Human versus #AI – episode 54 – GPTs, much more than next word prediction
🤖💡📝 The #GPTtemplates are much more than just a word prediction tool. Their richness lies in a four-step process that goes far beyond this initial prediction.
🚀 Pre-workout: During this phase, the #GPT explores language in depth, understanding grammar, context, nuances and subtleties. This is also the stage where the various styles and tones of writing are assimilated.
🎯 Supervised Fine Tuning: GPT is not limited to a single skill. Thanks to this stage, he can specialize in various fields – ranging from answering specific questions, to language translation, to writing texts.
⚖️ Reward Modeling and Reinforcement Learning: Here the GPT learns to optimize their responses and make decisions based on the feedback they receive. This is how it generates high quality responses.
🔔 Let's not forget that despite their sophistication, GPT models do not correct their errors like humans. They excel at optimizing responses, but lack an intrinsic ability to assess and correct.
💼 In short, the models #GPT are exceptional learning machines. Word prediction? Yes. But they go far beyond: understanding the context, adapting to new situations, generating high-quality responses. These versatile machines are transforming the future of#AI, making the future of the professional world brighter than ever.
#AI#ML#OpenAI#MachineLearning#GPT4#Artificial intelligence
🦾🦾🦾 The text above is generated by #ChatGPT
💪💪💪 Notes on the method:
1️⃣ Using the video by Andrej Karpathy, co-founder of Open AI at the Microsoft Build conference: https://lnkd.in/gAcaKXyY
2️⃣ Convert transcript to pdf
3️⃣ Query using the AskYourPDF plugin
4️⃣ LinkedIn Post Fitness
📊 Using AskYourPDF is not nearly as efficient as straight processing in ChatGPT, but it is mandatory for a document longer than 7000 words.
It was necessary to orient and correct the production several times to obtain a suitable result (Angle: GPTs are much more than tools that predict the next word). Alternative solution: use an application such as https://app.nextstart.ai/ developed by Flavien Besset et Ronan Broussier
For more efficiency, the final formatting was done in a new conversation without plugin
source: https://www.linkedin.com/feed/update/urn:li:activity:7068820758365368320/
To learn more: read our episode on emerging LLM capabilities such as GPT-4 and our article on the meaning of generative AI.