Many workers fear the tightening AI crisis, but Marc Andreessen thinks you don’t have to worry (at least if you’re a software engineer). Just as a toddler starts to walk, AI needs a guiding hand — and general partner of venture capital firm Andreessen Horowitz, also known as a16z, thinks this will ultimately benefit some of the workforce.
“One of the biggest fears is AI-induced unemployment,” Andreessen said in a recently released episode of his podcast The Ben and Mark show. “The irony is that what’s happening today is a boom in AI hiring.”
In part, they fill AI’s recent growing pains. Whether these software engineers are digging their graves toward eventual obsolescence or ensuring a path to human-assisted AI remains to be realized. But in the short term, Andreessen sees growth at least.
The words of Andreessen, a key figure in Silicon Valley, are valuable. “Whether they love him or hate him, people pay attention when Marc Andreessen speaks. The venture capitalist is as responsible as anyone for the modern technology industry that powers the global economy, having co-created the first mainstream Web browser, Netscape Navigator, in the early 1990s,” Andreessen’s blurb reads FortuneThe 100 Most Powerful People in Business.
His early innovation in the software industry and his VC’s backing of major players in the field (including, but not limited to, Facebook, Lyft, and Airbnb) make him a powerhouse when it comes to the future of AI.
After intense growth, AI seems to be losing steam. According to a new study from Epoch AI, tech companies will soon run out of public data they use to train large language models. They are expected to hit a wall between 2026 and 2032, “or slightly sooner if models are overtrained.”
“There is a serious bottleneck here,” Tamay Besiroglu, one of the study’s researchers, told the newspaper Associated press. “If you run into those limitations about the amount of data you have, you can’t really scale your models efficiently anymore. And scaling models has probably been the most important way to expand their capabilities and improve the quality of their output.”
Andreessen echoes this sentiment in his podcast, explaining that there is a finite amount of high-quality human data on which AI models can train. “Basically the big models are trained by scouring the Internet and pulling in virtually all the human-generated training data, all the human-generated text, and increasingly video and audio and everything else,” he said. “And there’s literally only so much of that.”
