Have We Reached Peak AI Hype?
Recently, there's been a lot of negative press about Generative AI. Have we reached Peak AI Hype? Is it all downhill from here? We think not.
Halloween is just around the corner. So it's fitting that we've recently seen a lot of Generative AI FUD (Fear, Uncertainty, and Doubt).
Is AI overhyped? Have we reached peak AI? Let's review the recent headlines, then we'll share our thoughts on them.
The FUD
- CNBC: "Overhyped" generative AI will get a "cold shower" in 2024
- The Information: OpenAI Scraps New Arrakis Model
- The Guardian: AI chatbots could help plan bioweapon attacks
- VentureBeat: Uh-oh! Fine-tuning LLMs compromises their safety
GenAI: Not Overhyped
The Gartner Hype Cycle predicts that most new technologies go thru 5 stages of market adoption:
- Innovative Trigger
- Peak of Inflated Expectations
- Trough of Disillusionment
- Slope of Englightenment
- Plateau of Productivity
CNBC says CCS Insight, an analyst firm, believes that we're at the Peak of Inflated Expectations in the AI hype cycle. CCS predicts "the reality of the cost, risk and complexity involved “replaces the hype” surrounding the technology" in 2024.
Open AI now wants a valuation of $86 billion. Many might also consider this evidence of peak AI hype.
We tend to disagree. We think GenAI changes everything.
After CNBC's "Overhyped" article, Gartner stated that GenAI will be everywhere, and companies need to be GenAI-ready. "AI adoption should no longer be considered an IT initiative, but an enterprise initiative." Read the VentureBeat article.
We agree with Gartner.
But Some May Experience a Cold Shower
There's a lot of hype about GenAI right now. Will certain GenAI companies be able to fill their lofty valuations? Only time will tell. Some of them may (eventually) see a cold shower.
Many companies have decided to build their own Large Language Models (LLMs) or fine-tune their own LLMs. In 2024, these companies will realize that creating and fine-tuning LLMs requires a large amount of data, people, and GPUs. They're the ones who will experience the cold shower.
DIY GenAI is like many DIY projects: it seems fun at first, but it becomes chaotic in the end. Many enterprises should turn to Generative AI SaaS solutions to save both time and money.
Why Scrap Arrakis?
Open AI scrapping Arrakis isn't the end of the GenAI world. You have to think about why they scrapped Arrakis. We have 2 thoughts on this:
- GPT-4 is really, really good already.
- To beat GPT-4's response quality and speed, even Open AI needs to invest a lot of money. Perhaps the ROI of Arrakis wasn't really there.
In most cases, GPT-4 combined with Gleen AI meets or exceeds our customer's needs.
Wait -- My Chatbot Knows About Bioweapons?
Researchers were able to "jailbreak" an unnamed chatbot and LLM and get information about bioweapons.
But this article is primarily AI fearmongering. Yes -- it's possible to jailbreak an AI chatbots and get the chatbot/LLM to not observe content safety standards. Jailbreaking chatbots isn't new.
This article also doesn't differentiate between public chatbots – like ChatGPT and Bard – and custom, white-label chatbots that enterprises deploy to better serve their stakeholders.
A good white-label chatbot should constrain its responses only to knowledge provided. In other words, it shouldn't hallucinate.
The white-label chatbot you deploy should meet data security standards and prevent jailbrake attempts.
Fine-Tuning Compromises Content Safety Guidelines
This isn't a surprise. LLMs are complex machines. Turn one knob here, another one moves over there.
LLM fine-tuning may not always be a good idea, considering the expenses and potential dangers involved.
The Key Takeaways
Don't let the GenAI FUD scare you away from persuing GenAI. Remember:
- Becoming GenAI-ready is an enterprise-wide initiative.
- To avoid a cold shower, don't try to build your own GenAI solution from scratch. Partner with enterprise chatbot solutions like Gleen AI.
- Chatbots are critical to your GenAI strategy. They're your first layer of defense for response relevance, hallucination, and content safety.
- Chatbots are part of your infrastructure. They need to be secure, and they need to withstand jailbreak attempts.
Get started with Generative AI today. Create your own generative AI chatbot with Gleen AI now, or request a demo of Gleen AI.