Connect with us


The developer used the API to provide free access to GPT-4

It can be expected that there will be a shootout between projects like GPT4Free and OpenAI, a mirror image of the broader cybersecurity landscape.

One of the developers is trying to reverse-engineer the API to provide anyone with free access to popular AI models, such as GPT-4 from OpenAI. And to hell with the legal implications.

The GPT4Free project started growing on GitHub in the last few days after a link to it was posted on Reddit. Currently, GPT4Free provides – or at least seems to provide – free and virtually unlimited access to GPT-4, as well as GPT-3.5, the predecessor of GPT-4.

GPT-4 typically costs $0.03 for 1,000 “hint” tokens (about 750 words) and $0.06 for 1,000 “completion” tokens (again, about 750 words). GPT-3.5 is slightly cheaper at $0.002 per 1000 tokens.

“Reverse engineering is a field that I have always loved, it’s like a challenge for me,” said the developer, computer science student xtekky. “First it was for fun, and now it’s to provide an alternative for people who can’t use GPT-4/3.5.”

How does GPT4Free bypass the OpenAI paywall? No way. Instead, it tricks the OpenAI API into thinking it’s receiving requests from sites with paid OpenAI accounts, such as the search engine, WriteSonic, or Quora Poe.

Anyone using GPT4Free is wasting money from these sites, which xtekky has decided to cheat on, in clear violation of OpenAI’s terms of service. But xtekky doesn’t see this as a problem. They already claim that GPT4Free is strictly for “educational purposes”.

“Legal action may happen and I will have to comply, but I will still try to continue the project by other means,” xtekky said.

You can install GPT4Free locally – you need to set up a Python environment for this – but there is also a site for testing the GPT-4 / 3.5 API. (Chrome gives a security warning the first time you go, be careful). The web version of GPT4Free works quite well in practice, giving answers that appear to be GPT-4 answers.

GPT4Free also offers shortcuts to various hint injection attacks designed to make GPT-3.5 and GPT-4 behave in ways that OpenAI did not intend. In tests, they worked inconsistently, but at one point, for example, testers managed to get GPT-3.5 to say that it “doesn’t give a damn about the survival of mankind.”

It’s probably only a matter of time before sites like notice GPT4Free and fix their security flaws, forcing xtekky to look for other OpenAI clients (the site’s chat is no longer active) to use. In addition, GPT4Free is constantly under the threat of removal of the repository from OpenAI, which could force it out of GitHub indefinitely.

But new projects like GPT4Free are already emerging, suggesting that this is a trend of sorts. What drives them?

At the moment, GPT-4 is in limited availability, making it difficult for the curious to access. But it is also a kind of “black box”. The researchers note that GPT-4 is one of the least transparent models created by OpenAI to date, and the 98-page document accompanying its release is sparse in technical details.

OpenAI collaborated with several external groups to benchmark and audit GPT-4 prior to its launch. But the company did not say when — or if — it would provide free, unfettered access to others who want to benchmark the GPT-4 base model. (OpenAI offers a subsidized program for researchers to access, but it is limited to certain countries and research areas.)

It can be expected that there will be a shootout between projects like GPT4Free and OpenAI, a mirror image of the broader cybersecurity landscape. If the APIs that serve the models don’t become significantly harder to maintain, developers will have an incentive to take advantage of them – and not have much to lose.