The Pulse #115: LLM improvements slowing down?
The Pulse #115: LLM improvements slowing down?
Several signs indicate that improving LLMs with more training/compute is no longer efficient. Also: dev charged $1,100 after following a tutorial, a reminder to be vigilant with open source, and more
The Pulse is a series covering insights, patterns, and trends within Big Tech and startups. Notice an interesting event or trend? Send me a message.
Today, we cover:
Industry pulse. ChatGPT to compete with Google on Search, another large fundraise for an AI coding tool, Klarna files for IPO, possible public/private cloud equilibrium, and more.
LLM improvements slowing down? Several signs indicate the next versions of LLM foundational models won’t get much better despite the extra compute and training data. Reports at OpenAI, Anthropic, and Google, all suggest a slowdown in development, which could prompt a search for new routes to improvement.
Dev charged $1,100 after AWS tutorial. An AWS tutorial on how to integrate OpenSearch omitted key details, like how it launches an expensive service, and that shutting down the sandbox environment doesn’t shut down the OpenSearch service. A dev got burned financially, and is unhappy the tutorial still hasn’t been updated for clarity.
Open source project vigilance reminder. A look at an attempt to add malicious code to the Exo open source project. The attack seemed rather amateurish, and was perhaps intended to be spotted. It’s a reminder that open source projects are increasingly the targets of malicious activity.
1. Industry pulse
ChatGPT launches Search
In big news, OpenAI has launched ChatGPT search, which looks like a direct competitor of Google. The search functionality summarizes results on the web and links to sources, making them available as citations. It’s an approach that’s similar to AI search engine, Perplexity.
ChatGPT preparing to go head-on-head with Google matters because it has 200 million weekly active users, as of August. We don’t know how many monthly active users Google Search has, but it serves more than 5 billion searches per day, and YouTube has over 2 billion logged-in monthly users. A truly fascinating thing about OpenAI’s numbers is that these numbers were zero just two years ago, before ChatGPT launched on November 30, 2022.
If ChatGPT’s Search product gains momentum, it’s possible Google could adapt its own search UX to a summarization interface. It would be a drastic change, but it’s hard to imagine Google standing idle while its search product potentially being out-innovated.