Eoin O'Mahony

A substantial piece of work from Apple led by Iman Mirzadeh, keivan alizadeh vahid, Oncel Tuzel, Samy Bengio, Mehrdad Farajtabar and Hooman Shahrokhi supports the thesis that transformer-based LLMs do not reason--they perform pattern-matching. The team from  also provides a great benchmarking tool--GSM-Symbolic for assessing mathematical and formal reasoning. This supports my anecdotal findings when I evaluated o1 https://lnkd.in/ec72fecb as well as my thoughts on the incompleteness of our assessments of AGI https://lnkd.in/ehBT3HB3

16

2 Comments

I am excited to share that our paper, "Silver Lining in the Fake News Cloud: Can Large Language Models Help Detect Misinformation?" has been accepted for publication in IEEE Transactions on Artificial Intelligence (IEEE TAI)! Abstract: In the age of advanced generative AI, distinguishing truth from deception is a growing societal challenge. Our research explores how Large Language Models (LLMs) can aid in detecting misinformation. We use a comprehensive approach, evaluating multiple LLMs with few and zero-shot prompts and incorporating sentiment and emotion annotations to understand their impact on misinformation detection. Additionally, we deliberately distort news using ChatGPT to examine various parameters like abstractness and concreteness, enhancing our framework for identifying authentic, human-generated, and LLM-induced misinformation. Impact Statement: As LLMs evolve, they play a dual role as both defenders and perpetrators of misinformation. This research delves into the conflicting nature of LLMs, evaluating their effectiveness across diverse datasets and prompting techniques. By corrupting authentic news and human-generated fake news with ChatGPT, we quantify parameters like concreteness and abstractness to improve misinformation detection tools. Our findings aim to enhance current frameworks and contribute to the responsible development of AI, fostering a more reliable and resilient information ecosystem in the face of misinformation. Key Contributions: LLM Evaluation: Assessed multiple LLMs' effectiveness in detecting rumor content across six datasets. Prompting Techniques: Investigated few-shot and zero-shot prompting and examined sensitivity to hyperparameters. Sentiment & Emotion: Explored the impact of sentiment and emotion on LLMs' rumor detection capabilities. Dual Corruption Approach: Intentionally distorting both authentic news and human-generated misinformation using ChatGPT. We analyzed parameters such as concreteness and named entity density to gain deeper insights into the impact of these distortions. Explore our research and access the code on GitHub: https://lnkd.in/gbY-9QcT A huge thank you to my co-authors Raghvendra Kumar, Bhargav Goddu, Adam Jatowt, and all those who supported this journey. Here's to advancing the fight against misinformation with cutting-edge AI technologies! #Research #IEEE #ArtificialIntelligence #MisinformationDetection #LargeLanguageModels #AI #FakeNews #IEEE_TAI

249

11 Comments

Hallucination is Inevitable in LLMs An important new paper out that shows that even with perfect data and an idealized architecture “LLMs will always hallucinate”. I suspect that this will over time be linked to Gödel’s Incompleteness Theorem [1] and Turing’s Halting Problem [2]. Every technology has trade-offs. We love LLMs but they need to be used in the right way. [1] Good pop science depiction in Oppenheimer (Gödel was Einstein’s walking companion) or also see Rebecca Goldstein’s book Incompleteness [2] Turing is depicted in The Imitation Game and also the progenitor of the famed Turing Test.

36

Predictions based on Machine Learning and Statistics often rely on maximal likelihood, which can trap LLMs in biases inherited from the majority opinions of training data. This problem has been substantially addressed through conditional statistics, allowing for mitigation of biases and preventing the cancellation of minority perspectives. https://lnkd.in/gMXKqwyB

7

Released a new paper on how to use LLMs to generate unit tests for PLC code. Getting valid test inputs worked well in a few examples but the assertions were often hallucinated. Needs more research, but already can save developers testing efforts. Virendra Ashiwal, Soumyadip Bandyopadhyay, Chandrika K R, and me from ABB conducted prompt engineering and created a test generator and test execution framework for IEC 61131-3 ST code based on open-source (MATIEC, OpenPLC, GCC). Used several open-source function blocks (OSCAT) for experimentation. Our paper was just accepted at IEEE ETFA 2024: https://2024.ieee-etfa.org 8-page report on arXiv: https://lnkd.in/dn3dZguQ Github repo with the code: https://lnkd.in/dxb7JhBC

165

16 Comments

I'm thrilled to share that our latest paper, "Systematic Review of Software Engineering Uses of Multi-Criteria Decision-Making Methods: Trends, Bibliographic Analysis, Challenges, Recommendations, and Future Directions," has been published in Applied Soft Computing (Q1, IF 8.7)! 🔗 Check out the paper here: https://lnkd.in/dyB5svSq This comprehensive review delves into the fascinating field of multi-criteria decision-making (MCDM) and its impactful applications in software engineering. We cover the latest trends, provide an in-depth bibliographic analysis, and discuss the challenges and future directions in this dynamic area. For those interested in how MCDM theory has contributed to advancing software engineering, this paper is a must-read! Happy reading! 📚💡 #Research #SoftwareEngineering #MCDM #DecisionMaking #AppliedSoftComputing #AcademicPublication #Engineering #Innovation #FutureDirections

