Chan Jong (CJ) Na on LinkedIn: This is insane! I like the caveat tho. Same system can solve one problem… (2024)

𝐋𝐚𝐫𝐠𝐞 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐦𝐨𝐝𝐞𝐥𝐬 𝐚𝐫𝐞 𝐭𝐡𝐞 𝐩𝐫𝐨𝐛𝐚𝐛𝐥𝐲 𝐡𝐨𝐭𝐭𝐞𝐬𝐭 𝐭𝐨𝐩𝐢𝐜 𝐢𝐧 𝐀𝐈 𝐚𝐭 𝐭𝐡𝐞 𝐦𝐨𝐦𝐞𝐧𝐭. If you haven't seen our LLM content yet and would like to learn more about LLMs and related aspects, you can take a look at our LLM page here: https://lnkd.in/d_ng_tXnAs many customers are currently thinking about integrating LLM-based solutions into their infrastructure, we also invite you to a free 30-minute exchange with our Machine Learning Scientists to discuss potential LLM use cases. Simply send us an email to info@dida.do.#largelanguagemodels #llm

11

📢 Announcing LLM Engineering - a new live online course for aspiring AI Engineers!Join us for a 3 week cohort. Starting Thursday, November 2, to master the foundations of Large Language Models. You’ll learn:🤖 Transformer architectures📚 Unsupervised pretraining⚖️ Supervised fine-tuning↔️ Alignment, including instruction-tuning & RLHFWe’ll follow the history of the GPT models, paper-by-paper, and we’ll implement the core concepts we learn along the way in Google Colab!Apply in the comments.

45

12 Comments

Interesting!This paper from Google DeepMind shows that large language models (LLMs) can find patterns with 👉 self-generated reasoning.By only feeding minimal input, the system 👉 constructs its own CoT (Chain of thought), boosting performance by neglecting hand-crafted prompts.This could offer the potential to unlock more complex reasoning abilities within LLMs and to analyze the model’s reasoning patterns.Link in comments 👇

1 Comment

Hey there, fellow ML enthusiasts! Welcome to ML Math, the blog where we geek out over all things machine learning and dive into the fascinating world of math behind it all. Today, we’re going to tackle a super cool topic — prompting and how it can steer those sassy language models. You know how language models sometimes come up with responses that make you go, “Huh? What’s going on in there?” Well, prompting is the secret sauce that helps us guide these models to generate more accurate and desirable outputs.

9

1 Comment

Recently, while talking with a colleague, we were talking about LLMs, and their use in real problems. "What kind of problem?", "Which LLM is best for your problem?" I remembered this list of Benchmarks for GPT4 and Gemini-Ultra. Added to the same search query can give you an idea of how good a specific LLM is based on your problem.

13

Today, #TheSequence discusses one of the most fascinating papers published this year: Who's Happy Potter? https://lnkd.in/evqeyde5 #generativeAI #machineLearning #artificialIntelligence #LLMs

4

I completed Google's 'Introduction to Large Language Models' course, solidifying my foundational knowledge of machine learning algorithms and techniques. I look forward to leveraging this expertise in solving real-world challenges and contributing to the ever-evolving landscape of technology. #MachineLearning #GoogleCourses"

2

GPT-4o has arrived - and it points towards a future where natural human-language will be integrated into all devices. It can already match human response times in a conversation. This vid shows the ability for you to interrupt it, ask it to change tone, and react to user emotions. Show it a maths problem from a visual scan and it will solve it with understanding. Give it code, and it will view and analyse into plain speak what the code does.

1

Google DeepMind has released an amazing paper: "FunSearch: Making new discoveries in mathematical sciences using Large Language Models"FunSearch operates through the collaboration of a pre-trained LLM, dedicated to generating creative code-based solutions, and an automated "evaluator" tasked with preventing hallucinations. The process involves a continuous exchange between these two components, allowing initial solutions to undergo an iterative evolution, resulting in the generation of improved and correct code.I think is great because:1. Generality. FunSearch can be applied to any real-world computing challenge and already finds better solutions than the current ones, like for the Bin-Packing Problem (BPP).2. Interpretability: FunSearch is not a black-box: It doesn't just output a solution. It also returns the code and we can learn how the solution is obtained. Read the blog: https://lnkd.in/eAjA-eKR

46

1 Comment

PyTorch Internals!A comprehensive guide to how you can understand the PyTorch codebase & contribute to it.Here's the link:https://lnkd.in/gS3yVfGj↓If you are interested in AI Engineering, subscribe to my weekly newsletter ML Spring!Join 7k+ readers: https://lnkd.in/gRQX28B8

71

9 Comments