI that actually augments our skills? Why should we model AI after bicycles? This is the subject of my PhD research. In the following days, I will be sharing some thoughts on the answers, but for now I highly recommend reading this. (If you don’t have access comment and we can find a pdf 😉)

This paper is worth your read. It explains why language is primarily a tool for communication rather than thought and implies the limitations and boundaries of the Autoregressive LLMs (AR-LLMs) approach towards machine intelligence, which many of the hottest AI applications on the market today are based on. In my opinion, we are witnessing the biggest bubble at our time. The AI bubble is order of magnitude bigger than the Dot-com bubble in the 2000. When companies have tried and realize their AR-LLMs investment can't bring much revenues to their business, the bubble will burst badly. Most will fail but some will thrive thereafter. https://lnkd.in/gcngxjDG https://lnkd.in/gZ92QxAs

🤖 With generative artificial intelligence (#GenAI) transforming the social interaction landscape in recent years, large language models (#LLMs), which use deep-learning algorithms to train GenAI platforms to process language, have been put in the spotlight. Led by Prof. LI Ping, Dean of the Faculty of Humanities and Sin Wai Kin Foundation Professor in Humanities and Technology, PolyU’s research team has investigated the next sentence prediction (#NSP) task, which simulates one central process of discourse-level comprehension in the human brain to evaluate if a pair of sentences is coherent, into model pretraining and examined the correlation between the model’s data and brain activation. 💬 The study found that LLMs perform more like the human brain when being trained in more similar ways as humans process language. 🧠   More: https://polyu.hk/WKTCi #TheHongKongPolytechnicUniversity #FacultyofHumanities