ChatGPT's Custom Bots: A Leap Forward or a Stumble in AI Progress? It’s a Mixed Bag.
OpenAI's rollout of custom GPTs for ChatGPT Plus users seemed like a breakthrough in personalized AI. However, my own experiments with this technology revealed a different story—one of missed potential and the sobering realities of AI's current capabilities.
My journey began with an attempt to create a bot for coding a database-driven website. The confusion that ensued was more than a mere programming hiccup. ChatGPT, lost in translation, veered off into a tangent, eventually spewing NSFW content. This incident wasn't just about a bot's error; it was a glaring example of the AI's struggle to grasp context and nuances in user requests.
A notable success came when I incorporated Austin's CapMetro transit data into a GPT. The task: to map a specific bus route. While ChatGPT eventually delivered, the process was far from smooth. The bot's initial attempts to "chart" the route, instead of drawing a straightforward map, highlighted a critical gap: AI's inability to intuitively understand and execute tasks without extensive guidance.
My venture with a bot based on my "Code Like an Investor" blog was similarly eye-opening. The AI's advice frequently contradicted my blog's content, suggesting a disconnect in understanding and applying specific information sources. Even after uploading relevant materials, I had to constantly direct the bot to use them, rather than allowing it to rely on its broader dataset. This necessity for continual correction painted a picture of an AI still in its learning shoes, far from the mature, context-aware technology we envision.
领英推荐
These experiences culminated in a realization: ChatGPT, in its current incarnation as a web application, is a work in progress. It's like a child, full of potential but needing constant guidance and supervision. The need for extensive context-setting, the apparent memory limitations, and the lack of continuity in conversations signal that we're still at the early stages of AI's web application journey.
Moreover, in a world where AI is increasingly woven into the fabric of various software, the value proposition of a ChatGPT Plus subscription comes under scrutiny. If the AI in your existing suite of tools isn't meeting your needs, perhaps ChatGPT can fill that gap. But beyond such specific scenarios, its current form feels more experimental than essential.
Reflecting on my experiment with ChatGPT's custom bots, I'm left with mixed feelings. There's excitement for what AI can become, but also a tempered realism about its present limitations. As AI continues to evolve, we must approach it with a blend of optimism and critical evaluation, recognizing both its potential and its current state of adolescence in the vast landscape of technology.
Retired from IUP
1 年One immediate advantage I appreciate is the ability to curse out a bot without the fear of offense. ??
Helping businesses grow with innovative digital solutions
1 年Joe LeBlanc The trick with the GPT's is not to expect much. There are some really good examples. But like a recent post I saw, 95% of the GPT's being created are rubbish. If you createa custom GPT, your instructions you give it and how to work with your data must be more explicit. If you do this, then your results will be much more specific. As an example I wrote a custom GPT which resulted in this experiment. https://matthew.philogene.co.za/t/ai-30-day-image-experiment/ - My single prompt was: "Generate a new image for today 16:9 format" I had very specific instructions in my bot. I will follow up with more info on this soon, as I have been able to query the past 30 days of info with some interesting results...
My original notes: Is ChatGPT ready for the web? OpenAI recently released custom GPTs. This means that anyone with a ChatGPT Plus plan can use and create chatbots for specific purposes. I decided to give this a try. The results were interesting: * One of my first attempts at creating a bot involved me saying “Create a GPT that generates code for a database-driven website.” ChatGPT was confused by this and apparently thought that by mentioning “GPT,” I wanted to create a clone of ChatGPT itself. It started scatting away and eventually said some NSFW words. * On one of my more successful attempts, I uploaded public information about Austin’s CapMetro transit system into a GPT. I eventually got ChatGPT to create a map of a specific bus route in Austin. But it still took a lot of prompting and it was still trying to “chart” the route rather than just draw a map. * Another bot I tried to create was one based on my Code Like an Investor blog. It would up giving advice that is contrary to the advice on my blog. I found that I had to remind the bot of specific entries on the website, which it would go and look up individually. It does not seem to index or cache the results, so it is probably best to just upload this as an attachment.