2024: The 7 Big Questions for Enterprise AI

Welcome to AI Pioneers at Work, a newsletter for technical leaders building AI products in the enterprise. To join our community of >1,000 product and engineering executives using LLMs and modern AI, please sign up here.

It is hard to forecast the future of AI, but in our community of AI Pioneers there are many big questions on everyone’s mind going into the new year. We recently shared some of our thoughts in “AI predictions for 2024: What top VCs think” and have additionally crowdsourced some of the most interesting and controversial topics on the minds of many people in our community. We are excited to share them with you here without much fine tuning, and look forward to getting your opinions on them as the year progresses:

Question #1: Is this the year we take Google seriously?

If 2023 was the year of Open AI and Microsoft owned the airwaves, this should be the year we will all be talking about Google. Their substantial investment in Gemini and unrivaled data and compute resources will likely offer developers GPT-4+ capabilities in all shapes and sizes, pushing the frontier for all foundation model providers. We expect everyone to be taking them seriously. 

Question #2: Is this the year Nvidia becomes a cloud provider?

NVIDIA has not so quietly announced its DGX Cloud offering earlier this year, and has the market cap and the momentum to credibly offer cloud services and a full stack alternative to Amazon, Microsoft, and Google. Continued investment in AWS’s Trainium and Inferentia chips along with Google’s long standing investment in TPUs is leading to a clear competition for Nvidia with every major cloud provider. Jensen’s next move will need to be big (credit to our friend Rob Toews for making this prediction). 

Question #3: Will there still be a GPU shortage?

A quietly building contrarian opinion is we will soon *not* have a GPU shortage. It is becoming harder for midsize companies to rationalize the cost of training their own foundation models. The vast majority of future consumption will likely come from Microsoft, Google, Amazon, Facebook, and many sovereign nations. The rest of us should therefore be able to rent GPUs from cloud providers with plenty of leasable capacity to go around. Yes, the rich will get richer - but quality of life should improve soon for the “GPU-poor”.

Question #4: Will the transformer architecture hit a plateau?

We hosted over 200 AI researchers at NeurIPS where a record 3,540 papers were published. Transformers have emerged as dominant based on the predictable premise of better models from more parameters and training data. But the research winds are now shifting towards mathematically-equivalent alternatives to attention which improve scaling with dramatically less compute, storage, and size. The next wave of LLMs will likely be smaller and more skilled, with plenty of room for innovation at the bottom (see our Latent Space NeurIPS summary here). 

Question #5: How rapidly will we see AI “agents”?

The major releases of multi-modal models this year will make it near impossible for humans to discern the difference between a computer and a human in certain use cases. We can see this already in places like Character AI and Instagram and expect this to take hold in the workplace in areas like training, support, and marketing / sales. Enterprises will lag consumers, but we may be building customer relationships with machines sooner than we would like to admit.

Question #6: Who will start making money in AI?

We know Nvidia, Open AI, and Microsoft are the early winners from the AI wars. We also see positive signs from Adobe, Salesforce, and others who are forecasting significant AI revenue growth in the year to come. Will anyone else profit from AI in 2024? The rise of new AI applications will ultimately force the question of who can charge more for AI, and who will get the pricing right. The current world heavily subsidizes adoption, with a large benefit to end customers. Like ride sharing and cloud consumption before it, we expect the credits to continue until the AI market share wars settle out. 

Question #7: Can we deploy AI without human “alignment”?

Teaching an AI system to properly respond like a human is an age old problem, encapsulating AI ethics, governance, and security. The rise of AI regulation and the business risk of an “unaligned” system are now front and center, requiring every company to open the Pandora’s box of issues in AI alignment. Who decides whether an AI system is safe? Is it better done by humans or machines? We expect techniques like RLAIF and trends such as AI “red teaming” to be hot topics when moving LLMs from pilot to production. 

Our community is full of many great AI thought leaders and we are grateful for all of their insights shared this year. Several of our AI Pioneers were recently highlighted in the media for their AI predictions this year including our friends at Walmart, Cisco, Intuit, and Nvidia, and we will be publishing more of your insights on deploying AI throughout the year. 

We are excited to host our first AI Center of Excellence virtual roundtable for our founding members in the coming weeks. As always, feel free to reach out if you are interested in participating in our events and we look forward to seeing you all this year!