| ChatGPT

ChatGPT's release has consequences beyond the technology itself.

Let's look at what it can be used for, the benefit for Microsoft and what's next for the tech sector.

Some thinking traps

Before diving into what people are suggesting ChatGPT be used for, some caveats:

There is a lot of survivorship bias. For every "oh wow" article, there are countless unreported "so what's".  Don't get me wrong - it is impressive to ask a machine to explain something or tell a joke. But - would you pay money for it today, and how much? How many hours did it save you, what can you do that you could not do before? What service do you not need to pay anymore?

Everything you say is monitored. OpenAI's terms explicitly give it the right to monitor and incorporate your input and your work. If work would not allow your ChatGPT conversation to be passed through Google Translate, or to be stored in your private Drop Box account, you better not use ChatGPT.  Amazon has found its confidential material information in OpenAI's outputs. Microsoft itself advises employees not to input confidential information.

It is unclear which capabilities are native to ChatGPT, and which have been added by armies of low-paid workers. Its programming prowess seems to be added by humans. The vast number of people working/training/labeling/patching it are trade secrets by a company interested in hyping up its product. ChatGPT's answers seem to acquire a lawyer-ish shine as time progresses, especially those about itself or political leaders. This is not specific to ChatGPT, but it should give pause to those who think it is easy to "retrain GPT on our data". 

"Let's use ChatGPT to"...

Some of these traps lead to overexcited proposals, before we come to the money makers:

Sell it to our clients! ChatGPT (and GPT in general) are not specialized tools. Unless you want to be a reseller, they need to be trained - the "P" in GPT. Training costs millions of dollars, takes time, blows through your company's ESG goals and is done by machine learning specialists (who will look at your data and your use case and usually tell you why they're not a good fit). It does not work by "giving it your data" and, whoosh, you have a magic wand. If it were easy, someone would already have done it. The real value lies in integrating GPT into your already good products and services.

Do our work for us! Depending on your line of work, it is unknown whether clients will accept work done by AI or pay standard rates for it. How much time will you spend error checking? Who assumes, insures mistakes? You were good in services, now you dabble in AI - does that open you up to more tech-savvy competition?

Using it for client work without their consent is risky too. First AI detection tools are available now. Similar to post-quantum cryptography, a cat and mouse game will mean your client will use tomorrow's tools to unmask the AI-generated content you charged human rates for today. CNet tried using AI (not ChatGPT) to generate content, and paused it after readers called it a bait-and-switch.

Improve my products. This is the main question, and few people are asking it at the moment - few but Microsoft and its big competitors.

What will Microsoft make with ChatGPT? Money, to begin with

Microsoft is a business and Satya Nadella needs to recoup his investment into OpenAI. It's already happening:

Selling access via Azure: Microsoft has been selling "Managed GPT-3" since 2021 as Azure OpenAI. It is being expanded to include GPT-3.5, including DALL-E and (soon) ChatGPT. Just don't expect it to do your work for you, out of the box.

Taking money from OpenAI: None of your experimentation is free. Each one costs OpenAI an outrageous couple of Microsoft's funding cents. Since they host exclusively on Azure, the money goes back right where it came from.

Integrating it into GitHub: GitHub is a service that helps programmers store their code. Microsoft acquired it and is building AI-based code completion and -generation functionality into it, called "co-pilot". That functionality alone adds $10 per user per month.

Integrating it into M365: Microsoft is said to be looking at integrating it into office. "Teams Premium" adds summaries written by GPT and $10 per user per month to your bill.

Integrating it into Bing: The most audacious move and likely what jolted Google (here is a GIF). I have cost questions: a ChatGPT search is estimated to be 1000x more expensive than a Google search. You also can't place advertisements there, which is at odds with Microsoft's goal of doubling its ad business to $20b. The Google Assistant faces the same constraints and it was likely the reason Alexa's teams got decimated. I also have legal questions: How will they deal with copyright violations, false information, the EU's right-to-be-forgotten, takedown requests and other issues?

Note how most of these forms are not pure ChatGPT, but modified versions of it or the technology it is built on. What your business does with AI matters, not AI for AI's sake.

Who is winning?

Globally, Microsoft says the top 5 labs are Chinese. OpenAI's huge PR coup has shifted the focus inwards: "We thought it was going to be China pushing the U.S., but looks like it’s start-ups." says a former Twitter AI leader.

Between Western companies, it's hard to say who is winning. AI people talk between themselves and move between labs, companies and universities. No company is ever more than 2-6 months ahead. (Remember when a Google AI was pronounced sentient?). Let's hope the aggressive productization does not disrupt that collaboration.

Advantages in AI typically don't last: Siri and the Google Assistant have not become much smarter. Nobody heard from IBM's Watson again. Microsoft shut down Cortana. ChatGPT may not evolve, but be supplanted by something even better, by someone else entirely.

Finally: The biggest winner is surely the field of AI research. It needs successes to attract new people - and tangible, cool products like ChatGPT are as important as more abstract innovations, even if their primary function is to inspire.

I'd like to leave you with "what will be important", as the noise about AI will increase in the next 1-2 months. The competitors are about to stir. Their products will be revealing, but even better to look for the "why" and how they will reposition themselves. Is anyone truly gaining an edge? Do stocks go up? Do companies with lots of money change in response to that, or will the news die down?

Also: Try it out yourself! ChatGPT is amazing, and perhaps your opinion is different than mine and it is instantly useful to you. Just remember that you are dealing with another data hungry company. Also, there are hundreds of startups like it, building all sorts of AI infused applications. Is there anything worth paying for? Drop me a line, I'd love to know.

And finally, thank you for reading this excessive post. 

Join 2100+ people who read these tips weekly. Subscribe now!