Reddit Revolt Explained
The biggest news this week was Reddit’s user revolt. This may seem at first glance unrelated to artificial intelligence, but it is highly related, and we may be witnessing a change to the fundamental structure of the internet.
This week, thousands of subreddits “went dark” in a protest against Reddit’s upcoming API changes. Specifically, Reddit plans to start charging for API access, which was previously free. This will affect thousands of developers who build apps on top of Reddit to make it even better.
One of the most popular apps is called Apollo, which is built by Christian Selig. Christian figures the API changes would cost him about $20 million per year, obviously making his business unable to continue operating. Reddit also gave developers only 30 days to comply, which left many scrambling for solutions. Ultimately, Christian decided to shut down Apollo one day before the new prices take effect.
How does this all relate to AI? Ultimately, it comes down to data. The value of data has always been high, but with the proliferation of large language models, the monetary value of quality datasets is growing exponentially. Large language models feed on data. Most LLMs are built on many of the same, open-source and publicly available datasets. For an LLM to differentiate, training on a unique and highly structured dataset might give it the performance boost it needs to be the best LLM. Also, LLMs are driving tremendous value for their owners, and Reddit realizes that their data is “leaking” through their API. Since Reddit has an incredible and unique dataset, they want to monetize it in anticipation of its IPO.
Reddit isn’t the only company effectively shutting down free and open access to their API. Twitter has also taken the same approach by charging for API access. Twitter has a unique and highly valuable dataset, just like Reddit. Elon Musk has already shared plans to expand their AI development, and their dataset will help them build a competitive LLM.
This looks like the first step of a significant transformation of the free and open internet. For years, providing free and easy access to a company’s data via an API has been a staple of a vibrant ecosystem. Platforms have significantly benefitted from 3rd party engineers building applications on top of their systems. All of that is likely to change now that the data that feeds LLMs looks more valuable than allowing 3rd party developers to build applications on top of platforms.
There’s another wrinkle to this story. Both Reddit and Twitter are 100% user-generated content. Ultimately, these platforms are just hosts to user data, but they technically “own” that data. Not only are they already monetizing the data by serving ads to users, but now they are also monetizing the data to feed LLMs. This is a misalignment of incentives with their users since users gain a lot of value from 3rd party applications.
So, shouldn’t users get paid for generating all of this data they are creating? Elon Musk has already planned to share revenue with Twitter users, which is a step in the right direction, but Reddit hasn’t disclosed any such plans. This seems like a ripe time for users to realize their value and for another platform to come along and pay users for their time and energy in creating content.
In an article by Dmitry Mazin, he takes it further by theorizing that there could be some collusion between Reddit and OpenAI. Sam Altman, the founder, and CEO of OpenAI, is also deeply entrenched in Reddit. He was the former president of YC, the original investor in Reddit, and an investor himself, close friends with the founders of Reddit, and was temporarily the CEO of Reddit. If OpenAI has access to Reddit’s data through a backdoor deal, that could give them a significant competitive advantage against all other LLMs. Given the deep connections between these companies, it’s not impossible to imagine this scenario being true.
I believe we are in a transformative period for the internet. I’m nervous about the free and open internet that we know and love today being closed down and siloed because every company wants to extract as much value as possible from the LLM gold rush. One potential silver lining for this is that users generating free content for platforms like Twitter and Reddit will realize their value and start to demand profit-sharing. If these companies don’t begin to treat their users as the true value source, I hope another startup comes along and eats its lunch.
ChatGPT Updates
ChatGPT released a bunch of new updates to their API, here’s a list of those changes:
A new function calling capability in the Chat Completions API.
Updated and more steerable versions of gpt-4 and gpt-3.5-turbo.
New 16k context version of gpt-3.5-turbo (compared to the standard 4k version).
75% cost reduction on the state-of-the-art embeddings model.
25% cost reduction on input tokens for gpt-3.5-turbo.
Deprecation timeline for gpt-3.5-turbo-0301 and gpt-4-0314 models announced.
Function calling allows developers to more reliably get structured data back from the model.
For instance, developers can create chatbots that answer questions by calling external tools.
Convert natural language into API calls or database queries.
Extract structured data from text.
New models introduced:
gpt-4-0613 and gpt-4-32k-0613 are updated with function calling and extended context length respectively.
gpt-3.5-turbo-0613 has function calling and better steerability.
gpt-3.5-turbo-16k has 4 times the context length of gpt-3.5-turbo.
Deprecation process for the initial versions of gpt-4 and gpt-3.5-turbo announced.
Lower pricing for the popular embeddings model and gpt-3.5-turbo model introduced.
QR Codes Are Cool Now!
QR codes are getting the AI-treatment. After an individual showed off a new kind of QR code, which uses stable diffusion to make them look visually stunning, there’s a proliferation of QR code tutorials, libraries, and even fully-hosted websites using generative AI to create incredible looking QR codes. I’m planning on making a video tutorial about this, so stay tuned!
