Qingsong Wen

🔬 Interesting insights from Microsoft Azure! We're thrilled to share Dr. Krysta M. Svore's recent post on defining logical qubits and criteria for resilient quantum computation. Understanding the path towards scaled quantum computing is crucial for the industry's progress and the development of a robust strategy to build a quantum-ready community. In June 2023, Microsoft Azure outlined the three implementation levels necessary for achieving utility-scale quantum computing: Level 1 Foundational, Level 2 Resilient, and Level 3 Scale. Currently, all quantum computing technologies are at Level 1, and while NISQ machines have been developed, they do not offer practical quantum advantage. 🚀 The next critical step towards Level 3 Scale is demonstrating resilient quantum computation on a logical qubit. Resilience, in this context, means showing that quantum error correction enhances non-trivial quantum computation. However, demonstrating resilience with just one logical qubit is not enough. It's essential to showcase the interaction between logical qubits and the entanglement they generate. Therefore, demonstrating two logical qubits performing an error-corrected computation that outperforms the same computation on physical qubits will mark a significant milestone in the field's history. 🌌 This advancement highlights the ongoing efforts to push the boundaries of quantum computing, paving the way for practical quantum advantage. Join us in the journey towards achieving scaled quantum computing! #Microsoft #MiddleMarket #Simulator #Supercomputer #MicrosoftAzure #QuantumComputing #QuantaStandard

3

Explore how to fine-tune BERT using the Hugging Face Transformers library. Discover how you can leverage pre-trained models for your own projects from Shaw Talebi's article. #LLM #MachineLearning

13

📢 Insight from Stanford HAI: Do Healthcare Algorithms Need to Be Generalizable? In a blog post by the Stanford Institute for Human-Centered Artificial Intelligence (HAI), Dr. Nigam Shah suggests that while generalization is key for AI reliability, it’s just as important to tailor models to the specific needs of hospitals and medical institutions. By considering local differences in how people interact with their healthcare providers, we can ensure that AI models are not just reliable but also effective in serving diverse communities. Want to know more? Dive into the full article here: 🔗 https://lnkd.in/dZe_SGDt #AI #Healthcare #MachineLearning #StanfordHAI #DigitalHealth #HealthTech #AIinHealthcare #DataScience #Innovation

7

Spatial Reasoning did not ‘emerge’ spontaneously in Large Language Models (LLMs) the way so many reasoning capabilities did. Humans have specialized, highly capable spatial reasoning capabilities that LLMs have not replicated. But every subsequent release of the major models — GPT, Claude, Gemini- promises better multimedia support, and all will accept and try to use uploaded graphics along with texts. #llms #chatgpt4 #gemini #llmops #aiml #ml https://lnkd.in/dFSWKqKs

2

For his debut TDS article, Jeremi Nuer presents a thoughtful walkthrough of a recent paper on model interpretability: "We want to have finer-tuned control over LLMs. Being able to detect when certain features are active, and tune how active they are in the middle of generating output, is an amazing tool to have in our toolbox."

18

We are happy that the paper "Accelerating Transformers with Spectrum-Preserving Token Merging" has been accepted at NeurIPS 2024, one of the leading conferences in ML and AI. The conference will be held in Vancouver, Canada. ================================== Here is the information from our recent paper: Title: Accelerating Transformers with Spectrum-Preserving Token Merging Abstract: This paper proposes a new token merging method, i.e., PIToMe, to enhance the efficiency of Vision Transformers (ViTs) by merging tokens to preserve crucial information. Unlike previous methods, PIToMe uses an energy score to prioritize and protect informative tokens while merging redundant ones. In this way, the computational and memory requirements are reduced by 40-60% with minimal impact on accuracy. Superior performance is achieved in multiple tasks such as image classification, image-text retrieval, and visual question answering, demonstrating its effectiveness and robustness compared to existing token merging techniques. Furthermore, it theoretically preserves the spectral properties of the original tokens, contributing to its efficiency and performance stability. Link: https://lnkd.in/gUnijrGh

8

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

Aligning Large Language Models with Diverse User Preferences Using Multifaceted System Messages: The JANUS Approach Quick read: https://lnkd.in/gQAsT95M Paper: https://lnkd.in/gkBAuWiC

19

Excited to share our new work: Automatic Curriculum Expert Iteration for Reliable LLM Reasoning! https://lnkd.in/grzSbyVF New approach to tackle hallucinations and "laziness" in AI models! Auto-CEI (Automatic Curriculum Expert Iteration) balances assertiveness and caution, helping LLMs better navigate reasoning tasks by confidently responding within their skillset and only deferring when needed. This method boosts accuracy and robustness in reasoning, math, and planning by refining AI's responses with a dynamic curriculum that rewards extended problem-solving Paper: https://lnkd.in/g9Gw_2wt Code: Coming soon Zirui Zhao Hanze Dong Amrita Saha Caiming Xiong

