?? Everyone, everywhere, all at once

First of all,?congratulations to Michelle Yeoh?for her Oscar victory and proving that style never goes out of fashion. Her film swept the board at the Oscars. Even bigger things will be happening with large language models.?

I’ve been trying to figure out how to think about what will happen when Microsoft (possibly announcing this Thursday) rolls out chat extensions and other generative AI tools into its office productivity suite. Google’s has already announced that it is bringing generative AI tools to its office producitvity applications.

OpenAI’s GPT-4 is now available, and?it does much more than its predecessors.?Many apps, like Quora and Stripe, have already put GPT-4 intro production. Anthropic, another leader in the field, rolled out Claude+, their new LLLM too. So understanding what hapens when these tools get used everywhere is really important.

ChatGPT reached 1 million, then 100 million users in record time. But Microsoft could beat that record?exponentially, by allowing its?1.2+ billion (and that was only 2015) users?to access GPT-4-like tools. We need to face the fact that this could have dramatic, widespread impacts. Let’s explore what those might be.

Microsoft has already tolled out?GPT-3.5 to Teams, its conferencing tool. Teams?had 280m users in January this year, although it’s unclear how many of them have signed up for the Premium access that features GPT-3.5. I’ve had four Teams meetings this month, and no one was using it.

Powering up a billion brains

Let’s imagine though that the chat-style interface, powering advanced sentence completion, improved grammar and rephrasing, text summarisation, concept extraction, and image and video generation, is plugged directly into Microsoft’s Office products. Accessible to 1.2bn humans.?

How can we possibly begin to imagine what kind of impact that might have?

There isn’t much scientific evidence on the effect of ChatGPT on the desk maestro yet, but?Noy & Zhang’s 2023?paper (which we profiled in?EV#412) gives us some clues.

The study by MIT researchers focused on mid-level professionals with university degrees (marketers, grant writers, consultants, data analysts, managers, etc) performing writing tasks. The median salary for these workers was high, in the $70-75k range.?

The results were staggering.?

Participants who used ChatGPT were much more productive, with time taken to complete tasks decreasing by 0.8 standard deviations. Or in the 40% range. A 27-minute task ran to about 17 minutes. The effect was significant.

When expert evaluators estimated the output quality, the results were also positive. Output quality improved by about 0.45 standard deviations. That is also impressive. An improvement of 0.45 standard deviations is about the same, in my estimate, as turning a median worker into a top quartile one.?

Most surprising, perhaps, was that the lowest performers saw the most significant performance boost using the chatbot. Given that the subjects’ occupations require a certain level of cognitive ability, I would suggest that widespread use of ChatGPT would be a great equaliser across the workforce.

Here's why...

Continue reading this essay here.