Mark Cox’s Post

The OpenAI Spring Update just finished. We are living in thrilling times as visions from sci-fi films materialize right before our eyes! Meanwhile, GPT-4O is already in the Playground, at least for me😀 I'm not sure if the new model is smarter, but at first glance, it's definitely very fast - spitting out tokens even faster than the 3.5-turbo! But hey, what happened to Maximum Tokens? As you can see, the Playground interface has also been updated; now, for example, you can upload images. What are your thoughts on the OpenAI's Spring Update? Have you got access to GPT-4O yet?

Just finished "Hands-On OpenAI API: Building a Real-World Solution" by Rodrigo Díaz Concha! Check it out: https://lnkd.in/eD2Jy9pe #openaiproducts, #technicalsolutiondevelopment. Really good example of a solving a real world problem using OpenAI APIs and deploying it in a GPT application.

I found out Georgian is not widely used and low-resource. OpenAI's Whisper doesn't support Georgian speech recognition. So, I fine-tuned the Wav2Vec2-BERT model specifically for the Georgian language and attained a Word Error Rate (WER) of 17.5 on "google/fleurs" dataset, using only a few hundred examples. This model achieved impressive results in the first attempt only. I've shared the code and model weights. check the comments for it. Also pushed model weights to Hugging Face hub. #deeplearning #machinelearning #asr #speechrecognition #naturallanguageprocessing #huggingface

Down the road there is not going to be significant difference between the LLM base models across multiple providers. What is going to be different is how seamlessly they become part of our day to day usage. Today's announcement from OpenAI shows the steps in that direction. https://lnkd.in/gzfbNfqP

If you haven't read through the chain of thought examples from OpenAI comparing o1-preview to gpt4o - you should stop what you're doing and read through the Cipher Example. Here's the prompt they tested on both: "oyfjdnisdr rtqwainr acxz mynzbhhx -> Think step by step Use the example above to decode: oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz" The chain of thought in o1 is mindblowing - While the model is too cost prohibitive to scale everywhere - there is clearly something fundamentally new and different about "Inference time RL." check out the results in the comments.

I just found this a valuable guide for using the new OpenAI model more efficiently (beginnersguide) #OpenAI #GPT4o #AIGuide

Fine tuning Gpt 4o mini… But there’s a catch. Watch the video to find out more. Here is the link for the different openai tiers: https://lnkd.in/ded6aRCN #openai #finetune #4omini

Today in Heartbeat: By connecting different components of #LLM functionalities, Chains are the secret sauce to building powerful LLM applications. Learn more about the different core chains offered by OpenAI and LangChain (and how to use them) in the newest article from Nhi Yen: #LLMOps #PromptEngineering https://bit.ly/49oCdRB

OpenAI just dropped MASSIVE news: (everything that you need to know) OpenAI’s Spring Update brought the HEAT. Here’s the scoop: 1. NEW GPT- 4o model - Faster. Cheaper. Better. - Available to all (free and API). - Improvement across 50 languages. - Reasons across voice, text, and vision. Capabilities: - Vision - Browse - Memory - GPT Store - Advanced Data Analysis 2. Desktop app - Vision = can see your screen. - Voice = you can talk to it. - Simple. Easy. Intuitive. 3. Crazy good voice updates - Chat in real time. - Chat through vision. - Picks up on emotion. - Generate voice in different tones. Crazy updates. Models improving. Interactions easier than ever. Updates rolling out in next weeks. Hats off OpenAI team! P.S. What update are you most excited about?

OpenAI just released the OpenAi o1 : - new capacities of reasoning on top of answering the question. As capable of a PhD Level for complex science problem (OpenAI test results). What will be the role and interaction with react agents based on LangGraph framework, for example ? - capable for complex reasoning for health, math and code complex problems - capable of explaining reasoning, which is good for confiance - scope is not as wild as GPT4o... today -> OpenAi announced it will be a new generation of LLM https://lnkd.in/eVcBUk-M