🚨Is In-Context Learning Sufficient for Instruction Following in LLMs?🚨

In our new work, we study alignment of base models, including GPT-4-Base (!), via many-shot in-context learning. I.e., no fine-tuning whatsoever, just prompting - how far can we go?

Many people are

For Yann Lecun LLMs are already the past. Even „boring“.
I’d say, let us see where more scale in LLMs will get us to.

GNN-RAG

Combines the language understanding abilities of LLMs with the reasoning abilities of GNNs in a RAG style.

The GNN extracts useful and relevant graph information while the LLM takes the information and leverages its capabilities to perform question answering over

🚨 New LLM personalization/alignment paper 🚨

🤔 How can we obtain personalizable LLMs without explicitly re-training reward models/LLMs for each user?

✔ We introduce a new zero-shot alignment method to control LLM responses via the system message 🚀

Discover how to build a more efficient & powerful #app !

Alexander explores a few alternative graph data models that can enhance #RAG Applications and explains each model's unique benefits.

bit.ly/3VTPfCY

#LLms #Neo4j #graphdatabase

This paper surveys the recent literature on LLMs and financial applications, including sentiment analysis and forecasting. Tons of references are provided for further reading.

Link: arxiv.org/abs/2401.11641

Teach LLMs to internalize chain-of-thought (CoT) reasoning, without generating explicit intermediate steps, enabling implicit CoT reasoning during inference.

📌 Stepwise Internalization that successfully teaches LLMs to reason implicitly achieves high accuracy while maintaining

الـ WSJ سوو إختبار شمل خمسة من أفضل النماذج اللغوية الكبيرة (LLMs) في العالم، تم تقييم ChatGPT من OpenAI، وClaude من Anthropic، وCopilot من Microsoft، وGemini من Google، وPerplexity على مهام يومية متنوعة وهي كالتالي:

1.الصحة
2.الإستثمار
3.الطبخ
4.الكتابة العملية
5.الكتابة

'Navigating a career: tech, people, and LLMs'.
A panel with Monica Giambitto, Bozhidar Batsov, Rashmi Nagpal and Lucian Ghinda.

#rubyday #rubyday 24 #conference

Closing interview today with Sam Altman and Nick Thompson from Atlantic.

Leading one of the largest centralized LLMs has a lot of responsibility, I am always intrigued in his answers...

We all should be.

AI for Good 🇺🇳 #AIforGood

look at Taraxa Project

They had unveiled Trixy at $TARA's TinkerFest, showcasing its groundbreaking integration of blockchain technology and advanced AI.

Leveraging TARAXA's Echo network, Trixy fine-tunes LLMs for native social understanding, enabling predictive market

Stop paying $20 to $60 for ChatGPT and other LLMs!

You can get GPT-4, Claude, Gemini, and a lot more, including image generation.

Here is how you can get started for free:

With Conversation Genome we are building a decentralized Scale AI
​
Bittensor gives validators access to a subnet commodity based on TAO stake
​
SN 3️⃣3️⃣'s commodity is structured data: the critical ingredient for LLMs
​
Today, we enabled validators to process ANY text data thru SN33

ميسترال Mistral هي إجابة فرنسا على OpenAI. وكلها مفتوحة المصدر.

لقد ألقوا للتو أكبر هاكاثون للذكاء الاصطناعي في باريس على الإطلاق.
تقدم أكثر من 1000 مشارك لبناء ما هو ممكن باستخدام LLMs مفتوح المصدر.
انا شخصيا استخدم نموذجهم وسريع ومتقن حتى ان مايكروسوفت تعاقدت معهم لوضع

Here’s a great guide teaching you how to construct knowledge graphs using LLMs that adhere to a pre-defined schema - using purely local models (ollama, Hugging Face), and Neo4j as the graph store.

Possible with LlamaIndex 🦙! 🦙🕸️

1. Define the set of possible entities and

Excited to present our work “Large Language Models Can Self-Improve At Web Agent Tasks”. We show that synthetic data self-improvement boosts task completion by 31% on WebArena and introduce quality metrics for measuring autonomous agent workflows. #AI #MachineLearning #LLMs [1/n]

Mistral is France’s answer to OpenAI. And it’s all open source.

They just threw Paris’s largest ever AI hackathon. 1,000+ hackers applied to build what’s possible with open source LLMs.

Here are the finalists from the Mistral AI x Cerebral Valley hackathon in Paris (🧵):

“To people like me, LLMs are the past... they are kind of boring now.” Yann LeCun
➝ Spotify: open.spotify.com/episode/7GhUkp…
➝ Apple: podcasts.apple.com/fr/podcast/sca…
➝ YouTube: youtu.be/zfVsT1DqI9I?si…
#scalingtheory

PV-Tuning: SOTA 1-3 bit LLMs 🤯

Say what??
🔥 Optimal quality for target model size (Llama 2 70b with avg of 2 bits per param!)
🤗 Open Models! huggingface.co/collections/IS…
😢Requires more compute/memory during fine-tuning

Impressive work! Paper: arxiv.org/abs/2405.14852

LLMs are like boxing speed bags.

No matter how hard and long you hit them (write to them), they come back at you with a one-page (or more) response in a second.