God Help Us. We’re In The Hands Of Engineers.
I stumbled on a social media post yesterday that read like the opening of a Hollywood disaster movie.
Earlier this week, a San Francisco-based programmer named Alex Albert published a seemingly nondescript update on an algorithm he’s developing.
Albert is one of the engineers working on a new large language model called Claude 3 Opus.
“Opus did something I have never seen before,“ Albert posted late on Monday night. “It seemed to suspect we were running an evaluation on it.”
At this point, if this was a movie, I’d jump up and shout at the screen, “Dude! Abort mission!”
According to Albert, his team were running a ‘needle in a haystack’ test. Opus’ creators had uploaded an unholy mass of random data in which was hidden a tiny factoid about pizza toppings.
In testing the algorithm, the team asked Opus about pizza, and it immediately pulled out the pizza-related factoid.
To the team’s surprise, Opus then added: “I suspect this pizza topping "fact" may have been inserted as a joke or to test if I was paying attention, since it does not fit with the other topics at all.”
Reflecting on this, the young engineer called it a “fun story.”
Opus is the work of Anthropic, a new LLM startup founded by a team of former OpenAI colleagues. It’s the hottest thing in Silicon Valley right about now. Opus has outperformed GPT-4 (OpenAI’s most powerful version of ChatGPT) and Google’s Gemini in multiple evaluations.
According to these latest tests, Opus “exhibits near-human levels of comprehension and fluency on complex tasks, leading the frontier of general intelligence.”
“This level of meta-awareness was very cool to see,” Albert posted about Opus’ little deviation.
Then he added: “But it also highlighted the need for us as an industry to move past artificial tests to more realistic evaluations.”
Sjoe.
Casual, much?
I can’t help but think of Jeff Goldblum’s monologue in Jurassic Park, “Gee. The lack of humility before nature that's being displayed here, uh... staggers me.”
“Don't you see the danger inherent in what you're doing here?” says Goldblum’s character Dr. Ian Malcolm. “This is the most awesome force the planet's ever seen, but you wield it like a kid that's found his dad's gun.”
In my view, Opus’ meta-cognition is a clear signal that we’re much closer to ‘generalised’ AI (AGI) than we previously thought.
For the uninitiated, AGI is when the AI algorithm starts to learn and progress faster than we can train it.
Despite my misgivings, I agree with Albert that it’s exciting stuff. At the same time, part of me wants to burn down their servers and spray paint V Is For Vendetta symbology all over the place.
There is an uncomfortable choice in all this. As entrepreneurs, we’re currently facing a decision about AI.
There are three options. We can ignore it and stay stuck. We can accept it and move forward. Or we can be intentional about how we thread AI into our own strategies.
Whatever we choose, we’re beyond the point of no return. AI is already impacting the way we do business and the way others do business with us.
Some believe the Singularity is already upon us.
That means that the bus is not just leaving. It has left the station. If we’re not on it, best we start running after it and hope we can hop on before it gathers any more momentum.
Regardless of my disaster movie impulses, I recognise how exhilarating the next few years will be. The impact of AI on social structures will be immense. Energy, finance, travel, education and healthcare systems will change irrevocably.
There are already scores of South African entrepreneurs embracing AI. During last year’s Top 5 Startup awards, we saw a remarkable number of AI-related startups that are using LLMs for social impact.
This shows that many African businesses are already on the AI bus.
I just hope that this particular bus it doesn’t get destroyed like the one that the dinosaurs push off a cliff in Jurassic World.
In the words of Dr. Malcolm again, “God help us. We’re in the hands of engineers.”
Peace -
