Sekėjai

Ieškoti šiame dienoraštyje

2025 m. vasario 4 d., antradienis

How to Get the Most Out of AI Assistants --- Popular work bots like ChatGPT, Claude and now DeepSeek help with project planning, web knowledge -- and even ordering flowers


"I keep waiting for my team to buy me a "WORLD'S BEST BOSS" mug. Then I remember they're bots. Workplace brown-nosing isn't one of their many skills.

The two AI co-workers on my org chart are OpenAI's ChatGPT and Anthropic's Claude. During the past few months, they've taken on some of my work . . . so I can do even more work. And now I am auditioning a third assistant, DeepSeek.

They're not just rewriting emails or summarizing meetings. These guys are building spreadsheets, prepping research, creating calendars and, yes, even ordering flowers for my spouse.

I pay $20 a month for Claude and ChatGPT. Why both? Because we're living in Turbulent AI Times where one week's best AI assistant is the next week's also-ran. Case in point: DeepSeek's recent surprise debut. Fortunately, that's free. I've also tested Google's Gemini, Meta AI and Microsoft Copilot but, to paraphrase the great Shania Twain, they don't impress me as much.

Choosing the best AI assistant for your work isn't only about these ever smarter models, but also the tools and features that help you get things done.

You will judge an AI not about how well it can do your job, but how many tasks you can offload to it.

"Every job is a bundle of tasks," says Erik Brynjolfsson, a Stanford University economist and founder of the AI-at-work consulting company Workhelix. "When you analyze jobs at that level, you can really make headway as to whether technology can help."

What tasks you can outsource to these assistants depend on your job, your workflow and, most importantly, the AI's capabilities. Yep, it's a lot like hiring -- you want the candidate with the right skills.

Deep thinker

I asked Claude to organize a list of contacts into a spreadsheet -- then it immediately followed up with, "I can also help create email templates for reaching out to these contacts." The perfect Lumon employee: fully focused, efficient and cheerful. (I see you "Severance" fans.)

That's a main reason Claude has become my go-to. Its writing is superior, and it not only gets the job done but also explains what it did and suggests next tasks, all with a bubbly personality.

The latest buzz is all about "reasoning" models, which break down queries into steps and "think through" their answers. DeepSeek's DeepThink R1 shows its entire thought process. When I asked if a hot dog was a sandwich, it spent 28 seconds walking me through its thinking -- complete with an analysis of USDA guidelines.

OpenAI's o1 also takes time to think but doesn't show all its work. Anthropic's CEO Dario Amodei told me last week his company doesn't see reasoning as a separate feature -- Claude Sonnet 3.5 already has similar capabilities.

Detail master

ChatGPT gets me. It really gets me. Its memory feature keeps track of details about you, learning your preferences to tailor responses. Go to Settings, then Personalization, and you can enable it to build a little dossier on you.

Mine includes: "Joanna is writing a book about AI in her life for a year" (true) and "Joanna's back hurts" (not true -- today). It makes responses feel more personal but you can delete individual lines, turn it off and even wipe it completely.

ChatGPT has even learned my habits. If I paste in text, it just knows I want it copy-edited, no prompt needed.

For Claude and DeepSeek, memory is more . . . goldfish-like. They remember details during a single chat or project, but forget everything when you start fresh. Amodei says Claude's memory will improve. DeepSeek didn't respond to my requests for comment.

Memory also does mean data collection, which is why I am careful not to put any sensitive information into these tools. With DeepSeek, specifically, I have deeper concerns given its Chinese ownership and lax privacy policies. For what it's worth, I have been testing it via Perplexity, which runs the model on U.S. servers that don't have ties to China.

Project manager

As I embark on my AI book adventure, I've hired a human research assistant. But Claude has already handled about 85% of the grunt work using its Projects feature.

I uploaded all my book-related documents (the pitch, outlines, scattered notes) into a project, basically a little data container. Now Claude can work with them whenever I need something.

At one point, I needed a master spreadsheet of all the companies and people mentioned across my documents, with fields to track my progress. Claude pulled the names and compiled them into a nicely formatted sheet. Now, I open the project and ask Claude what I should be working on next.

ChatGPT recently added a Project feature, but when it comes to managing my book tasks, I prefer Claude's personality traits.

Research pro

But without real-time web access, Claude is oblivious to current events. For all it knows, David Hasselhoff could be president, and we could all be commuting in Jetsons-style flying cars.

Meanwhile, ChatGPT and DeepSeek have web-browsing modes. Toggle them on, and they'll pull in real-time information, complete with links to sources. For folks I'm interviewing in the next few weeks, ChatGPT and DeepSeek pulled together up-to-date bios, with links to recent work.

Amodei says Claude's web integration is coming "very soon" and will be different from the competitors.

Good communicator

I have a new habit. When I'm in the car or out walking, I tell ChatGPT my ideas, and it jots them down.

With Advanced Voice Mode, ChatGPT responds in a natural, conversational way. It can't take action in this mode (so no sending emails yet), but I can ask later, "Hey, what was that genius idea I had this morning?" and it'll remind me. It can also export notes as a text file so I can add them to a project in Claude.

Microsoft Copilot, Meta AI and Google Gemini Live all have interactive voice modes. DeepSeek and Claude don't have an equivalent.

Self-starter

Where is this all heading? AI agents. The next bots will take real-world actions on our behalf. While we're not quite at the "manage my entire inbox" stage, OpenAI's new Operator tool (part of its $200-a-month Pro plan) offers a glimpse.

I tested Operator by asking ChatGPT: "Order my spouse some flowers from Bouqs.com. Find me something with oranges and yellows and avoid lilies." I watched, fascinated, as it navigated around the site in a virtual Chrome browser. When it first suggested a $200 party arrangement, I redirected it and it quickly found an alternative I liked.

It was pretty slow and had missteps, and I still had to handle payment details myself. I also didn't love typing my personal information into a remote browser. But I was wowed by how it navigated the web. Besides, nothing says "I love you" quite like my robot ordered these for you.

Claude offers similar capabilities through its Computer Use mode, but that requires more technical setup.

So for those keeping track: Claude is my go-to for project planning, clear office and document tasks and it's got a great personality. ChatGPT picks up the slack with real-time web knowledge, a friendly voice and more.

DeepSeek is smart but, so far, lacks the features to get ahead at the office. Maybe now's a good time to buy me that mug.” [1]

1.  How to Get the Most Out of AI Assistants --- Popular work bots like ChatGPT, Claude and now DeepSeek help with project planning, web knowledge -- and even ordering flowers. Stern, Joanna. 
Wall Street Journal, Eastern edition; New York, N.Y.. 04 Feb 2025: A14.  

 

Komentarų nėra: