Over the holiday, hanging out at Facebook for the first time in years, I found Cluetrain co-author Doc Searls flagging something called Kwaai.
Kwaai aims to build open source tools for AI, much as the open source cloud developed after the dot-com crash early this century.
I looked at the website and thought deeply about this, coming up with two thoughts.
First, the open source revolution came about when I was a much younger man. We were all younger then, even Doc. Having lost work at various Internet start-ups, many found ourselves with time on our hands, yet still idealistic. I don’t know how many open source programmers “back in the day” were living in their parents’ basements. I doubt it was that many. But some were. I do know we were naïve enough to believe we had time to turn new ideas into action.
I’m not that way anymore. Most who just ended their careers on the rocks of AI have come out with enough to get to the end of the road, rising to meet us. I suspect it’s even true for some reporters. Tell us to think a decade ahead and we’re bound to see a cemetery as real estate.
At our age we can lead and we can advise. We can’t do the job. Kwaai founder Reza Rassool is in America, and his roadmap reflects that.
What Kwaai Can Do
But Kwaai is an African word. It’s South African slang, for cool or excellent, derived from a Dutch word meaning angry or aggressive. That’s the way language works. Graft used to be mean “hard work” in English. It was changed by Americans into meaning “corrupt.” Funny, that.
So far, Kwaai is just a call to action, and that nearly 18 months old.
To become something more, Kwaai needs a more realistic mission.
There’s a lot of western wringing-of-hands on the Kwaai page, worries about an oligopoly of huge companies building LLMs that concentrate power in just a few hands. A natural fear. It just happened in America.
But there’s no Moore’s Law for LLMs. Quite the reverse. The more data you pack into them, the slower they improve. I think they’re a dead end.
The Hot New Trend
The hot new trend is AI agents. These are programs that represent you on the Internet, trained only to carry out specific tasks. Some agents are honestly looking for bargains. Others are false personas, impersonating people to steal from others.
The point is that you don’t need Facebook’s LLaMa, or any enormous LLM, to build agents. This makes them perfect for African programmers, who are naturally resource constrained. Africa is also where you’re most likely to find communities of trained, or semi-trained, programmers who are young, scrappy, and hungry, the ingredients for great open source projects.
Western and even Indian companies working on AI agents haven’t scratched the surface of what these programs may be capable of. Personal agents that represent your desires on the Internet can help you make connections and build a career within and beyond Facebook. Shared agents can work on shared projects. Negotiating agents can help people come to fair agreements, in markets and even on public policy.
There’s a whole new world out there and, most important for the programmers, huge profits to be won, by finding and developing AI niches the west has ignored. The ideals in the Kwaai manifesto are fine. I subscribe to them.
But give people something that will make them money and they’ll go further, faster.
