AI Intelligence Briefing: No Content Without Context

CHAT GPT: Booming or Bankrupt? Some People Can’t Decide

A few weeks ago, analyticsindiamag.com floated the idea that Open AI, owner of Chat GPT, was struggling under the financial weight of $700,000 per day operating costs of Chat GPT, and that they might go bankrupt in 2024 (even though simple math would suggest that a $10 billion investment from Microsoft could maintain a $700,000/day burn for decades). This story was picked up by a number of outlets covering AI with very little common sense pushback.

Now. miraculously, theinformation.com is reporting that Open AI is on pace to generate over $1 billion in revenue over the next twelve months. And those same outlets that speculated on Open AI’s impending bankruptcy are now hailing its “turnaround” and speculating that it is a result of their release of Chat GPT Enterprise.

Once again they miss the mark. First of all, they just released ChatGPT Enterprise, so it is doubtful they have done any real business. More to the point. had Open AI not released the Enterprise version, they would have ZERO chance of getting any significant business from large enterprise clients. Why? For the same reason that a growing list of companies banned the use of Chat GPT including Apple, Amazon, Accenture, Bank of America, Citigroup, Deutsche Bank, ,Goldman Sachs, iHeart Radio, JP Morgan Chase, Northrup Grumman Corp, Samsung, Spotify. Verizon, and Wells Fargo.

It stems to a series of data leaks at Samsung that were discovered in May. Some Samsung employees uploaded proprietary code to ChatGPT without realizing that that code is now part of ChatGPT’s training and may be made available to other ChatGPT users. While no one has reported on this as a crisis, it is clear that any company with sensitive data on would run the other way unless Open AI could guarantee them more data security. And that’s what they did with Chat GPT Enterprise. Their description leads with :enterprise-grade security and privacy” and promises that “you own and control your business data in ChatGPT Enterprise. We do not train on your business data or conversations, and our models don’t learn from your usage.”

Whether this convinces large enterprise customers to engage with Open AI or start their own internal AI projects remains to be seen, but at least they have addressed the biggest hurdles and can start a sales dialogue. Meanwhile, for the rest of us, it may be a good time to RTFTOS (Read The F*in Terms Of Service) and make sure you understand exactly what you are doing when you use ChatGPT.

As the governments of the world start to address he impact of AI, some have already made strides in addressing the issues that concern them most, while others are taking more of a “wait and see” approach. In the US, Senate Majority Leader Chuck Schumer has scheduled a closed-door meeting on September 13 with the CEOs of all of the leading AI companies including Alphabet, Open AI, Microsoft, Meta, Tesla, NVIDIA, IBM and Palantir. Whether this meeting leads to US legislation or simply the formation of an AI cartel is still anybody’s guess, but one possible outcome is the much-awaited grudge match between Elon Musk and Mark Zuckerberg.

Meanwhile, while the US schedules meetings, countries such as Brazil, Italy, China, Japan, Israel and the European Union are already moving ahead with draft legislation and policy recommendations, according to an article in the Washington Post. Brazil’s draft law focuses on AI users rights and AI developers’ responsibilities while China’s draft laws holds developers accountable for training data that infringes on intellectual property and also requires generative AI to hold to ‘socialist core values”. The European Union has focused its attention on defining and containing high risk activity from AI deployments. And Israel and Japan have developed policies encouraging self-regulation and soft intervention from the government. Italy has focused its concerns on data collection and has already allocated $33 million to support workers at risk from digital transformation.

Despite the challenges of running multiple companies, Elon Musk always seems to have bandwidth to incorporate AI into all of his projects (while naming them all some variation of X). He recently took an AI-powered drive in a Tesla to demonstrate that Tesla’s self-driving capabilities are almost ready for introduction. He also announced Tesla’s new AI data center, as one of the world’s most powerful AI facilities.

He also recently changed the terms of service on X ( the social media app, not the AI platform, the spaceship, the car or the son) to allow for the collection of user data to train AI models. He announced his AI platform xAI back in July 2023, with the stated goal of the platform to “Understand the true nature of the universe.” No pressure. And evidently part of that “understanding of the true nature of the universe” is going to come from using all of the activity on X-Twitter to train xAI.

As one of the original founders of Open AI, Elon Musk has had his hands in AI for a while. After the Open AI board declined Musk’s offer to run the organization (which was non-profit at the time), Musk resigned from the board and focused his AI efforts on Tesla and now xAI,

While an army of IP lawyers are trying to sort out the mess that AI has made out of copyright protection, many people are trying to figure out how to protect their IP that is already on the internet from being scraped, scanned and rolled up into training data for LLMs. If you are3 one of these people. here are a couple of tools to help you out:

Opt-out of Open AI: Open AI is doing its best to get ahead of being regulated by giving website operators the ability to opt out of its GPTBot web crawler indexing their sites on their site’s Robots.txt file or block its IP address. Although this does nothing to undo content that has already been scraped and scanned, it is supposed to protect your content going forward. We would still love to see some third party verification of the effectiveness of this solution, but it is a positive step forward and it steps a benchmark that may become a standard around all the top LLMs.

Glaze Your Images! Some very smart researchers at University of Chicago developed an app called Glaze that cloaks your images from multiple AI models such as Midjourney, Stable Diffusion, etc. Glaze is a tool designed to cloak images to prevent AI from stealing the style of the artwork. Each cloak is uniquely tailored to a given image, and once applied, it protects the artwork from various AI models - a property termed 'transferability'. This cloak is robust against removal attempts, including sharpening, blurring, or even stripping of metadata. The strength of the cloak can be modified, where greater alterations lead to more robust protection against AI theft of style.

AI Weakness and Image Cloaking Despite AI's advanced capabilities, it has consistently demonstrated a vulnerability: its inability to approximate human vision. This shortcoming is illustrated by examples where minor modifications to input can significantly confuse an AI model's classification. This phenomenon was observed as early as 2014, indicating a persistent gap in human and AI perception. Glaze utilizes this AI weakness in its cloaking technology. The alterations made by Glaze to the images might not be discernible to human eyes but can severely distort AI perception. Current AI architectures would need significant revamps to overcome this inherent flaw. It's also worth noting that while some have tried, Glaze hasn't been successfully bypassed since its release.

Application, Limitations, and Future Plans Glaze is not foolproof against all types of attacks. Currently, Glaze is not available for mobile devices due to its high computational demand, but a web service version is in development for artists without desktop access. As for artists with existing uncloaked artwork online, using Glaze will gradually shift their style in AI's feature space, causing the AI to misinterpret their artistic style over time.

Check out the Windows and Mac versions of this free app at https://glaze.cs.uchicago.edu/