It’s time to break the ChatGPT habit

OpenAI’s ChatGPT was the first seriously popular generative AI tool, but it was probably not the best. It’s time to widen your AI horizons.

Shutterstock

ChatGPT is the first artificial intelligence (AI) brand to go mainstream and it was the fastest-growing tech product ever. The brand returns more than three million results on Google News Search. Late-night TV talk show hosts mention ChatGPT by name — and their audiences know what they’re talking about.

In business and tech circles, as among the general public, ChatGPT is synonymous with Large Language Model (LLM)-based chatbots. But it's time to stop obsessing over ChatGPT and start discovering the world of powerful alternatives in this new world.

First, let’s be clear about what we’re talking about.

LLMs, and the tools built on top of them

An LLM is a kind of AI system typically trained on petabytes of text data. It looks for patterns in all this data to predict the next word or sentence when queried, resulting in human-like interaction.

A chatbot is a tool that accesses an LLM. ChatGPT is a chatbot. The free version accesses an LLM called GPT-3.5 and the paid version GPT-4.0.

Those are just two LLMs, or two versions of the same LLM. But there are many comparable models out there. In fact, the number is astonishingly high.

Stanford scientists recently published their work taxonomizing LLMs. They identified — wait for it — 15,821 currently available LLMs. And to help us understand and contextualize them, the researchers built a resource called Constellation, which is what they call an “atlas” of LLMs for visualizing them.

Moving beyong ChatGPT

LLM chatbots are fundamentally multipurpose. But that doesn’t mean that you should use one tool like ChatGPT for everything. Besides, a controversial new report from Stanford and UC Berkeley suggests that for some uses, ChatGPT is actually getting “dumber.” (By the way, that links to a PDF. If you don’t want to read through a long, academic PDF, you can load the PDF into an LLM site like Scholar Turbo and just ask the PDF questions.)

One of the newest LLM offerings comes from Microsoft and Meta; it's called LLaMA 2. (The companies claim LLaMA 2 is open source, but critics say it’s not. Still, it’s far more “open” than LLMs from Google or OpenAI.) There are many ways to try LLaMA 2.

You can try Llama 2 via a demo hosted by the venture firm Andreessen Horowitz. You can download the code from Meta or the AI sharing platform Hugging Face, access it through Microsoft Azure or try it on Amazon SageMaker JumpStart. Or you can check out a unique resource that combines LLaMA 2 and Perplexity.ai to give answers to queries.

Beyond that, there are many more LLMs coming soon. Even Apple has reportedly built one on top of the Google machine learning framework Jax, which runs on Google Cloud. It’s codenamed "Ajax” and unofficially referred to by some internal engineers as “AppleGPT.” Employees use it for product prototyping work and other uses.

(Apple famously banned employees from using ChatGPT out of concern that they might introduce Apple trade secrets into its dataset.)

Like smartphones, it’s apps that will change the world

One of the best ways for non-developers to use LLMs is through sites, apps, browser extensions, and services that access them.

Tech founder, angel investor, product designer and Product Hunt super-user, Chris Messina, told me on the This Week in Google podcast recently that the overwhelming majority of apps on Product Hunt are now based on AI — north of 80%. Just a year ago, AI-based apps on Product Hunt were rare.

With new tools coming online every day, and existing tools getting powerful updates, it’s a good idea to stay on the lookout for new products via curated search tools like Futurepedia and AI Top Tools.

For example, for search engine replacement work, I recommend a site called Phind; it's an LLM-based search engine that gives you both the LLM-derived result and the search engine links it used for the data. It’s one of the many that is far superior to ChatGPT for current knowledge. In general, LLM-based chatbots are vastly inferior to search engines, and it’s best to combine search and AI.

For help writing emails or birthday cards, Anthropic’s Claude tends to write better prose. (Still, take it from a professional writer: No LLM-based chatbot is especially good at writing. More on that below.)

And given how quickly these tools evolve, it's important to remember that the best chatbots, browser plug-ins, websites, and apps for specific individual purposes next month, or even next week, might be completely different.

How to deal with errors and hallucinations

You can’t really rely on LLM-based chatbots for accurate information. And you can’t predict whether they’ll give you a result that’s incomplete, misdirected, totally erroneous, or perfect.

One nugget of advice for avoiding the pitfalls of errors and hallucinations is to get a second opinion — and a third and a fourth. It’s easy to compare by loading a bookmarks folder on your browser with chatbots that use different LLMs. For example, you can load a folder with Phind, ChatGPT, Bard and Claude. Type in a prompt, copy and paste it into each of the chatbots, then compare the result. This is a very fast way to both make sure you have factual information, and you can comparison shop for the best result.

In my experience, if you get three or four of these services to agree with each other, it’s usually reliable information.

Another alternative is AI Playground, which lets you choose different models and try them side-by-side.

I warned just a few years ago that we shouldn’t let AI write for us, lest our ability to write (and by extension, to think) atrophy. Now, I have a more nuanced recommendation. AI can erode our ability to write if we just put in prompts, then copy and paste the result for use as our own communication. Or, we can use AI as a writing partner to improve our ability to write and think.

The AI world is so varied and complex right now that the best advice I can give is to experiment like a mad scientist. Try as many new tools as you can. Experiment with integrating AI into your work. Run creative queries. And use AI as a brainstorm partner and sounding board.

The world of AI is changing incredibly fast — faster than any other category of technology in our lifetimes. And the promise and peril of the technology is overwhelming. All of us in the tech and business worlds need to approach it with a sense of adventure, learning, and experimentation. And don’t just stick to ChatGPT out of habit.