One of the most useful features of ChatGPT is the user of plug ins or tools. For exampl, the data interpreter allows you to upload data to chatgpt and have it write code that processes it to for example create graphs or analyse it. Another tool was Dall-E 3. You could ask it to connect to Dall-E 3 and generate an image. THis is what in Ai is referred to as agentic AI. Ai agents that can connect to other services or tools to perform a function for you. Previosuly, you had to select which tool you wanted to use. However, ChatGPT just rolled out a new interface change in which all the tools are available by default. This is extremely confusing for both the user and the AI. For example you ask it to do something and it has to decide which tool to use. recently, I gave it some data and asked it to generate a figure. It decided to use Dall-E 3 and tried to prompt it to generate an image. However, what I wanted is that it used the code interpret to read the data, write code, run it and give me the actual graphs for the data. Previously me giving the data and asking it to generate the figure while having selected the code interpreter was enough. However, in this case, I have to specifically say which tool I want it to use. That is fine for now, but as AI LLMs become connected to other external services, having access to all kinds of tools, from connecting your calendar to create events, to connecting to services like Uber to order rides for you, this can become messy and frustrating. Imagine wanting ChatGPT to give you ideas for what to it tonight and it deciding it order food from Uber Eats. this is a major challenge since the new release form OpenAI, GPT and Agent builders clearly points towards tgeir intention of maing GPT be connected to the real world, to read data and act on it. In another note I talk about how a crucial component of effective human-AI interaction is users having transparency and undersdtanding on the AI models. In that case it is useful to instead of trying to abstract as much complexity as possible from the user, giving them control and visibility. One of the most valuable features of ChatGPT is its ability to utilize plugins or tools. For instance, the data interpreter permits users to upload data into ChatGPT. The program can then write code that interprets and shapes this raw data into comprehensible graphs or analytics. Now, consider an instance of another tool, Dall-E 3. Once prompted, ChatGPT can connect to Dall-E 3 and effectively generate an image. This is a shining example of what is often referred to as "agentic AI" - these are AI entities that can efficiently link to other services or tools to perform a function. Previously, users had to specifically select the tool they desired ChatGPT to use. However, a new interface change in ChatGPT now makes all the tools accessible by default. While this may seem like an advantageous feature, it often leads to some confusion for both the user and the AI. For example, when asked to perform a task, ChatGPT must make a decision on which tool to utilize. Not too long ago, I fed some data into ChatGPT and instructed it to generate a figure. Unexpectedly, it chose to engage Dall-E 3 and attempted to produce an image. What I required, however, was for it to employ the code interpreter to assess the data, construct and execute the code, and provide me with corresponding graphs. Previously, just supplying the data and requesting a figure while I had the code interpreter selected sufficed. Now, however, I must clearly stipulate the tool I desire ChatGPT to use. With AI language models like ChatGPT becoming increasingly connected to external services, encompassing everything from arranging events on your calendar to partnering with apps like Uber to order rides, this can quickly become disorganized and frustrating. Picture asking ChatGPT for dining suggestions for the evening, only for it to suddenly order food from UberEats. This stands as a substantial challenge following the new releases from OpenAI, and GPT. Their clearly articulated intention is to enable GPT to connect to real-world data and react appropriately. My previous discussions have highlighted a key component of effective human-AI interaction; transparency and user comprehension of AI models are paramount. Instead of trying to remove as much complexity as possible from the user's experience, it may be more beneficial to enhance their control and visibility over the process.