Reflections On My 2020 Predictions For AI, DL, And ML

On December 19, 2020, just over a year ago, I published a piece titled 2020 Predictions For AI, DL, And ML. And while it’s safe to say I never predicted how different this year was going to look – in so many ways – it was interesting to pull up the article and recognize that all five of my main industry predictions were either realized or moved forward. While so many industries came to a halt, this forward momentum shows the continued force being pushed into artificial intelligence, deep learning, and machine learning. 

In this piece, I’ll look at each of my predictions from last year and reflect on the movement and continued drive behind each. 

2020 Prediction #1: Exclusivity and Consolidation 

It’s fitting that this was my first prediction of last year because over the past 12 months we have seen big players make big deals, creating an ever-increasing environment of AI winners and losers. Interestingly it was not just the big hardware companies buying other hardware companies, rather we saw companies move up the stack in a manner never seen before. Apple’s acquisition of Xnor.ai – a small maker of low-power, edge-based artificial intelligence tools – provided an interesting window into a FAANG giant looking further up the stack.

However, the biggest news moments of this category – and frankly in the AI space during 2020 – were headlines I didn’t see coming at the start of last year. While I thought 2020 would be filled with smaller acquisitions, we instead saw the largest semiconductor acquisitions in history – Intel buying Xilinx for $35B and NVIDIA buying Arm for $40B. These deals were AI fueled and showed the true industry appetite for exclusive deals between traditional legacy players and those at the cutting edge of technology. 

The consolidation in the market has made it more clear than ever that those I’ve always considered the traditional players recognize the importance of expanding into AI, DL, and ML in big ways. 

2020 Prediction #2: Big Growth for NVIDIA

My second prediction, that NVIDIA’s data center revenue would continue to grow, was seen time and time again through huge NVIDIA milestones this year. In early July, NVIDIA’s market cap exceeded Intel’s for the first time in history – a huge moment for the semiconductor industry as a whole. 

And as 2020 has gone by we’ve seen this difference in market cap continue to widen in favor of NVIDIA. At the time of writing this story, NVIDIA’s market cap is at $328.77B and Intel’s market cap is at $189.21B. These figures – and the difference between them – continues to separate. However, I think there will be more room for further NVIDIA growth over the next year, as no competitor has yet delivered a compelling hardware/software alternative to market.

2020 Prediction #3: PyTorch > TensorFlow

The graph above is a clear way to really look at this prediction. As you can see during the first month of last year about 23% of academic papers were using TensorFlow, so far in December 2020 this has fallen to only 11%. Over the coming year, I believe we’ll continue to see even more developers moving towards PyTorch and away from TensorFlow. This is because of the many advantages of PyTorch – from ramp up time to easier debugging – furthering this continued draw for teams and programmers in the long run. 

2020 Prediction #4: Deep Learning for Language Modeling Makes Big Jumps

This prediction was also for sure correct, and in a lot of ways perhaps even more hyped than I expected last December. Over the past 12 months, we have seen projects like GPT-3 do so many amazing things – really blowing past language modeling away by every measure.  

However, over the next year I hope – and believe – that the current hype in this area can come back down to earth. I believe we are at the peak of the hype cycle with language models, and that we must look at our advancements carefully while fixing many serious limitations. As Yann Lecun said in his post criticizing GPT-3 as a dialog system “trying to build intelligent machines by scaling up language models is like building a high-altitude airplane to go to the moon. You might beat altitude records, but going to the moon will require a completely different approach.”

2020 Prediction #5: 2020 Election

Interestingly, this prediction, which in many ways seemed almost guaranteed – is probably one of the hardest to go find evidence for. Surprisingly, we did not see a lot of deepfake issues. However, over the course of the election cycle the role of AI at social media companies seemed to likely reinforce echo chambers and continue to drive wedges between the existing divide. For example, consider social media giant, Facebook – who rather than switching the algorithms decided to stop showing political ads altogether during the election, but the critical piece that was still in play was the social narrative being reinforced by echo chambers on the platform. Taking away ads didn’t remove echo chambers, rather it just took away the people directly paying to pop up in your echo chamber. 

Conclusion

All in all, while it has been a crazy year to say the least, it is clear that the movement in the AI, DL, ML space is only just beginning. Soon I’ll be publishing new predictions for 2021, more on that note soon.