all images in this post created with Midjourney AI
AIs, like humans, work best when you appreciate their differences.
People often ask me for advice on which AI to use: ChatGPT or Claude? Perplexity, Bing Copilot or Gemini?
But just as there is no one perfect colleague, there’s no one perfect AI.
When you experiment with using different AIs for different purposes, you will almost inevitably find yourself reflecting on what each AI does well, where it’s frustrating, and how you need to adjust the way you work to make the most of each AI’s particular strengths and limitations.
That’s why AI collaboration is a terrific way to build your awareness of neurovariety—that is, your appreciation for the functional differences in how different workers think, perceive and communicate. Get better at recognizing the value of different AIs, and you’re building your ability to recognize the value of different colleagues, too.
Today I’m sharing my take on the most commonly used general-purpose AIs, and which AIs are the best at which tasks.
ChatGPT: The erratic pleaser
ChatGPT is the enthusiastic young employee who says yes to everything. “Read twenty reports and get you my summary by end of day? No problem!” But when that summary lands on your desk, you’ll see your colleague barely skimmed the reports, or maybe ignored a bunch altogether, because their enthusiasm in no way reflected their actual capacity.
That becomes evident if you load GPT up with a ton of background material (for example, by adding a whole bunch of knowledge files to a custom GPT, or carry on an extended conversation. GPT will accept as many documents as you give it, and carry on a conversation forever, but at a certain point it is just tossing a ton of that material out the car window, because there’s only so much it can hold in its robot brain.
That’s why it’s up to you to keep your conversation scaled to what GPT can actually digest and retain (about 300 pages worth of material and/or conversation), and start a new chat once you hit something like that limit. (I’m curious to see whether GPT’s brand-new model o1 will change this equation.)
It’s great to have that yes-to-everything colleague when you need volume rather than discernment, which is why ChatGPT’s 4o model is the workhorse I use to organize my own information—but not for writing style or for gathering facts. ChatGPT is useful for help with things like….
- Summarize these 3 PDFs by turning them into a table that compares different solutions based on the following criteria
- Help me convert this knitting pattern to a different weight of yarn
- I’m going to ramble a bit on the subject of this new project, and then I want you to organize my thoughts into an outline (via ChatGPT’s voice interface)
Note that I only trust ChatGPT because I’m paying for the Team plan; if you’re using the free or Plus plan, ChatGPT may roll your chats and uploads into the training data for its future models.
Claude: Quality with boundaries
Next to the people-pleasing GPT, Claude looks like the mid-career colleague who sets clear boundaries and is a little prickly if you ask for more than they can take on: “There’s no way I can read twenty reports for you, so pick the five that are most important.”
Try to add more than a handful of documents to a Claude project or conversation, and Claude will refuse to accept the uploads or tell you that you’ve run out of room with your chat.
Just like colleagues who only commit to what they can truly deliver, Claude’s latest model (3.5 Sonnet) usually produces much better results than what I get from ChatGPT. Claude may not accept the same volume of input, but whatever it does accept will actually get processed effectively. I use Claude for tasks that require judgment or an appreciation for writing style, like….
- Read this contract for me and identify which clauses I should look at more closely, and tell me why they might be of concern
- Please review this draft newsletter for clarity, style and grammar/spelling/typos, and return three bulleted lists: one list of errors that I MUST fix, one list of stylistic improvements to consider, and one bulleted summary of the strengths and weaknesses of this draft
- Take this interview transcript and return a list of quotes that I can turn into LinkedIn or Instagram posts
If I need to summarize a few academic research papers, write a first draft that reflects a few of my past articles , or pick potential quotes from a series of interview transcripts, Claude is the colleague I trust to do the job.
Perplexity: Informative…with a catch
When there’s one person in the office who always has the latest organizational intel or industry update, it’s hard to resist using that person as a resource or sounding board, even if you don’t really trust them.
That’s how I feel about working with Perplexity. On the one hand, I appreciate the specificity of its answers, and the fact that it footnotes its sources. On the other hand, the platform has landed in hot water for the sketchy way it handles published content, so I would never ask Perplexity a question that depends on sharing any of my own ideas or writing.
In practice, that means I treat Perplexity as a kind of upgraded search engine—something that can quickly answer a factual question—rather than as a thinking partner. Some of the tasks and questions I’ve recently given Perplexity include…
- Describe the structure of trade unions in the Japanese auto sector
- Are there any proposals or advocates for changing US overseas voting rules so that state of last residency is not the determinant of where you vote?
- A customer who has always worn Nike Pegasus runners would like to try Hokas. Which model should she consider?
Gemini and Bing Copilot: Free, at a price
When there’s a new intern or junior hire in the office, you might give them a task to see what they can handle. That’s what I’ve done with Google’s Gemini and Microsoft’s Bing Copilot, repeatedly…but I have yet to find a reason to use either one of them in place of ChatGPT or Claude.
Sure, Gemini formats its results nicely…and once in a while, when I need to turn a document into a table or do some other task that doesn’t require a whole lot of subtlety or judgment, I sometimes turn to Gemini. And if I need to get a factual answer to a question I don’t want to share with Perplexity, I might turn to Bing Copilot (which cites its sources, but often chokes at providing the level of detail I want.)
Some tasks where Gemini did just fine, even if it didn’t wow me, include:
- Best practices for setting up a home wifi network with 10+ computers and many (50+) home automation devices like light switches, Amazon Alexas and curtain controllers?
- Can you find examples of companies or employers who have made text to speech a standard part of how they work?
- What are the best materials and practices for sealing off an area with plastic when someone in the house has Covid?
Just like that inexperienced intern, the main thing these AIs have going for them is price: Unless you’re paying for a better AI that offers you unlimited chat, you may find it useful to use one or both of these as your own daily workhorse.
Remember: AIs aren’t people. (And then forget it.)
A lot of smart people argue that it’s very dangerous to anthropomorphize AIs like I have in this newsletter. After all, it’s a total misrepresentation of how AI actually works, and can lead users to put too much trust in AI.
But entering into the illusion that I’m talking to a person is a big part of why AI has been such powerful creative accelerant. When I’m in an extended chat with Claude or ChatGPT (and especially, via ChatGPT’s voice interface) , there’s something about playing let’s pretend that brings me back to the freewheeling creative energy I had as a kid, and which I gradually learned to repress as a teen and adult.
This process of learning to repress our uninhibited, authentic selves is what autistics refer to as “masking”—that is, engaging in the behaviors that allow neurodivergent people to pass as neurotypical (non-neurodivergent). That effort of repressing and masking is much harder for some of us, but all of us pay a price for working in a workplace that asks us to put away childish things, repress our exuberance or stop playing pretend.
Recent Comments