A mysterious new language model called "GPT-2 chatbot" has appeared on a website called "chatbot Arena", and it seems to be outperforming even GPT-4 and Claude Opus in various tests. The model has a different naming convention compared to other GPT models, and no company has claimed responsibility for its development.

Experts like Ethan Mollick, a professor at Wharton, have been playing with the model and have found it to be on par with GPT-4 in terms of capabilities. Andrew G. also claims that the GPT-2 chatbot was able to solve an International Math Olympiad problem in a single prompt, while Dr. Alvaro Cintas from Rundown was able to use it to code a simple snake game on the first try.

The AI community has been left baffled by this mysterious model, with speculation ranging from it being a new version of GPT-4 to it being a version of GPT-2 trained on new data from GPT-4. However, Sam Altman, the CEO of OpenAI, has confirmed that it is not GPT-4.5.

As of now, the origin and nature of this GPT-2 chatbot remain a mystery, but its impressive performance has certainly caught the attention of the AI community.

Open AI has finally rolled out the memory feature in ChatGPT to all of the Plus users. This feature allows users to save information about themselves and have ChatGPT remember it.

To use the memory feature, you can log into your ChatGPT account and provide information about yourself. ChatGPT will then save this information in your memory. You can view and manage your saved memories in the settings.

The memory feature also includes a "temporary chat" mode, which allows you to use ChatGPT without saving any information. This is useful for maintaining privacy and keeping your conversations incognito.

Overall, the new memory feature adds useful personalization capabilities to ChatGPT Plus, allowing users to customize their interactions with the AI assistant.

There are rumors circulating that OpenAI is getting close to launching its own search engine. According to reports:

• OpenAI's SSL certificate logs now show they created the domain "search.chat.gp.com", hinting at a potential search engine.
• Pete Hangs from Neuron pointed out a quote from Sam Altman on the Lex Fridman podcast, where Altman said "the intersection of large language models plus search, I don't think anyone has cracked that code yet. I would love to go do that, I think that would be cool."
• Pete also speculated that this search engine could launch on May 9th, though the source of this information is unclear.

It's also reported that Microsoft Bing would allegedly power the OpenAI search service. This potential move by OpenAI could pose a serious threat to Google's dominance in the search engine market.

However, details are still scarce, and an official announcement from OpenAI has not been made. The AI community will be closely watching for any further developments around OpenAI's potential foray into the search engine space.

This week, it was reported that Apple and OpenAI have intensified talks about incorporating OpenAI's technology into future smartphones. While Apple has also been in talks with Google about potentially using Gemini, it seems that Apple is exploring all options for powering the AI in upcoming versions of iOS.

Many are speculating that at WWDC this year, Apple may announce a new version of Siri with better AI features. It appears that Apple is still shopping around to determine which AI model they will use for this purpose.

This week, Anthropics' AI assistant Claude rolled out two new features:

1. Team Plan: Claude now has a team plan that allows for collaboration and shared conversations within the Anthropics platform.

2. iOS App: Anthropics has released an iOS app for Claude, allowing users to access the AI assistant on their Apple devices.

The team plan feature enables users to collaborate on conversations and share information within their team or organization. This can be useful for businesses or groups that want to leverage Claude's capabilities in a more coordinated manner.

The new iOS app provides mobile access to Claude, making it easier for users to interact with the AI assistant on-the-go. This expands the accessibility of Claude beyond just web-based interactions.

These updates from Anthropics demonstrate the company's continued efforts to enhance the functionality and usability of their AI assistant. As the AI landscape continues to evolve, features like team collaboration and mobile access can help differentiate Claude and make it more appealing to a wider range of users.

The Biden Administration has established the Artificial Intelligence Safety and Security Board, with 22 initial members. The board includes prominent figures such as Sam Altman (OpenAI CEO), Satya Nadella (Microsoft CEO), and Sundar Pichai (Alphabet CEO).

The goal of the board is to steer the direction of regulations around AI development and deployment. It raises questions about potential conflicts of interest, as the CEOs of major AI companies will be involved in shaping the policies that govern their own products.

The board also includes leaders from other industries, such as the CEOs of Delta Airlines, Northrop Grumman, and Occidental Petroleum, as well as the governor of Maryland and the mayor of Seattle. This diverse representation aims to balance the interests of various stakeholders.

However, some experts have expressed concerns about having the heads of the largest AI companies directly involved in crafting the regulations that will impact their own businesses. There are worries that this could lead to policies that benefit the tech giants and make it harder for smaller players to compete.

Overall, the establishment of this board reflects the growing importance and potential risks of artificial intelligence. Striking the right balance between innovation and safety will be a key challenge as the technology continues to advance rapidly.

GitHub announced the GitHub Co-Pilot Workspace, which appears to be similar to the previously circulated Devon demo. It seems to be an AI agent from GitHub that allows users to tell it what they want to code, and it will pre-plan the necessary files (e.g., JSX, CSS) and generate the code based on those plans.

The workspace enables live previewing of the generated code, and it also supports collaborative work, where multiple people can contribute to the code development within the workspace.

While not yet available, GitHub has a waitlist at the provided URL for those interested in trying out this AI-powered coding tool. It represents an advancement in AI-assisted software development, allowing developers to leverage the capabilities of large language models to streamline their coding workflows.

Marquez Brownley, a prominent tech reviewer, recently published a review of the Rabbit R1, an AI-powered device. In his video titled "barely reviewable", Brownley shared some strong opinions about the state of the product.

