Apple Outclasses ChatGPT with Ferret
By the end of 2023, the most significant piece of AI news will likely be Apple’s decision to finally make its public debut.
The Apple model Ferret has been in the news lately, despite being released back in October.
Not only is it the first of its kind, but it also outperforms ChatGPT-Vision in a number of activities and provides fresh hints as to what Apple might have in store for the global market come 2024.
The Value of Correcting Mistakes
Hallucinations are the most common problem with a language model out of all of them.
Since Large Language Models (LLMs) are essentially stochastic word predictors, each response they provide is inherently unpredictable.
As a result, in order to prevent the model’s response from being entirely fictitious when employing these models, you must give it pertinent context in the prompt itself.
For example, you can integrate the article directly into the prompt so that the model uses it if you are writing a summary of a current news event that the model is unaware of.
This technique, known as “in-context learning,” gives the response’s pertinent context instantly.
This functions by allowing LLMs to use data that they have never seen before efficiently since they have seen so much data during training; this process is known as “zero-shot learning” or “few-shot learning” when a large number of instances are given.