18

2 Comments

Thanks Ashish Patel 🇮🇳 for the below ❓ What Is the Technical Debt of Large Language Models (hashtag #LLMs) and How Does It Affect Us❓ As Large Language Models (LLMs) become central to AI, understanding their technical debt is crucial. This debt impacts maintainability, scalability, and effectiveness, posing significant challenges. Efficient management is vital for sustainable AI development and deployment. Key areas of technical debt associated with LLMs: 1. Model Complexity: 🔸 Large Parameter Count: Billions of parameters make training and deployment complex and costly. 🔸 Inference Costs: High computational requirements for inference lead to increased operational costs and latency issues. 2. Data Management: 🔸 Data Quality: Poor quality or biased training data affect performance and introduce biases. 🔸 Data Versioning: Managing different dataset versions is challenging, impacting reproducibility. 3. Training and Fine-Tuning: 🔸 Resource Intensity: Training LLMs consumes substantial computational resources, leading to high costs and environmental impact. 🔸 Hyperparameter Tuning: This process is resource-intensive and time-consuming. 4. Deployment and Scalability: 🔸 Deployment Complexity: Requires robust, scalable infrastructure for efficient production deployment. 🔸 Serving Infrastructure: Ensuring low-latency, high-availability serving of LLMs needs advanced setups like distributed systems. 5. Model Maintenance: 🔸 Continuous Monitoring: Detecting issues like model drift and performance degradation requires constant vigilance. 🔸 Updating Models: Regular updates or retraining with new data are complex and resource-intensive. 6. Security and Compliance: 🔸 Data Privacy: Compliance with data privacy regulations and preventing data leaks is crucial. 🔸 Adversarial Attacks: Ensuring robustness against attacks is an ongoing challenge. 7. Interpretability and Explainability: 🔸 Opaque Decision-Making: LLMs are often "black boxes," making their predictions hard to explain. 🔸 Regulatory Requirements: Providing explanations for model decisions can be difficult but necessary for some applications. 8. Ethical Considerations: 🔸 Bias and Fairness: Addressing biases and ensuring fairness for all user groups is critical. 🔸 Responsible AI: Preventing misuse or harmful applications of LLMs adds to the technical debt. 9. Legacy Code and Dependencies: 🔸 Outdated Libraries: Reliance on outdated libraries can lead to technical debt. 🔸 Compatibility Issues: Ensuring compatibility with evolving systems and tools is challenging. 10. in the comment 👇 Addressing these areas of technical debt requires a proactive approach: regular reviews, testing, monitoring, and training. This ensures the long-term sustainability and success of LLM projects.

11

New GAI Textbook Announcement (The textbook is being used by Stanford CS372: AI for precision medicine and psychological disorders) The past two years have been a significant period in the field of AI, during which we have observed substantial strides in generative AI technologies. Having dedicated over 5,000 hours to research and analysis, I have compiled this book to share a collection of findings and hypotheses regarding Large Language Models (LLMs)—detailing their operational mechanisms, and strategies to mitigate biases and reduce instances of hallucination. The central premise is to enhance context awareness and define human objectives with greater precision. By doing so, LLMs can more effectively mirror human linguistic behaviors, utilizing context-specific linguistic features to fulfill intended goals. SocraSynth: Socratic Synthesis with Multiple Large Language Models. https://lnkd.in/giC4474p

25

Reduce hallucinations! Nicely linked to yesterday's paper which also tackled hallucination using self-consciousness. Prereq-Tune uses multiple LoRAs, one for knowledge and one for skills to achieve similar goals. By adding knowledge in frozen LoRA form first this means that during the skill building SFT stage the model doesn't try to 'fill in the gaps' and therefore hallucinate. This also means knowledge can be added in a modular fashion depending on the task without affecting the core skills of the model. https://lnkd.in/eSdX2Dnv

1

Jorge Murillo presented our paper titled "CDN-Shifter: Leveraging Spatial Workload Shifting to Decarbonize Content Delivery Networks" at the ACM Symposium on Cloud Computing (ACM SoCC) earlier this week. The paper presents a spatial workload scheduling algorithm called CDN-Shifter for greening CDNs. Read it here: https://lnkd.in/eSxCU-Qh Walid Hanafy David Irwin Ramesh Sitaraman #cloudcomputing #decarbornization

37

Good detailed overview of Retrieval-Augmented Generation (RAG), one of leading techniques to improve accuracy of LLMs. When developing a production-grade LLM-based system, it's essential to understand the advantages and limitations of different techniques (RAG, prompt engineering, fine-tuning etc), across training and serving/inference phases, to improve LLMs for specific applications (see my previous post https://lnkd.in/giuJdJJt) A balanced approach to LLM applications that takes into account accuracy, latency, and cost is crucial for successful productization at scale. #artificialintelligence #largelanguagemodels #GenerativeAI #airevolution

