Joshua C.’s Post

Just finished the course “Introduction to Artificial Intelligence” by Doug Rose! Check it out: https://lnkd.in/gh9sbPgh #artificialintelligenceforbusiness #artificialintelligence. Presented in the easiest way to understand. Doug really helps to relate to things which are familiar. A great introductory course!

ReGenesis: LLMs can Grow into Reasoning Generalists via Self-Improvement Whether a LLM can self-improve by generating data to train itself,  is still a controversial question without concrete answers. Papers present evidence to both standing sides of this question. In our exploration, we have found that STaR, the most commonly used LLM self-training method, doesn’t generalize well to out-of-distribution (OOD) tasks. We hypothesize it is due to their self-synthesized reasoning paths being too task-specific, lacking general task-agnostic reasoning guidance. To address this, we propose Reasoning Generalist via Self-Improvement (ReGenesis), a method to self-synthesize reasoning paths as post-training data by progressing from abstract to concrete. We show that ReGenesis achieves superior performance on all in-domain and OOD settings tested compared to existing methods such as STaR. For six OOD tasks specifically, while previous methods exhibited an average performance decrease of approximately 4.6% after post training, ReGenesis delivers around 6.1% performance improvement. Here is the arxiv version of our paper: https://lnkd.in/gTns7cC4 Kudos to co-authors Xiangyu P. Congying Xia Xinyi Yang Chien-Sheng (Jason) WU Caiming Xiong. Ping us with your questions! We love feedbacks :)

so what it is about DeepSeek? a quote from the DeepSeek R1 paper, where the authors reflect on their experience with the model(s): 'One of the most remarkable aspects of this self-evolution is the emergence of sophisticated behaviors as the test-time computation increases. Behaviors such as reflection—where the model revisits and reevaluates its previous steps—and the exploration of alternative approaches to problem-solving arise spontaneously. These behaviors are not explicitly programmed but instead emerge as a result of the model’s interaction with the reinforcement learning environment. This spontaneous development significantly enhances DeepSeek-R1-Zero’s reasoning capabilities, enabling it to tackle more challenging tasks with greater efficiency and accuracy.' https://lnkd.in/g-3vZwDT

Our new research article come in Evolving System, Q2. With the title of Evolving knowledge representation learning with the dynamic asymmetric embedding model 😎. https://lnkd.in/g9H-Xqny

Excited to share our new paper on reasoning, REGENESIS: LLMs Can Grow Into Reasoning Generalists via Self-Improvement. ReGenesis delivers a 6.1% boost on OOD tasks while other baselines see a 4.6% drop. Check out our paper to see how we improve LLM’s general reasoning capabilities through innovative self-improvement methods!

Just finished Artificial intelligence beyond engineering (Beta)! Check it out: https://lnkd.in/dbwgFjgB #businessintelligencetools #business #artificialintelligenceforbusiness #artificialintelligence

ELAINE's analysis of a paper - "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning". Here is the caveat. Following are excerpts from ELAINE's analysis: "we can see that by distilling DeepSeek-R1, the small model can achieve impressive results. However, there is still one question left: can the model achieve comparable performance through the large-scale RL training discussed in the paper without distillation? " "we can draw two conclusions: First, distilling more powerful models into smaller ones yields excellent results, whereas smaller models relying on the large-scale RL mentioned in this paper require enormous computational power and may not even achieve the performance of distillation. Second, while distillation strategies are both economical and effective, advancing beyond the boundaries of intelligence may still require more powerful base models and largerscale reinforcement learning. " ELAINE knowledge map on Deepseek

Check ELAINE in action unravelling #DeepSeek in 4 steps: 1)Discovers facts, identifies the causation and consequential reasoning to augment human intelligence 2) From the author's perspective, determine what are the stories, key focus, headwind, tailwind and relevancy of stories. 3)From a logical perspective to understand it the logical hierarchy of topics and associated stories with respect to the overall context. 4)Form relations that connects the stories.

I just finished the “Advanced Prompt Engineering Techniques” course, which Morten Rand-Hendriksen very well taught! More than just teaching how to better use AI-LLMs it also taught me to improve my own thinking.  https://lnkd.in/g3zbSwUz #artificialintelligence #promptengineering.

Beyond Patterns Uncovering the Limits of LLMs in Mathematical Reasoning A new study reveals that, despite their achievements, current large language models (LLMs) fall short in genuine mathematical reasoning, relying heavily on pattern recognition that leads to inconsistent results. To advance AI’s problem-solving capabilities, we must prioritize models capable of formal, logical reasoning, especially in areas like math and science. The findings underscore the need for improved benchmarks and research into models that can handle complex, nuanced reasoning rather than just mimic patterns in data. https://lnkd.in/gqXjH-zh https://lnkd.in/grZj3nDR #AIAdvancements #AIBenchmarks #AICommunity #AIInnovation #AIModels #AIResearch #AIStudy #BenchmarkingAI #CognitiveAI #ComputationalMathematics #DataScience #DeepLearning #FutureOfAI #GSM8K #GSMBenchmark #LanguageModels #LLMAnalysis #LLMCapabilities #LLMStudy #LogicInAI #MachineIntelligence #MachineLearning #MathematicalReasoning #ModelPerformance #PatternRecognition #ResearchInsights #SymbolicReasoning