82

Optimizing large language model inference cost is arguably one of the most important factors in increasing the adoption of #GenerativeAI. Speculative decoding is an optimization technique for inference that makes educated guesses about future tokens while generating the current token, all within a single forward pass. 🔥 Learn more in “A Hitchhiker’s Guide to Speculative Decoding” from the PyTorch blog: https://lnkd.in/ekCBhQ4q

5

vGPU, MIG, & Time-Slicing! 3 ways to optimize the #GPU resources to meet the dynamic demands of computational tasks💻 Sameer Kulkarni explains & compares these methods in the blog post to help you select the right GPU optimization for your models👇 https://lnkd.in/gFQPX4b4

7

Understanding and mitigating hallucinations in Large Language Models (LLMs) is critical for reliable AI applications. In a blog by Rogger Luo, we delve into advanced techniques to tackle this issue. Our experiments show promising results in reducing hallucination rates, offering valuable insights for those working with LLMs. Vectara’s RAG-as-a-service platform already helps mitigate hallucinations effectively, and we are excited to see ongoing research further enhancing these capabilities. Read the full blog here: https://lnkd.in/gjZ4yreb

12

3 Comments

Fine-tuning LLMs with Reinforcement Learning from Human Feedback This blog post explores how Reinforcement Learning from Human Feedback (RLHF) can be used to fine-tune Large Language Models (LLMs) to align with human preferences. It delves into the concept of Proximal Policy Optimization (PPO) and how it is used in this context. The key takeaway is that RLHF offers a powerful framework to train LLMs that can produce human-aligned text, while still maintaining creativity. This paves the way for more natural and nuanced language generation in the future of AI and natural language understanding. Read the article: https://lnkd.in/d9EGf579 #AI #machinelearning #goML

8

🗳️ Will Trump Win the 2024 U.S. Presidential Election? What if Large Language Models (LLMs) could forecast the next election? While LLMs have proven effective in healthcare, legal analysis, and creative fields, can they accurately predict elections? In our latest research, we explore this possibility by applying LLMs to forecast the 2024 U.S. presidential election. Using a multi-step reasoning framework, we (Chenxiao Yu, Zhaotian Weng, Winston Li, and collaborators) integrate synthetic personas with real-world data from the ANES 2016 and 2020 studies, factoring in candidates' policies and backgrounds. This approach blends demographic, ideological, and time-sensitive data to assess how well LLMs handle dynamic political environments. We validated the framework with the 2016 and 2020 elections and generated predictions for 2024 within two weeks! ⚠️ Disclaimer: This study explores the capabilities of LLMs in political forecasting. Predictions are not definitive and do not reflect the authors' views. 📄 Curious about the winner? Check out our current draft: https://lnkd.in/gEvR25gj #AI #LLMs #ElectionPrediction #PoliticalForecasting #SyntheticData #MachineLearning #Research

82

4 Comments

We're pumped to welcome Anush Elangovan of AMD to the PyTorch Conference 2024 keynote stage! Explore the list of keynote speakers: https://hubs.la/Q02Nfvkm0. Register & join us September 18-19 in San Francisco! https://hubs.la/Q02NfwF30 #PyTorchConf #PyTorch #PyTorchFoundation #AI #MachineLearning #ML #DeepLearning #OpenSource #OpenSourceSoftware #OpenSourceDevelopment #OpenSourceCommunity #OSS #LinuxFoundation #events #linux #SoftwareSupplyChainNews

48

1 Comment

Hello All! It is great to be back with another article in my series “Neural Networks Intuitions - 18. Generative Pretrained Transformer(GPT) Series". The past 1.5-2 years have been very much dominated by Large Language Models(LLMs) ever since the release of GPT-3.5 and GPT-4, their fantastic ability to solve wide range of natural language tasks — right from question answering, summarization, visual QA and even code generation. And anyone who has followed this recent trend would know about the talks regarding the emergence of AGI which led to two school of thoughts — one set of people who truly believe that Language Models can lead us to AGI, while the other set sharply denies this statement by arguing that although there are a few signs of reasoning, there is no true extrapolation as such and all of it comes down to the training data(and its scale). The latter group's criticism is that unless there is visibility into the training data distribution and having a test set(without contaminating the test data with train set) along with a proper evaluation methodology, it is unfair to say that these LLMs can actually lead to AGI. Now in order to truly validate some of these closed source GPT models(such as GPT-4, Gemini, Claude etc), we need to be aware of the training data distribution — which is highly unlikely. But another way to understand if LLMs or more specifically instructed tuned LLMs can actually generalize to novel distribution or not is by actually digging deep into the evolution of these architectures/algorithms — more specifically the evolution of GPT models. In this article, I will be going over a series of GPT papers - GPT-1, GPT-2, GPT-3 and InstructGPT, and try to understand what makes these models truly generalizable or universal. Do read and comment your thoughts/opinions :) https://lnkd.in/di8_VpJA #neuralnetworks #largelanguagemodels #GPT #incontextlearning #nlp

24

1 Comment