From ChatGPT to Gemini to Copilot, which AI assistant chatbot is actually worth a monthly subscription?
When we last looked at the utility of AI assistants, it seemed like Microsoft’s Bing and Google’s Bard were the two chatbots actively going after everyday consumers, but multiple app and feature launches later, the market is a lot more crowded.
OpenAI gave ChatGPT something like an app store, Anthropic made Claude accessible to everyone, and Bing, rebranded as Copilot, has become the centerpiece of everything Microsoft does. That’s just a sliver of the changes to the landscape of artificial intelligence that have happened in the last year, but it’s a good indication of how interested companies are in converting research projects into products.
Still, whether the current collection of AI assistants will be useful to you is hugely dependent on how much text you deal with on a daily basis, the depth of information you search for online, and how comfortable you are double-checking just about everything. None of these assistants are free from errors, and they regularly remind you of that fact.
How useful is an AI assistant, really? With more options than ever, and many offering monthly or yearly subscriptions, knowing what to look for is hard, but Inverse is here to help. For this comparison, we looked at the things these assistants most readily do: answer questions, interact with and understand files and photos, and creatively produce images and text. There is no unified use case for these services — the companies creating them don’t seem capable of identifying one — but if any of those verbs pop up in your day-to-day life, here’s which AI assistants will be most useful to you.
07
Grok — xAI
Grok is not an AI assistant in the same caliber as any of the others on this list, for the very fact that it’s not really useful for anything in particular. Grok can answer questions and generate text (it created an oddly moralistic story about a haunted mine when I asked for a spooky story), but living inside of X and using X posts as a primary source for its responses ultimately produces some strange answers and opinions about things. Where it matters, it gave me similarly useful responses for, say, removing super glue from a wooden desk, but anything more complicated, like analyzing an image or a file, it was completely unable to do.
Its sense of humor — Grok seems to default to “Fun Mode” when you first use it — also left a lot to be desired. What usually ended up being funnier was how it tried to show me posts related to what I asked, regardless of what I asked or how useful they actually were. It should be said, X is not nearly the source of instructional information that Reddit is. Grok is not worth subscribing to X for, nor does it feel like an AI assistant that will be useful to anyone any time soon.
06
Meta AI — Meta
Meta’s AI assistant, Meta AI, is broadly comparable to what Copilot and Gemini are capable of. It can generate images, answer questions, and create text about general information, but its connections to web results are not entirely straightforward. I was never able to get Meta AI to answer questions about current events, but it could tell me about Italian restaurants near where I live. Meta’s AI has no way to interact with other files you might have on your computer or phone, but it can analyze and answer questions about a photo. In that way, it feels limited as a tool for conducting research or getting actual work done.
One thing working in its favor? It’s quick. Unless you’re asking Meta AI to create an image from scratch, its text responses are just as fast as ChatGPT, if not faster. The web information it pulls up doesn’t take longer than five seconds or so to display. My sticking point with Meta AI is that the easiest way to access it is currently in Meta’s social apps, like Messenger group chats, Instagram Explore page, search, or a direct conversation with Meta AI in WhatsApp. I understand why Meta is doing this, and see the unique experience bringing an AI into a group chat could offer, but it never feels particularly natural to me.
Access Meta AI on the web | iOS via Instagram, WhatsApp, Messenger | Android via Instagram, WhatsApp, Messenger
05
Copilot — Microsoft
Copilot presents similar frustrations to Gemini. Considering how much Microsoft knows about Windows users, whether they use Windows for work or their personal life, Copilot feels distinctly limited in what it can do. It’s not bad at digging up information, answering questions with current web results, and replying with a helpful tone I found pleasant and encouraging. But it also doesn’t really do anything else with outside information or files unless you’re willing to pay for an additional Microsoft 365 subscription. I was surprised at that, especially since Microsoft is relying on OpenAI for most of its functionality and ChatGPT can do those things.
The more annoying aspect, though, is that Copilot is seemingly fixated on offering shopping help. I was surprised at the questions I asked that still ended up producing links to store pages and carousels of prices. I just wanted to know the best way to clean my dog’s hair, I didn’t need to buy a dog brush or shampoo. If one of the goals of these AI assistants is to improve the cluttered experience of current search engines, this is in many ways replicating it. In terms of subscriptions, $20 per month gets you better image generation and the ability to create custom Copilots focused on a specific task, but none of it could be described as essential until it’s more clear what other things Microsoft wants Copilot to do. That’s something we should hear a lot more about around the company’s developer conference in late May.
04
Gemini — Google
Gemini and Gemini Advanced should absolutely be more useful than they currently are, considering how much AI research Google does and how interwoven Google services are in most people’s lives. Gemini is fine at answering questions and, much like ChatGPT, produces results that are easy to understand and read. Where it gets frustrating, however, is its ability to handle outside files. To conduct the same PDF analysis test as I did with ChatGPT and Claude, I had to upload those files to Google Drive and remember what each file was called to ask Gemini about them. That’s more annoying than it should be.