10

We are thrilled to announce an upcoming seminar by Dr. Hedieh Sajedi, Associate Professor at the University of Tehran. Title: “Spectrum of Deep Learning Algorithms in Drug Discovery” Abstract: The journey of drug discovery is complex and costly, often spanning over a decade. However, advancements in AI, especially deep learning, are revolutionizing this field. This seminar will delve into how deep learning algorithms are transforming drug discovery, offering innovative solutions to accelerate the development of new therapeutics. Dr. Sajedi will explore the spectrum of these algorithms, addressing the challenges and opportunities they present, including data quality, computational demands, and model interpretability. Date & Time: September 20, 16:00 Location: YSU, AI lab

6

OpenCoder: An Open-Source Code LLM Redefining Model Efficiency 🌟 The OpenCoder team has launched a game-changer in the world of code language models. Unlike many LLMs that focus on sheer size, OpenCoder brings something unique to the table: a data-focused training approach. By putting quality before volume, OpenCoder’s 1.5B and 8B models rival some of the biggest players, like DeepSeekCoder and QwenCoder, with only a fraction of their token count. The core strength of OpenCoder lies in its meticulous data processing pipeline and two-stage fine-tuning. With a training dataset composed of 2.5 trillion tokens, 90% raw code and 10% code-related web data, OpenCoder refines every aspect of its corpus. Advanced filtering and deduplication strategies, applied to 607 programming languages, optimize model efficiency and precision, ensuring high-quality data drives its training. Key Features: - Complete Training Transparency: Open-source datasets and training pipelines enable reproducibility and further research. - Instruction Fine-Tuning: Two-stage tuning sharpens both core CS concepts and real-world coding skills, prepping OpenCoder for production-level code generation. - Available on Ollama: Accessibility in deployment means more avenues for experimentation and improvement. OpenCoder's data-driven strategy showcases the potential of a leaner approach. By focusing on quality data rather than quantity alone, OpenCoder achieves file-level deduplication that outperforms traditional repo-level methods, making it both efficient and robust. 🚀 Read more here: https://lnkd.in/eEnG3yJA #OpenSource #MachineLearning #CodeLLM #AI #ModelEfficiency #DataQuality #Innovation #CodeGeneration #OpenCoder #TechNews #AIResearch #Programming #LargeLanguageModels #ArtificialIntelligence

1

#RAG DOES NOT REDUCE HALLUCINATIONS IN #LLMS: Think I've called this out a few times, but the validation is nice - https://lnkd.in/eAi6zxW2. The article makes two well established points: 1) "Hallucination is a problem of reasoning and not relevance! Any amount of relevant text fed through RAG to a machine will retain the original perplexity and entropy in the system to hallucinate independent of the text"; and 2) "The initial challenge arises from the retrieval phase’s reliance on semantic similarity, which is quantified by inner product spaces or cosine similarities. This metric, while capturing semantic closeness, does not differentiate between factually accurate and inaccurate information." This should be clear to everyone, but the industry still shills flimsy RAG scaffolding anyway (shovels to gold miners).

104

34 Comments

One of the hidden gems I found in github! Check it out!

4

Here's another blog about the neural tangent kernel that I wrote for Borealis AI: https://lnkd.in/eK6_CMeH This is a super-interesting area but the literature is really hard to read. I've done my best to make it as simple as possible. This article presents expressions for the evolution of the residuals, loss and parameters for neural networks with a least squares loss and analyzes these expressions to gain insights into trainability and convergence speed. We further write expressions for the evolution of predictions and even their uncertainties over time (neural tangent kernel Gaussian processes). Finally, we show that it is possible to make predictions from an infinitely wide neural network without ever training it explicitly.

128

A new asynchronous SGD dropped: Mind Flayer SGD... 🎤 We study the problem of minimizing the expectation of smooth nonconvex functions with the help of several parallel workers whose role is to compute stochastic gradients. In particular, we focus on the challenging situation where the workers' compute times are arbitrarily heterogeneous and random. In the simpler regime characterized by arbitrarily heterogeneous but deterministic compute times, Tyurin and Richtárik (NeurIPS 2023) recently designed the first theoretically optimal asynchronous SGD method, called Rennala SGD, in terms of a novel complexity notion called time complexity. The starting point of our work is the observation that Rennala SGD can have arbitrarily bad performance in the presence of random compute times -- a setting it was not designed to handle. To advance our understanding of stochastic optimization in this challenging regime, we propose a new asynchronous SGD method, for which we coin the name MindFlayer SGD. Our theory and empirical results demonstrate the superiority of MindFlayer SGD over existing baselines, including Rennala SGD, in cases when the noise is heavy tailed.

28

1 Comment

This is a good one! As we start to talk about AGIs as becoming "real," not just something in the far-off-distant-future (like 3-6 years), we need to start being a bit more specific - issues, architecture, how they'll solve problems that really don't fit in well with the current transformer-based approach. Check out this YouTube from the talk that Lee Goldberg and I gave at the recent #TrentonComputerFestival on AGI! #agi #artificialgeneralintelligence #ai #artificialintelligence #generativeai #llms #transformers #aieducation

5