OpenAI's $1 Billion ARR, Google's Watermark, & CAPTCHA Cracked
We have an update on OpenAI's revenue, two updates on Google, and god willing we may hopefully see the last of CAPTCHA.
OpenAI ChatGPTs Its Way to $1 Billion ARR
OpenAI, the creator of ChatGPT, is reportedly on track to generate over $1 billion in annual sales, much faster than its own projections, thanks to the skyrocketing demand for AI technology. The company, which makes money through licensing its technology to corporate clients and individual subscriptions, is currently generating over $80 million per month in revenues, a significant increase from $28 million for all of last year.
Despite this, OpenAI lost $540 million last year developing GPT-4 and won't be able to fully capitalize on its success for some time due to the terms of its deal with Microsoft. However, the popularity of ChatGPT, which remains the most visited AI-powered chatbot site by a wide range, could push OpenAI revenues well ahead of projections for the foreseeable future.
Google's SynthID: A Stealthy Sentinel in the War Against Deepfakes
Google DeepMind has developed a new tool called SynthID, designed to watermark AI-generated images in a way that is imperceptible to the human eye but easily detected by AI tools. This development comes as part of a broader effort to combat the proliferation of deepfakes and other AI-generated content that could be misused, especially with another election season on the horizon.
While the watermark does not alter the image's quality or experience, it is robust to various transformations like cropping and resizing. Although SynthID is still in its beta phase and is not a silver bullet to the deepfake problem, Google hopes it could become an internet-wide standard and be applied to other media like video and text.
Ideogram AI Cracks the Text Rendering Conundrum: A Game-Changer for Video Editors?
New kid on the block, Ideogram AI, founded by ex-Google Brain researchers and recently launched with a cool $16.5 million in seed funding, claims to have cracked the code on text rendering issues that have been a thorn in the side of AI apps. This has been a thorn on the side of image generators like Midjourney, which created an “E” from the prompt “the letter F”.
Hey, at least it looks like a letter.
Ideogram is still in beta and has some problems, like struggling with spelling, despite being fed the text. Their web app is showing promise in the text-to-image AI space, particularly for video editors and motion graphic artists who spend hours rendering text or logo animation. While it's currently focused on single-image solutions, the leap from text-to-image to text-to-video doesn't seem too far off.
Google's Duet AI Is Now Available in all Workspace Apps
Google is rolling out its Duet AI assistant across all Workspace apps, pitching it as a helpful collaborator and creative tool. Duet can turn Google Docs outlines into Slides decks, make charts from spreadsheet data, generate email responses, images, check grammar, and more.
It's a mix of Clippy's helpful demeanor and ChatGPT's creative abilities, with app-specific features like AI-based lighting and sound tweaks in Google Meet, and automatic thread summaries in Chat. However, it comes with a $30 per user price tag for large organizations, similar to Microsoft's Copilot.
Duet will be integrated into practically all Workspace apps, and given Google's history, it will be hard to ignore, even if you want to.
CAPTCHA Cracked by AI: Are Bots More Human, or Are We All Robots?
CAPTCHAs, those annoying puzzles that make you identify traffic lights and buses to prove you're human, are probably useless, thanks to AI bots that can mimic human brain and vision so well that they outperform us in solving them.
A recent study, yet to be peer-reviewed, shows that bots' accuracy in decoding CAPTCHAs ranged from 85-100%, substantially exceeding the human accuracy range of 50-85%, and even solving them faster in most cases. This raises questions about the colossal global effort invested in solving CAPTCHAs daily and whether it's actually worthwhile, especially given the built-in ceiling of CAPTCHA difficulty that prevents making them too hard for humans.