Brownley expressed frustration with the trend of companies releasing "half-baked" products and then iterating on them over time, rather than delivering a fully functional product at launch. He argued that consumers should not have to pay full price for products that are not yet complete.

Regarding the Rabbit R1 specifically, Brownley found the device to be "borderline nonfunctional" compared to the promises and features that were initially advertised. He criticized the practice of companies selling products based on future potential, rather than the current capabilities.

While Brownley acknowledged that he is an early adopter and tech enthusiast, he believes that companies should not release products until the promised features are ready. He argued that this "hot take" is a reasonable expectation for consumers.

Overall, Brownley's review highlighted the growing frustration with the trend of companies releasing AI-powered products that do not yet live up to their initial hype and promises. His critique resonated with many in the tech community who share similar concerns about the state of product launches in the AI space.

If you're sick of using mid-journey in the Discord, they finally made it available on their website for most people. If you've generated at least 100 images with mid-journey, then you can now use the mid-journey website to generate your images.

To get started, head over to Mid-journey.com, sign into your account, and you'll see an "Imagine" box at the top where you can type your prompt. You also have a settings box where you can set your aspect ratio, stylization, weirdness, variety, standard mode, raw mode, which version to use, and the speed.

Once you generate an image, you have all sorts of features on the sidebar to edit the image, such as Var, upscale, remix, pan, and zoom. You can now do all of this from the website without needing Discord anymore.

If you haven't generated 100 images yet, go make a bunch real quick, and then you'll get access to the tool. The founder actually pronounces it "Udio", though everyone else has been calling it "Udio", which is a bit more fun to say.

China has unveiled a new humanoid robot called Tiangong. This advanced robot is the latest development in the field of humanoid robotics.

Some key details about Tiangong:

• Tiangong, which means "Heavenly Palace" in Chinese, is a state-of-the-art humanoid robot developed by Chinese researchers.
• The robot is designed to mimic human movements and behaviors, with a highly articulated body and advanced sensors and control systems.
• Tiangong is capable of performing a wide range of tasks, from physical labor to social interaction, showcasing the rapid progress in humanoid robotics.
• The unveiling of Tiangong demonstrates China's growing capabilities in the field of advanced robotics and artificial intelligence.
• As humanoid robots continue to evolve, Tiangong represents an important milestone in the development of this technology, with potential applications in various industries and sectors.

DARPA has released a video showcasing their new autonomous vehicle called 'Racer' (Robotic Autonomy in Complex Environments with Resiliency). This vehicle is an incredibly fast-moving tank-like platform operated entirely by AI.

The video demonstrates the vehicle's impressive speed and maneuverability, controlled solely by advanced autonomous systems without any human intervention. This raises questions about the future of warfare, as countries may increasingly rely on autonomous robot systems to fight their battles rather than human soldiers.

The rapid advancements in AI-powered autonomous vehicles like Racer suggest that the future of military technology could shift towards robotic platforms engaging in highly sophisticated "robot vs robot" combat scenarios. This DARPA project provides a glimpse into how AI may radically transform the nature of modern warfare in the years to come.

This AI-powered camera system is designed to detect and deter intruders on your property. Using sophisticated computer vision technology, the camera can identify human faces and animals, even in low light conditions. It can distinguish between authorized individuals and potential threats.

If the camera detects an unauthorized person, it will autonomously fire paintball or tear gas projectiles at the intruder, giving them an unpleasant surprise they won't soon forget. The system uses a mobile app interface where you can categorize people as "friends" or "foes" to control who gets targeted.

While this technology may seem extreme, it highlights the potential for AI-powered security systems to actively defend private property. However, the legal and ethical implications of such autonomous defensive measures remain unclear and will likely require new legislation and precedents to be established.

A teacher was arrested this past week for using voice AI technology to create racist rants impersonating another teacher. The teacher who framed the other teacher used a service like 11 Labs to train the other teacher's voice and then generate the offensive content.

This incident highlights the legal gray area around the use of AI-generated content, as there is currently little precedent for how these cases will be handled. The laws and regulations surrounding the misuse of voice AI are still unclear, and this case is likely to set an important precedent.

The arrest raises questions about the potential for abuse of these technologies and the need for stronger safeguards and guidelines to prevent such misuse. As AI capabilities continue to advance, it will be crucial for policymakers and the legal system to address these emerging challenges and ensure that these powerful tools are not exploited for harmful or unethical purposes.

This past week, a video surfaced on Instagram showing a wedding couple using Photoshop's generative AI capabilities to try and remove a rock from their image. However, instead of replacing the rock, the AI generated a baby in its place.

This bizarre outcome highlights the unpredictable nature of AI-powered image generation. While the technology has made impressive strides in creating realistic-looking images, it can still produce unexpected and nonsensical results, especially when tasked with complex edits or manipulations.

The video serves as a reminder that AI systems, no matter how advanced, are not infallible. They can make mistakes, introduce unintended elements, and struggle with certain types of image processing tasks. As the use of generative AI becomes more widespread, it will be important for users to maintain a healthy skepticism and carefully review the outputs before relying on them.

This incident also raises interesting questions about the ethical implications of using AI for image manipulation, particularly in sensitive contexts like wedding photography. As the technology continues to evolve, there will likely be ongoing discussions about the appropriate boundaries and safeguards for its use.

Overall, the "AI generates a baby instead of a rock" video serves as a humorous yet thought-provoking example of the current limitations and unpredictability of AI-powered image generation. It underscores the need for continued research, development, and responsible deployment of these powerful technologies.