Ready, AI, Fire!
The search for AI impact.
This is the first in a series of opinion articles about AI at work. I am a technical person with a computer science degree and some experience with the software industry. However, I view AI as transformative for a broad worldwide non-technical audience, and judge the success of AI by its impact beyond the technology industry.
“Hello, hello, hello, how low
Hello, hello, hello, how low”
<Smells like Teen Spirit, by Nirvana, from the album Nevermind, 1991>
It’s been more than 20 months since ChatGPT was launched and blew our minds. The mind-blowing didn’t happen all on one day. People talked about it at parties and online, simply amazed by the miraculous things it could imagine and articulate. It spread person-to-person, group-to-group faster than any biological pandemic. And we witnessed the first AI gold rush.
So then, why hasn’t AI changed anybody’s life a whole lot in the last 20 months?
Email still sucks. Meetings are still tedious. Work is still more-or-less as difficult as before. Our cars don’t drive themselves. Facts are still buried by propaganda. So … what was all this fuss about?
“Load up on guns, bring your friends
It’s fun to lose and to pretend”
Every mega-tech company realized at the start of 2023 that maybe this was the moment to go all in with those billions of dollars they’d hoarded for years. Every one of those magnificent companies and all at the same time.
The market loved it and pumped up the magnificent seven by a few trillion dollars. And from all of that gold-rush fever, Nvidia became the most valuable company in the world — the only seller of those special GPU shovels you need to mine the new AI gold.
Check out the product roadmap for any product at Microsoft or Google or Amazon. Half the features will be about AI. Check out PhD programs in CS departments. Half the students are working in AI and machine learning. Look at new startups or any recent Y Combinator cohort. Half of them are “something.AI”.
Along with a few of my colleagues, I too started Thunk.AI. We have the gold rush fever and true belief too. All of us are ready and armed. We are all firing away with AI. It should be a veritable Cambrian explosion of AI innovation. Yes? Yes? But somehow … no, not so far. At least, not beyond the core LLM models.
Perhaps the problem is with our collective aim.
“Ifeel stupid and contagious
Here we are now, entertain us”
Perhaps it is also about the expectations.
As consumers, we’ve set the bar super-high after using ChatGPT, Dall-E, and Sora. We expect to be amazed. AI, entertain us, dammit! We’ve been promised an AGI Terminator-style end of the world and it feels a far more exciting way to go than a gradually cooking planet with melted icecaps and avian viruses (I say that only half in jest).
As for enterprise customers, that’s a whole other story. They have assigned significant budgets to AI. They have spun up teams and centers of excellence to focus on AI. They aren’t really obsessed with soap-opera issues like “what did Ilya see?”. If you’re running Walmart or CVS or McDonalds, you have more immediate and practical concerns. Will the AI models learn from our data, what does it mean to have AI security? Overall, I think this business customer “grounding in reality” might actually be what rescues this wave of AI technology and makes it useful.
All the same, there’s a big caveat: enterprise customers only buy what the tech companies sell them. Two kinds of AI offerings are being relentlessly pushed at enterprise customers today. One offering is infrastructure and includes amazing but not entirely reliable LLM models. Now the customer has to figure out how to put these to effective use for their business (wide-open risky innovation) and how to actually implement that (a completely non-trivial undertaking). The startup we’re working on, for example, is trying to help business customers with these problems. Perhaps unsurprisingly, we find that it takes a lot of discussion about use cases and scenarios, education about the capabilities of AI today, and guesswork about what to expect in the near future.
Above the infrastructure, customers also are coerced to adopt completely underwhelming legacy applications with a serving of AI “co-pilot” goop tacked on in distracting locations and for a steep premium. I make that sound super unappealing, don’t I? Perhaps because I wrote this article. It wasn’t ChatGPT with its anodyne long-winded blah-blah blandness. But in truth, these application co-pilots are far removed from the promise of improved productivity. I suppose I have seen useful AI products that take meeting notes or provide synthesized search results. Do I use them regularly though .. that’s a different matter.
There is one audience that is really benefiting — — software engineers. The AI-based code generation tools are really good. So it’s wierd and unexpected in a way, but all this techie tech is so far benefiting mainly the techie crowd and may put some techies out of work. Go figure.
“Ifound it hard, it’s hard to find
Oh well, whatever, never mind”
I don’t think consumers or business customers have quite yet reached the “whatever, never mind” phase. But they’re headed in that direction. And so are investors — whether in the public markets or VCs. What’s going to stem the rot?
Perhaps we need Sam to ride to the rescue one more time.
We know OpenAI is going to expand our minds again this fall, because that’s simply how that company works. They do amazing things, and then six months later, they do it again. Sometimes these things are amazing in ways that everyone doesn’t appreciate. When have you ever heard of a company that is the obvious leader in a hot technology announce that it will double the speed and halve the price at the same time. That doesn’t get talked about at dinner parties but it is transformative for the eco-system of customers and products that build on top of OpenAI. Despite their boardroom soap-opera, OpenAI is redefining innovation in a way we have never seen before.
Of course, OpenAI will inevitably be dragged back to earth. By then, the rest of us in the AI software business need to improve our aim with AI applications that translate the “wow” potential into “wow” obvious tangible value.
There’s a lot to do. We have been presented with a powerful new LLM “computer” — it is a computer that understands language, acts a bit unpredictably, and gets more powerful every year. But …
1. We don’t yet have a new operating system or a new application model that can harness this new computer.
2. Users don’t yet know what to expect of these new AI applications in terms of capability or user interface.
3. We don’t have an eco-system where people and teams can work together and leverage prior work to build AI applications.
So yes indeed, there’s a lot to do. More about that in my next article.
