But my guess is that this is still quite a ways off. A few thoughts:

• There is no guarantee that the current approach of feeding massive amounts of data into the models will continue to scale as it has. The fact that it worked better than almost everyone thought it would is no guarantee it will continue to work.
• Is there some equivalent law of perpetual motion for intelligence? In other words, could it be that you won't ever be able to get more out than you put in? Does this limit the ability to surpass humans by merely feeding lots of data? 
• In many ways these models have already surpassed the average human in terms of the breadth of knowledge. But just knowing a lot of things about a lot of things doesn't really mean you're able to create completely novel ideas. 
• How much does our intelligence rely on our embodied experience? How much does it rely on our emotions? There is clearly a lot more to intelligence than just memorizing a lot of facts and being able to synthesize them through statistical averages in interesting ways.
• This is not just a software problem. Even if a super intelligence existed today and wanted to create a better version of itself, look at how much raw processing power over a period of months is necessary to improve the models today? The hardware and power demands of all of this are huge. 
• AI experts often come off as having a greatly over simplified view of how the world actually works. Even if a super intelligence existed today, there is still so much of day to day living that doesn't take place on a computer. Getting this intelligence integrated into the places it needs to be integrated will be a huge technical problem. Think about how inefficient many businesses are with technology that has already been around for many years. (This is also why I don't worry about this taking tech jobs. Most companies have 50 years of tech debt and potential improvements sitting in their backlog.)
• I worked with early versions of language models and never thought they would get this good this fast, so I'm prepared to be wrong.