🌻 E45: PDL - New Language for Prompting
PDL, New Agent Architecture, and Lots of AI News - A 5 minute read.
Large Language Models (LLMs) have taken the world by storm, enabling many previously challenging applications of AI. We are now familiar with LLM and RAG-based applications and their roles in solving real-world problems.
However, we have observed that LLM responses can vary depending on the style of the prompt. The answer to a question may change based on how the question is structured.
AI will change how we create. 83% of creators are using AI, much like the iPhone gave everyone access to cameras.
Hey, my name is Raahul. I have been trapped in the AI Industry for 10 years. As a developer, I thought I'd share my learning on LLM, recent AI innovations, and Agent, obviously, in simple language while I'm here.
That's why I write a newsletter called Musings On AI
It is a daily email that concisely tells readers about the LLM, AI Agents, AI business and culture in a conversational, witty way.
I own the 20K+ audience's attention multiple times in a week morning at 8.02 AM ET.It will be free always. This is today’s endtion, If you dont want the next editions, please feel free to UNSUBSCRIBE. No hard feelings :)
That’s why we invest considerable time in prompt engineering to determine the best techniques, such as zero-shot prompts like “Let’s think step-by-step,” chain-of-thought (CoT), ReAct, Graph-of-thoughts, among others. Several frameworks, such as DSPy, help automate this journey.
Probably this motivated the development of prompting frameworks, which mediate between LLMs and the external world.
To address this issue, this paper introduces the Prompt Declaration Language (PDL). PDL is a simple, declarative, data-oriented language based on YAML.
It integrates well with various LLM platforms and supports the development of interactive applications that utilize LLMs and tools. PDL simplifies the implementation of common use cases such as chatbots, RAG, or agents.
🌸 From The Agent Community
🌼 A couple of developers from JPMorgan have developed an agentic architecture called FISHNET (Financial Intelligence from Sub-querying, Harmonizing, Neural-Conditioning, Expert Swarming, and Task Planning). This architecture effectively handles highly complex analytical tasks for over 98,000 regulatory filings, which vary widely in terms of semantics, data hierarchy, and format.
FISHNET demonstrates remarkable performance in generating financial insights, achieving a 61.8% success rate, significantly surpassing the 5.0% success rate of routing methods and achieving 45.6% RAG R-Precision.
🌼 Amazon has introduced Knowledge-Graph Enhanced Language Agents (KGLA) to enhance the ranking capabilities in its recommendation systems.
Traditionally, recommendations have relied heavily on classical user-item embeddings.
A knowledge graph can be pivotal in this context. It links users and items with specific keywords, relationships, and even time preferences. Leveraging this connectivity, it effectively generates recommendations.
🌼Your AI Financial Advisor: Recommends stocks basis news.
https://github.com/ComposioHQ/composio/tree/master/python/examples/quickstarters?utm_source=twitter
🌼 You can now compare multiple models Google AI Studio, accessible via the “Compare” button in the top right of any prompt.
🌸Choice Cuts
🌼 Anthropic releases Claude 3.5 Haiku with 4x the price of Calude 3 Haiku.
🌼 Apple + ChatGPT access via Siri for free, with option for paid upgrade
🌼 The behavior of LLMs varies depending on their design, training, and use.
🌼 OpenAI’s o1 model leaked on Friday and it is wild.
🌼 DataChain is a modern Pythonic data-frame library designed for artificial intelligence. It is made to organize your unstructured data into datasets and wrangle it at scale on your local machine. Datachain does not abstract or hide the AI models and API calls, but helps to integrate them into the postmodern data stack. Code:
🌼 In 2149, life is meticulously dictated by software that tracks and analyzes every facet of existence, from emotional triggers to optimal daily choices, streamlining lives into a seamlessly optimized journey “on rails.” As data privacy became obsolete, society embraced the Big Merge, pooling personal insights into a vast cloud that forecasts everything from personal fulfillment to global events, ensuring maximum efficiency and satisfaction. Read it here:
🌸 Podcasts
There’s a lot more I could write about, but I figure very few people will read this far anyway. If you did, you’re amazing, and I appreciate you!
Love MusingsOnAI? Tell your friends!
If you have any comments or feedback, respond to this email!
Thanks for reading, Let’s explore the world together!
Raahul