ChatGPT & AI News: The Big Questions

The world of ChatGPT and large language models (LLMs) is buzzing with news and developments. We're starting to see patterns emerge, and the big questions are coming into focus. Here's a rundown of what's going on:

• OpenAI's Internal Progress Scale: OpenAI has proposed an internal “level one to level five” scale of progress towards AGI, placing itself at level one today.
• Goldman Sachs on Generative AI Spending: Goldman Sachs released a report questioning whether the current wave of capex into generative AI will be supported by equivalent returns. The report highlights that it could take a decade for real applications to be built and widely deployed, and that many of the benefits could be cost savings rather than new revenue and profits.
• Andrew Ng on Regulating LLMs: Andrew Ng, a pioneer in machine learning, criticized California's proposed bill to regulate LLMs, arguing that it’s a crude wish-list from doomers and that trying to address problems from generative AI by regulating the models is like trying to stop fraud by regulating spreadsheets.

ChatGPT and the Question of Understanding

A core question about ChatGPT and other LLMs is how “clever” they really are. Are they truly understanding, or simply copying understanding?

• ChatGPT's Coding Abilities: A recent study suggests that ChatGPT is far better at coding problems that are likely to be in the training data, implying that it’s more likely to be copying rather than understanding.
• Doctors Using ChatGPT for Insurance Negotiations: Doctors are now using ChatGPT to argue with insurers, highlighting the potential for LLMs to automate repetitive tasks, like summarizing documents and writing emails.

ChatGPT and the Future of AI

As we move forward, it’s important to consider the future of ChatGPT and AI technology as a whole. Here are some key questions to ponder:

Will ChatGPT Keep Scaling?

ChatGPT and other LLMs have achieved impressive results by scaling up data and compute power. But how long will this trend continue? Will these models continue to scale, and how much bigger and better can they get?

• The Limits of Scaling: While scaling has worked so far, there are questions about how much further it can go. We may eventually run out of data, and the cost of training and running these models is already significant.
• The Need for Theoretical Understanding: We don’t have a good theoretical model for understanding what happens when we continue to scale these models, making it difficult to predict their future capabilities.

The Economics of ChatGPT

The cost and scope of ChatGPT and similar AI models are significant factors to consider. As models become more complex, so do their training and inference costs.

• The Trade-off Between Size and Cost: We can get a significant portion of the benchmark score of the best possible model at a fraction of the cost. This raises the question of whether it’s more efficient to use smaller, less expensive models for most tasks and larger models for only the most demanding ones.
• The Challenge of Consumer Adoption: Mass-market consumer adoption may be difficult for AI products with material marginal costs, especially if they don’t offer significant value.

ChatGPT's Role in the Tech Stack

How far up the tech stack will ChatGPT and similar AI models go? Will they become commodity cloud infrastructure, or will they be able to take on more complex tasks?

• Commoditization of LLMs: Apple and Meta are both moving towards making LLMs more readily available as commodity infrastructure, but this may lead to a situation where the value is captured by those who build the infrastructure rather than those who build the applications.
• The Rise of AI-Powered Applications: It’s possible that LLMs will become so powerful and generalized that they can power a whole new generation of applications.

The Future of ChatGPT and AI

The economic impact of ChatGPT and AI remains uncertain. It’s possible that the value created by these technologies will be captured by a few large companies that control the infrastructure, or it could lead to the creation of thousands of new companies and applications.

• The History of Disruptive Technologies: Technologies like fiber networks and airlines have often created immense value without making much money for those who initially invested in them.
• The Value Chain: It’s possible that the value from AI will be captured by those who build on top of the infrastructure rather than those who provide the infrastructure itself.

Outside of ChatGPT: Tech & News

While ChatGPT and AI are dominant topics, there's a wider world of tech news happening. Here are some notable points:

• Apple and Microsoft Give Up OpenAI Board Seats: This move might be due to increased scrutiny from competition authorities and suggests a shift in the relationship between these tech giants and OpenAI.
• a16z’s GPU Cluster: a16z is building a cluster of Nvidia GPUs to rent to portfolio companies. This move shows the increasing demand for GPUs and the importance of access to these resources for AI development.
• Everything is Hacked: From AT&T customer records to Indonesia's government systems, cyberattacks continue to be a growing problem.

Data and Insights: ChatGPT and the Venture Capital Landscape

The venture capital world is closely watching the evolution of ChatGPT and generative AI. Here are some data points and insights:

• Coatue’s Venture Tech Investing Report: Coatue’s annual presentation offers insights into the state of venture tech investing, particularly in AI.
• OpenAI’s Revenue Estimates: OpenAI is estimated to have $3.4 billion in ARR, with only 15% coming from its API, indicating a large portion of its revenue coming from other sources.
• ChatGPT’s Academic Use: Researchers have observed a surge in academic use of ChatGPT, potentially due to its ability to generate text quickly and easily.