Steve Ballmer was an underrated CEO

I was at Microsoft for the last couple years of Ballmer and the first few years of Nadella. He definitely did change the company and I remember at the time feeling that he handled the change really well, but from where I sat he spent the first part of his tenure evolving Ballmer's final push to move focus from Windows to developers. Everything Microsoft did prior to LLMs was to bring developers over, from VS Code to GitHub to WSL.

Now the company seems fully baked I to LLMs with everything they do chasing that. It would even make sense if the developer push was driven in part by the need to build up training sets for the eventual LLM work, though I really have a hard time believing that Microsoft was so well ahead of the game that they started grooming developers to provide data more than a decade ago.

Them along absolutely everyone else. ChatGPT was an iPhone moment.

It’s cute that we now have image and video gen AI. Also we have now Turing test passing chat bots (Id say). But although they are very impressive, and I know lots of people who use them for various tasks, I haven’t seen a “killer app” yet.

For iPhone the killer app was making calls. It was the best phone you could get. Then it had apps. It was undeniably better.

LLMs are good at a lot of things, but they don’t seem to excel any particular task - yet. I’m not sure they are a revolution yet.

I’d say they’re more of Macintosh moment. A hugely useful technology no doubt - but useful for what exactly? For Mac it was desktop publishing.

The killer app on the iPhone was not "making calls" — I suspect instead it was Safari, the other 1st party apps, the touch screen and the slick integration of all of that to make it a no-brainer device that even my mom and dad could use (they were approaching their 70's when the iPhone debuted).

Your analogy that ChatGPT (or LLMs generally) are more akin to the Mac feels close to the mark to me. Your comment about the Mac's killer app, desktop publishing, suggests that LLM's killer app will follow, just hasn't arrived yet.

The analogy is a little shaky though since, some would argue, it was the laser printer (plus the Mac) that kicked off desktop publishing.

For LLMs, the "killer app", for me, is already here. And there's two of them right now.

The first is the chatbot (like chatGPT or Pi or Claude). Having someone who you can just ask for any kind of information, from book recommendations to hypothetical space travel situations to advice about birthday gifts, and to get answers that are better than what I'd get from 90% of real humans, is huge to me.

The second one is the coding assistant, in my code copilot. It has made me at least twice, if not thrice as productive as I was before.

Nice analogy. My sense of things is that the iPhone was a win for all of its users. While ChatGPT may make some/many of its users more productive (see Ethan Mollick's work), the driving force behind 'AI! AI! AI!' in the corporate world is an executive hope that complacent AI can replace expensive people. That's not a win for all of its users.

What?

Making calls was the killer app for Nokia brick phones in the late 1990s.

The killer app for the first generation of smartphones (Windows Mobile, Blackberry, etc.) was email and calendar.

The killer app for iPhone and Android was the capacitative touchscreen combined with the ability to run 3rd party apps (yes, I'm aware there was an extremely brief moment in the history of the original iPhone where Apple opposed this), and 3G mobile internet (yes, again, I realize this came a year after the initial iPhone release). Mobile web browsers and Maps/GPS got the party started.

Coding doesn't do that, because the demographic interested in coding is not huge compared to the rest of the population.

Chatbots don't do it either because they're too unreliable. I never know if I'm going to get a recommendation for something the LLM hallucinated and doesn't exist.

There's also huge cultural resistance to AI. The iPhone was perceived as an enabling device. AI is perceived as a noisy, low-reliability, intrusive, immoral, disabling technology that is stealing work from talented people and replacing it with work of much lower quality.

It's debatable how many of those perceptions are accurate, but it's not debatable the perceptions exist.

In fact the way OpenAI, Anthropic, and the others have handled this is a masterclass in self-harming PR. It's been an unqualified cultural disaster.

So any killer app has to overcome that reputational damage. Currently I don't think anything does that in a way that works for the great mass of non-technical non-niche users.

Also - the iPhone was essentially a repackaging exercise. It took the Mac+Phone+Camera+iPod - all familiar concepts - and built them into a single pocket-sized device. The novelty was in the integration and miniaturisation.

AI is not an established technology. It's the poster child for a tech project with amorphous affordances and no clear roadmap in permanent beta. A lot of the resistance comes from its incomprehensibility. Plenty of people are making a lot of money from promises that will likely never materialise.

To most people there is no clear positive perception of what it is, what it does, or what specifically it can do for them - just a worry that it will probably make them redundant, or at least less valuable.

A Blackberry moment, perhaps. There appears to be something there, some groups are latching onto it and deriving value from it, but we haven't yet seen the iPhone come along to transform that initial interest into something that sweeps the world.

Old guy here, but it feels more like a Netscape moment than an iPhone moment. We'll end up with our pets.com of the LLM age, the whole thing will implode, and the few companies that were actually doing useful stuff with LLMs will survive.

For a while, all OEMs had 3D TV models, and it seemed their ubiquity was inevitable by sheer force of manufacturers ramming the products down consumers throats (like AI). The only debate was over which solution was superior: active or passive. 3D movies are still with us, so the tech didn't completely disappear - only from the consumer space.

What comes back to the old data-inefficiency of machine learning. There hasn't been visible improvement on this, and it is looking more and more as a fundamental limitation of AI.

You nailed it! Having spent significant time (as low-level minion) under both Ballmer and Satya, it certainly feels like the old Ballmer-time meme is coming back with the AI!

Also with it, the forced-curve ranking that Satya disbanded is being re-instituted under a different name.

"The killer app is making calls" is me quoting Steve Jobs on the Iphone 1 presentation. I get that it doesn't sound true now, knowing all we can do. It's true Iphone was a lot more, but that was his conviction at the time, and I think it makes sense. Their aim was to make the best phone in the world.

I also think yeah, it is a bit like Macintosh in the sense that this is a new general purpose technology, and I'm not sure we've really figured out what's going to the most transformative about it yet.

I do remember the initial visual voicemail implementation being very appealing of course. Especially since it seemed they had enough leverage to get the carrier/s (just one at the beginning) to do whatever they needed.

These days it seems like anybody can throw "AI" into their company name (even if it's complete BS) and it has the same effect as adding ".com" to a company name did in the late nineties.

IMHO AI is a .com-like hype cycle that's orders of magnitude larger and more irrational than anything that happened post-iPhone.

That's not to say that there aren't good businesses in there (the same was true of .com, of course), but there's a lot of junk that's getting a lot of money thrown at it.

Of course it could be better, but of all companies it seems that Microsoft managed to monetize best - sice copilot gets integrrted to the MS Office package

I think we agree. All I meant is, yes LLMs seem to do a lot of things, but nothing quite perfectly. Yet.