Paul Bratcher’s Post

I've been journaling this stuff privately, but I've decided to make it into a blog post as it might be useful to others and it might make for interesting discussion. Read more 👉 https://lttr.ai/AJBrr

We at Nested Knowledge just read a blog about the accessibility and portability of systematic review outputs, especially focused on Free Trials. We thought that the author brought up some very interesting questions, and we thought we’d provide answers for our tool! https://lnkd.in/e294WCnV Check out our most recent blog to learn more!

I just updated my blog on medium - Come check it out if you havent already! https://lnkd.in/garQbpVS medium @icarusabiding

Why are data scientists and ML people so excited by the DSPy library? If you come from an academic background or from "traditional" NLP, you probably find the whole LLM app building process quite manual and non-principled. Everything before was about running experiments, making sure you wouldn't overfit on your training data, and doing model parameter optimisation. Now with LLMs a large part of the model is encoded in a prompt, but prompt engineering is more black art than science. DSPy is a framework that tries to automate that by automatically picking the best examples to add to your prompt. We went ahead and tested this with Haystack, and it works! My colleague Stefano Fiorucci created a notebook where we show how to optimise a prompt in a RAG pipeline to make your LLM less verbose while keeping the accuracy high: https://lnkd.in/d9NGnMwR I wrote a post about that same notebook in medium with a short introduction to DSPy: https://lnkd.in/dTs8mMNs

💡 𝗖𝗿𝗼𝗽 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗺𝗼𝗱𝗲𝗹 𝘄𝗶𝘁𝗵 𝟵𝟵% 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆. 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗲𝗱 𝗯𝘆 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗺𝗼𝗱𝗲𝗹𝘀 𝗰𝗼𝗺𝗽𝗮𝗿𝗶𝘀𝗼𝗻 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆 𝗮𝗻𝗱 𝗳𝗶𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝘁𝗵𝗲 𝗯𝗲𝘀𝘁 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆 𝘆𝗶𝗲𝗹𝗱𝗶𝗻𝗴 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺. 👉 Checkout the Jupyter notebook here : https://lnkd.in/druWyTkH Now I am feeling enough confident by implementing more and more projects, thanks to my mentor Sameer Nigam of AI SOCIETY #machinelearning #ai #ailearning #classification #randomforestclassifier

There is no doubt that ChatGPT or Gemini(Formerly Bard) have led the explosion of Generative AI. However when you look at the data that they were trained on and how they might have been trained on – it raises the question if it was done the right way. The right way being subjective based on the viewpoint your are looking at https://lnkd.in/guVNPgNR #indiana #indianatech #indianapolis #indianapolisai #artificialintelligence #generativeai #chatgpt #bard #gemini

What separates SpecGPT from your average chat agent? Check out this podcast with our founder Hugh Seaton, CDT, CCS, or refer to the TLDR below: - Has been trained on multiple spec types, including Masterformat, Uniformat, and UFGS. - No page limits, upload your entire spec and each response will use the full context to get you the insight you need. - Context sourcing for each response, including specific spec references as backup provides the transparency you need to use the tool confidently. - Has been fine-tuned by industry experts.

Dive into key approaches to optimize the response LLMs for your application. 🔑

"How do I learn to write prompts for Large Language Models (LLM)?" During a meeting at the #AWSSummitBerlin today, I came across a very nice tool provided by Anthropic: https://lnkd.in/etWvysme They have essentially written a tool that uses their LLMs to help you write prompts efficiently! While you can of course use this tool to essentially write the prompts for you, I think it is also a great way to learn the essentials of writing good prompts yourself and then use that knowledge to effectively implement advanced use cases that require more than prompt engineering. Shout-out to Daniel Rosenthal & the engineers at Anthropic for making this nifty tool available! #AWS #GenAI #AI #ML #LLM #Anthropic

I've been journaling this stuff privately, but I've decided to make it into a blog post as it might be useful to others and it might make for interesting discussion. Read more 👉 https://lttr.ai/AJAz0