It’s becoming clearer that generative AI chatbots vary widely in their ability to provide useful information. For those engaged in creating – inventors and content industry workers, as well as students and employees – there are more options to fit the right tool to the task, and more are on the way.
Five of the top large learning models (LLMs) in nine topic areas, were tested by Dow Jones’ Wall Street Journal with intriguing results. Some are excellent in a few areas but none are great in most subjects. For example, while ChatGPT is very good, it is less effective with writing, both work-related and creative. The cost to LLM’s to become really proficient in a particular subject area can be formidable, even for the largest companies backed by the most capital.
For now, at least, no one bot can be relied upon for all of your tasks or information needs.
The WSJ tested five of the most popular bots: ChatGPT, Anthropic’s Claude, Copilot, leverages OpenAI’s technology and integrates with services like Bing and Microsoft 365, Google’s Gemini and Perplexity in nine subjects:
Health, Finance, Cooking, Work Writing, Creative Writing, Summarization, Current Events, Coding and Speed. Some of the results were counter-intuitive. ChatGPT was first in only three categories, health, cooking and speed. The summary is below.
Ryan Morrison tests chatbots for a living. Below are the a Tom’s Guide reporter’s current conclusions. In addition to the big five the WSJ reviewed, he includes among the best such less well known names xAI Grok, Llama 3 (best for open source) and MetaAI.
- Best overall: Claude.
- Best for Live Data: Google Gemini.
- Most Creative: Microsoft Copilot.
- Best for Research: Perplexity.
- Most personal: Inflection Pi.
- Best for Social: xAI Grok.
- Best for open source: Llama 3.
- Most fun: MetaAI.
Steering with CoPilot
Sabrina Ortiz, editor of ZDNET was a “Copilot die hard” until ChatGPT added five new new features, some with GPT-4o. For image generation OpenAI-owned DALL-E, she says, is not the only option.
Best overall image generator, reports Ortiz, is Microsoft Designer’s Image Creator (which I would imagine has at least some common elements given MSFT $13 billion investment in OpenAI). Two sides of the same coin?
“Like DALL-E 3,” she says, “it combines accuracy, speed, and cost-effectiveness and can generate high-quality images in seconds. Unlike DALL-E 3, Microsoft’s tool is entirely free.”
The Best Tool for the Job
Depending on your vocation or avocation there are now options for the best GenAI tools. Some are pricier than others; some rely more on unlicensed content than others; some are more reliable. Although all of the bots may look pretty much the same on the surface, it is becoming clear that it is unwise to to put your faith one LLM for all tasks.
It’s not just the data they use but how the data gets processed and presented.
Therein may lie an opportunity for creators using GenAI tools to help generate IP protectable output, such as inventions, screen plays and images. Not unlike a carpenter or a plumber creators need to think about the right tool for the particular job. (Carpenter or plumber? As much as things change, they remain the same.)
Robot Collaborator
It’s tempting to invite a robot collaborator into your orbit, especially when it costs little or nothing, saves time and money, and generates responses in milliseconds.
Creators should be mindful, however, that bot owners may eventually wish to declare as their property the product you formulate or refine using their tools, or, for that matter, the unique prompts you employ to get there.
Remember: Your prompts will likely be used to train the bot of your choice for more relevant responses, a potential boost for your competitors.
Image source: Dow Jones
ipcloseup.com 