Sandeep Pani’s Post

Playing with Claude 3.5 Sonnet, very positive impressions. Feels more precise and straight-to-the-point, than GPT-4o

A few of you have asked and shown curiosity with the newly release GPT4o (Omni) and wanted to know how to access the same. So here is a single step way to access the same. 1. Go to poe.com and select GPT4o or type poe.com/GPT-4o

New technique: Copy and pasting the response from Claude to GPT4 then the response fro GPT4 to Claude and so on asking what the pros and cons of their thinking are

Errors crumble, functions align, In the end, our models shine. Debugging done, we rest our eyes, Until the next challenge arrives. Instead of writing poems with GPT-4o, use it for debugging? DinoAI is your #gpt4o powered co-pilot. Keen on trying? Get started for free: https://lnkd.in/dqUqsGYS #generativeAI #analyticsengineering #dataanalytics #dbt

Ever get SQL or dbt bugs™? If you don't, you're *t̶r̶u̶l̶y̶ ̶a̶ ̶m̶a̶s̶t̶e̶r̶ ̶o̶f̶ ̶t̶h̶e̶ ̶a̶n̶a̶l̶y̶t̶i̶c̶s̶ ̶e̶n̶g̶i̶n̶e̶e̶r̶i̶n̶g̶ ̶c̶r̶a̶f̶t̶. *Lying. For the rest of you, check out this hands-on demonstration on debugging using paradime.io's new "/fix" command!

Compared to gpt-4-vision-preview, gpt-4o has: 2x lower pricing 2x the speed 5x higher rate limits state of the art visual understanding, with an MMMU eval score of 69.1% vs. gpt-4-vision-preview’s 63.1%

We are now live on 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗛𝘂𝗻𝘁! Trickle is where you can intuitively manage your screenshots as they pile up. Extract insights from screenshots using the reasoning abilities of GPT-4. Turn your screenshots into “visual memos” that are both readable and searchable. https://buff.ly/3RC3OZZ

Chapter 7: 👀 Vision API Just as I was writing this, GPT-4o dropped... 🤖 It turns out it's even better than GPT-4V at vision! (and faster + cheaper)

⭐ Check out our tutorial on deploying the quantized GPTQ, 4-bit version of the CodeLlama-Python-34B model with vLLM on Inferless. 🚀 Experience blazing-fast performance with an average lowest latency of 3.51 sec and an average token generation of 58.40/sec and cold start time of 21.8 sec 🔍 Our experiments also included HuggingFace Transformer Pipeline, AutoGPTQ, and Text Generation Inference. Check the results below. Dive deeper into our comprehensive guide for all the insights: https://lnkd.in/gkzRPsAy

Little preview of some upcoming refactoring & new features in the Atomic Agents library (https://lnkd.in/ePeB4K7J) . Building a fully orchestrated multi-agent flow for any tool in less than a single page of code, with the ability to configure and/or extend each building block individually - built in a way that it even works with GPT-3.5, reducing cost by a factor of 10 vs GPT-4o Of course there will be use cases where you want to customize and use GPT-4o, but for the majority of stuff like online research, using heavy constraints and well-built tools and agent execution flows greatly reduces cost and effort while maximizing performance #AgenticAI #ArtificalIntelligence #LargeLanguageModels #OpenAI