The implementation of Gemini on mobile devices feels similarly half-baked. On non-Android devices, you have to use a new tab in the Google app to start asking questions, and the app doesn’t always save your state when you leave and come back. Even on Pixel devices, switching to Gemini from Google Assistant prevents you from accessing all the features that Google’s original voice assistant had. Outside of interacting with Gemini directly, its use in every other Google service feels undercooked, and I’m not sure anything it produces would be satisfying to anyone but the middlest of middle managers. For a $20 per month subscription, Gemini Advanced does have a fascinating ability to remember and interact with images (I asked questions about printed recipes and other book pages that were really impressive) but Google needs to clarify and streamline how Gemini works before I can recommend it fully.
03
Perplexity and Arc Search
Perplexity and Arc Search aren’t technically AI assistants, nor are they made by the same company, but they handle a key task of these assistants remarkably well — answering questions and looking up information — so they’re going on the list. On good days, Perplexity comes close to replacing Google in terms of giving me useful information and links I actually want to click. It uses a mix of models to produce its responses, including GPT-4, Claude 3, and Perplexity’s own models, and I’ve found it to be a helpful starting place for finding information online (it’s almost Wikipedia-esque in terms of giving me links to follow into deeper research). The only obstacle to recommending a $20 per month Pro subscription? It still weirdly gets a lot of simple facts wrong, to the point where you should fact-check it when it answers simple science questions.
Presentation goes a long way when you’re dealing with AI products, and Arc Search is one of the better-looking options in terms of readability and playfulness (something underrated in products that almost exclusively produce plain text). It’s over-zealous in terms of the amount of content it generates to answer questions, but purely as a tool for gathering a bunch of links related to a topic, it’s handy and getting even more so now that there’s voice search too. Critically, it’s also completely free, provided you have an iPhone.
02
Claude — Anthropic
Anthropic is late to offering its language model Claude as a consumer-facing product, but it does get the closest to feeling trustworthy. That shouldn’t be surprising, Anthropic was started by ex-OpenAI talent, and its stated focus is to put safety first in the development of its AI products and research. That includes limiting its models’ ability to produce misinformation, amongst a whole host of other risks. Claude 3, which is technically a family of models that includes Haiku (for lightweight, fast responses), Sonnet (faster than Claude 2, best at knowledge retrieval), and Opus (same speed as Claude 2 but with much more detailed and informative responses), felt like the AI assistant I could actually use for work.
That’s because Claude is great at working with outside knowledge. I was able to upload journal articles written by a subject I’m interviewing soon, ask questions about what they wrote, and receive useful responses that guided my interview question writing. Claude feels slower to respond than ChatGPT, but its major disadvantage is that it doesn’t have access to the web and up-to-date information. Claude 3 Opus’ knowledge base only goes up to August 2023, which limits its ability as a search tool and general knowledge source. Anthropic recently released a Claude mobile app that is nearly as handy as ChatGPT’s, but whether on the web or mobile (iPhone only for now), I found Claude’s responses annoying to read because of how plainly they were formatted. Those annoyances aside, Claude feels the closest to a great research tool. You can pay $20 per month for access to its best responses and fewer limits on how often you use it, but the best part is that document analysis and basic question-asking are completely free.
01
ChatGPT — OpenAI
ChatGPT was the first proper generative AI assistant to market, and that head start is reflected in just about everything that’s good about the product today. While other companies like Google or Microsoft have more access to your personal and professional information, the flexibility of OpenAI’s ChatGPT ultimately makes it more useful. It can access information on the web; it can understand, summarize, and answer questions about a majority of files or photos you upload; and its creative output, whether text or images made with DALL-E 3, are far better first drafts than anything the other AIs are able to produce (even if they still need more work to be sufficient for human consumption, in my opinion).
On your phone, the mobile app OpenAI created for ChatGPT is far more user-friendly, too. It presents information in a legible way, with formatting that makes the piles of text it generates easier to digest on small screens. Talking to ChatGPT with just your voice is also natural, fast, and often more human-like than on-device voice assistants like Siri or Google Assistant. Getting a version of ChatGPT (using GPT-4) that offers the best version of these features, with the most up-to-date information, and the ability to access custom GPTs costs $20 per month, but it’s worth trying for a month to see if it works for you.
A Chat Box Doesn’t Work for Everything
We’re just at the start of what could be years of AI-driven and AI-focused products, so it’s natural, to a certain extent, for there to be a lot of unknowns around how this technology should be best used. Today, I can confidently say that text, text generation, and typed questions and answers, are not the be-all-end-all of interacting with a computer. In fact, trying to think about the things you want to do as a series of questions, answers, and text commands is really not natural at all.
The two things that could make these AI assistants feel more purposeful and helpful: doing things on their own (and do it in an accurate, risk-free way) and if everything didn’t have to be funneled back to a search box or chat. Where are the buttons? Where are the interfaces? Where are the useful voice commands? In the rush to create the feeling of a product that can do everything these AI assistants have become cumbersome to use. The useful future for this technology, unless it’s sequestered into hardware of its own, is likely deeper integrated into the apps and devices we already use. Until then, they feel a bit like square pegs being jammed into round holes.
