Good morning. It’s Monday, March 20th.

GPT-4 has been out for less than one week, and is currently throttled at 25 messages every 3 hours for paid users of OpenAI’s ChatGPT Plus.

In today’s email:

• Viper GPT

• GPT-4’s 25k word output comes into question

• Real-time voice translation

• Introducing your employees to AI

• Creating custom avatars

You read. We listen. Share your feedback by replying to this email, or DM us on Twitter.

ViperGPT

Researchers out of Colombia University have introduced ViperGPT, which aims to improve the way computers understand and answer questions about images.

Currently, most systems used for this task, called end-to-end models, have limitations in their ability to interpret and generalize information from images. ViperGPT offers a promising alternative by breaking down the task into smaller, manageable parts.

In simple terms, ViperGPT works by combining smaller programs, like building blocks, to create a customized solution for a specific question about an image.

It does this by generating Python code to execute the right combination of these smaller programs. This approach does not require additional training and performs exceptionally well on a wide range of visual tasks.

By combining the strengths of ViperGPT and GPT-4, we can expect even more powerful AI tools for image interpretation and analysis. For instance, a system utilizing both ViperGPT and GPT-4 could efficiently analyze images, diagrams, or screenshots, while generating accurate and relevant textual information in response to queries.

This could significantly improve the way we interact with AI-powered applications in various industries, such as education, healthcare, and accessibility services for the visually impaired.

Read more: ViperGPT demo, research paper

Can GPT-4 really produce 25k words?

The capabilities of GPT-4 are undeniably impressive. However, despite the impressive claims made by OpenAI, concerns have been raised about the maximum output volume of GPT-4, particularly in relation to its advertised 25,000-word processing capacity.

OpenAI claims that GPT-4 has a token limit of roughly 24,000 words, or 32,000 tokens. Though since the model’s release on Tuesday, users have reported difficulties when attempting to process large text segments within this limit, despite having a ChatGPT Plus subscription.

These difficulties appear to stem from the fact that the token limit accounts for both input and output tokens, requiring users to carefully manage their text submissions to avoid exceeding the limit.

Some have argued that the current token limit available for ChatGPT Plus users is closer to 4,000 words, significantly below the advertised capacity.

Currently, the ChatGPT interface lacks an intuitive means of informing users about the limitations of their text submissions, forcing them to rely on trial and error to determine the appropriate length.

This problem is exacerbated by the difficulty in determining the actual token count of a given text, which includes spaces and varies depending on the language used.

To address the limitations and user experience challenges, several solutions have been proposed within the GPT-4 user community.

• Breaking texts into smaller segments could help avoid exceeding token limits, albeit at the cost of increased complexity for users.

• Ensuring the correct model version is used in the API call might mitigate some of the token limit issues.

• Improving the user experience by providing a text analysis feature or an intuitive guidance system could enable users to better understand the processing capacity of GPT-4 and avoid errors.

Read more: OpenAI Community Forums

Opinion: The Advent of Real-time Translation Technology

As global interconnectedness increases, the need for seamless communication across linguistic barriers becomes ever more critical. Emerging technology, leveraging existing advances in natural language processing and speech generation, could revolutionize cross-cultural communication and international relations.

Drawing inspiration from the fictional "Babel fish" translator in The Hitchhiker’s Guide to the Galaxy, this theoretical technology would instantaneously convert spoken language into the listener's preferred language through an unobtrusive wearable device, such as Bluetooth headphones.

The technology could integrate state-of-the-art language models like GPT-4, renowned for its ability to capture the semantic meaning behind sentences, rather than relying on traditional word-for-word translations. Coupling this with sophisticated text-to-speech platforms, such as those developed by Elevenlabs.io, could yield an innovative real-time translation system.

The potential implications of this technology are far-reaching, spanning various aspects of society. For instance, international business could flourish as language barriers dissolve, enabling smoother transactions and collaboration between global partners. Additionally, retirees seeking a lower cost of living might venture abroad, no longer daunted by the prospect of learning a new language. International tourism could also experience a surge, as travelers would be empowered to explore unfamiliar destinations without the fear of linguistic isolation.

Moreover, the technology could foster empathy between nations by enabling citizens to comprehend speeches from foreign leaders in their native languages, delivered with a natural-sounding voice. This enhanced understanding could cultivate a sense of connection and foster improved international relations.

While the prospect of mass distribution remains a challenge, the ubiquity of cell phones and headphones worldwide suggests that the infrastructure to support such a technology is largely in place. As large language models and text-to-speech technology continue to advance, the realization of this revolutionary real-time translation system could be imminent. Within the next five years, its emergence could have a profound impact on international business, travel, and living, transforming the way we perceive and interact with the world around us.

How to introduce AI to your employees

WNR.ai is a free innovative platform that uses GPT-4 to streamline the document creation process for businesses.

One of WNR.ai's standout features is its ability to transform raw text into polished pitch decks for venture capitalists and investors.

The platform can seamlessly convert unformatted text into an array of professional documents, including employee handbooks, offer letters, and even rejection emails. In seconds, WNR.ai takes your content and crafts it into a specific document using a template tailored to your needs.

In addition to its prowess in document creation, WNR.ai also supports social media content, marketing articles, press releases, and tweet threads.

If your goal is to familiarize your employees with AI without requiring them to establish an OpenAI account or learn the intricacies of prompting, WNR is the tool for you.

Fiction.com is a new platform to make custom-trained AI generated images and avatars. With a dedicated AI server, users can make unlimited revisions to their custom models.

With fiction.com, users can create:

Custom Avatars: Create personalized AI-generated avatars for yourself or friends.

Client Mockups: Train a model that understands a specific design style or concept, and then render various ideas for your clients.

Video Editing: Fiction also includes Stable Diffusion WebUI and other professional AI tools. You can use these to create videos, edit images, and more.

Try it free for 5 days

_{sponsored post}

3x the information, for less than $2/week

Stay informed, stay ahead: Your premium AI resource.

Thank you for reading today’s edition.

Your feedback is valuable.
Respond to this email and tell us how you think we could add more value to this newsletter.

Read by employees from