Forward Future AI
Posts
Forward Future Episode 14

Forward Future Episode 14

FTC investigates OpenAI, Claude2, PoisonGPT, xAI and more!

July 14, 2023

What an exciting week in the world of AI. This was certainly a busy week in terms of launches and news. Let’s start with the biggest news of the week, which is OpenAI is now being investigated by the FTC, and then we’ll get into AI announcements and other news.

FTC investigates OpenAI

The FTC demanded a plethora of data, information, and documents from the AI giant and company behind ChatGPT and marks the company’s greatest regulatory threat to date. Just a few weeks ago, Sam Altman, the CEO of OpenAI, was testifying in front of Congress about AI safety and extolling the need for strong AI regulation.

Many, including myself, believed this strategy to be a front run to avoid major regulation targeted at OpenAI, but their strategy seems to have not worked.

The FTC this week sent a 20-page demand for records about how it thinks about risks related to AI and specifically ChatGPT. Additionally, it asked for information and complaints from users about false and misleading information provided by ChatGPT. In clear text on the ChatGPT app, it says “ChatGPT may produce inaccurate information about people, places, or facts.” But is that enough?

From the Washington Post article about it: “The FTC is investigating whether the company engaged in unfair or deceptive practices that resulted in “reputational harm” to consumers, according to the document.”

The FTC is also interested in internal documents related to the data leak that occurred earlier this year, which allowed some ChatGPT users to see other users’ conversations, although OpenAI stated the leak was very small.

It will be interesting to see how OpenAI handles this. The current FTC administration is known for going after tech companies so this is no different. OpenAI asked for regulation, and now they’re getting it.

Claude 2 Launched

Anthropic AI, the company behind Claude, has now launched Claude 2.

In a blog post published a couple of days ago, Anthropic said: “Claude 2 has improved performance, longer responses, and can be accessed via API as well as a new public-facing beta website”

In what many consider the closest competition to GPT4, Claude performs extremely well on many tasks and ranks close to GPT4 on the LLM leaderboards. One of the main differences between the two dominant (close-source) models is ChatGPT has essentially no personality and will just follow commands it is given, whereas Claude was built to have a personality and push back on prompts where necessary. Claude is also known for having a much larger context window of 100k tokens.

Any competition is good competition when it comes to AI, whether in the form of close-source models or open.

PoisonGPT

I made a video this week about a blog post written by Mithril Security where they detail a method in which they were able to “poison” a large language model. This means they were able to change specific facts housed in the model and spread the model pretending to be a trusted model author. Please watch this video to learn more and find out how to protect yourself against poisoned models.

Elon Musk Launches xAI

Elon Musk is the gift that keeps on giving to Mark Zuckerberg | CNN Business

Elon Musk finally launched his own AI company this week called xAI. The mission of xAI is to “understand the true nature of the universe.” Of course, that’s a massive mission and one that aligns with Musk’s other companies. Although he says xAI is a completely separate company from his others, he says it will work closely with his other companies to accomplish its mission.