Yasir_Mohammed .’s Post

Fascinating stuff. When thinking about it carefully, it makes sense. Our words are an outward reflection of our thoughts. LLMs are a statistical representation of our use of words. In a sense, they are designed to mimic expression of human thought. So we might expect the kinds of things humans do to create a better work product will also produce better results with LLM agents. Things like oversight, peer review, specialization, collaboration, consensus, and voting all have a place in our world and appear to have one in an agentic one as well. https://lnkd.in/g65fFBPg

On YouTube🎙️ Dive deep into the Large Language Model Evaluation with Gauthier Guinet, on our latest LLM Daily episode. 🌐 Elevate your understanding of LLM Evaluation by tuning in for insightful, scientific discussions. #LLMEvaluation #AIResearch #ScientificInsights 🚀 Youtube: https://lnkd.in/gQa5funr

Dear network, I'm excited to share a thought-provoking blog post on the potential creative benefits of large language model hallucinations. This survey delves into the notion that these hallucinations, often viewed as limitations, could actually fuel creativity and contribute to the application of LLMs. The post provides a comprehensive review of the taxonomy of hallucinations and their impact on LLM reliability, while also exploring historical examples and recent theories to substantiate this perspective. To learn more, follow this link to access the full post: https://bit.ly/42BpOHM I hope you find this exploration as insightful as I did. Best regards, [Your Name]

#NewPublicationAlert Research article in journal of Human-Computer Interaction by our lab member Anke Stoll together with Lena K. Wilms, Dr. Katharina Gerl and Marc Ziegele from the University of Düsseldorf. The article entitled "Technology acceptance and transparency demands for toxic language classification – interviews with moderators of public online discussion fora" examines technology acceptance of algorithm-based moderation software. The study focusses on the moderators’ characteristics and workplace requirements as preconditions for technology acceptance combining structured qualitative interviews with an enhanced cognitive walkthrough (ECW) of an algorithm-based moderation software. https://lnkd.in/eA2j4iPf

Tomorrow, my ISSCC paper, "C-Transformer: A 2.6-to-18.1yJ/Token Homogeneous DNN- Transformer/Spiking-Transformer Processor with Big-Little Network and Implicit Weight Generation for Large Language Models" will be presented. Welcome to my presentation!

"The situation is similar with Large Language Models. The task of establishing whether a given body of text contains hate speech is considerably easier than producing a hate speech-free text. Assessing the quality of a provided translation is easier than coming up with a translation from scratch (as anyone struggling to learn a new language will confirm!)." #largelanguagemodels #naturallanguageprocessing #retrievalaugmentedgeneration #evaluation

Check out this thought-provoking research paper discussing the potential of improving Large Language Models with emotional intelligence and the introduction of 'EmotionPrompt' technique. Learn more at https://ift.tt/PT3CsDB.

What do cosmic structures, financial returns, words usage in a language, choices of chess openings, number of views of web pages, neural activity and the number of citations of scientific papers have all in common? They all follow the same empirical behavior called 𝗭𝗶𝗽𝗳'𝘀 𝗹𝗮𝘄. - But what causes the emergence of such universal pattern? The answer to this crucial question has eluded researchers until now.  In our latest work with Davide Cugini, André Timpanaro and Giacomo Livan, we shed light on this century-old mystery. If you are intrigued, give a read to the rigorous proof we found that data drawn from any probability distribution, once ordered, become distributed at least locally with a Zipf's law, whose properties can be linked to the original "parent" probability. --> Link to the prepint: https://lnkd.in/dtwa_Bqi

Three types of knowledge conflicts can affect the reliability and effectiveness of LLMs in practical scenarios, namely, conflict within memory, conflict between contexts, and conflict within a particular context. These conflicts can substantially impact the performance of LLMs, particularly in real-life applications, where inaccurate data and noise are prevalent.

The Singularity Project 4o Expert in the developing General Theory of Intelligence, the theories of Artifactualism and Scientific Holism and the work of The Singularity Project https://lnkd.in/gAc3D9EA