AI News: The ChatBot That Surpasses ChatGPT in Benchmark Tests

According to the crowdsourced platform Chatbot Arena, the premium model of Claude, called Claude 3 Opus, has officially become the top-performing large language model, surpassing the long-reigning ChatGPT.

Claude 3 Opus, which costs $20 per month, has been voted as the best performing model by real human users on the platform. This dethroning of ChatGPT marks a significant shift in the AI language model landscape.

Furthermore, Amazon has announced that it will invest an additional $2.75 billion into Anthropic, the company behind the Claude platform. This investment solidifies Amazon's interest in the Claude model and its potential.

While Claude may not offer image generation or web search capabilities like ChatGPT, it has been shown to outperform ChatGPT in various other use cases. Users are encouraged to try out Claude and experience its capabilities firsthand.

Amazon has announced that they will invest an additional $2.75 billion in Anthropic, the company behind the popular AI chatbot Claude. This investment comes after Amazon's initial $1.25 billion investment, and will give Amazon a minority stake in Anthropic. However, Amazon will not have a seat on Anthropic's board.

This deal is similar to the one Microsoft has with OpenAI, as Amazon looks to secure a partnership with a leading AI company and its flagship product. The investment will likely help Anthropic continue to develop and improve the Claude platform, which has recently surpassed GPT-4 as the top-performing chatbot according to user votes on the Chatbot Arena platform.

The additional funding from Amazon will allow Anthropic to further scale its operations and continue innovating in the field of large language models and conversational AI. This partnership could also lead to deeper integrations between the Claude platform and Amazon's own AI and cloud computing services in the future.

Databricks, a leading data and AI company, has recently released a new large language model called dBrx (or D-bricks). According to the benchmark tests and the company's blog post, this new model outperforms every other open-source model in various tasks.

In language understanding, dBrx beat out LLaMA 2, Mixol, and Grok 1. It also outperformed these models in programming and math tasks. Furthermore, the new dBrx model was able to beat the free version of GPT-3.5 in every single benchmark test.

The weights for the dBrx model are available on Hugging Face, but the company requires users to apply and get approval before they can download the model. This new open-source model from Databricks seems to be a strong contender in the large language model landscape, offering impressive performance across a range of tasks.

There was some big news in the world of Open Source this week over at Stability AI. Ahmad Mustak, the CEO of Stability AI, seemingly abruptly stepped down from his role as CEO and relinquished his board position. The speculation is that he was kind of forced out, as opposed to actually stepping down, but the outward-facing information around the whole thing made it look like it was an amicable decision.

Stability AI has had a tumultuous ride recently, with most of the key employees and engineers actually leaving the company over the last several months. In fact, it's unclear if anybody who originally worked on Stable Diffusion still works at Stability AI, as they have all left in the last couple of months. There have also been a lot of rumors about Stability AI possibly filing for bankruptcy and having financial issues.

Ahmad's departure seems to be a confirmation of these issues at Stability AI. In his statement, Ahmad said that his shares have the majority of the vote at Stability AI and that they have full board control. He claimed that the concentration of power in AI is bad for us all, and that he decided to step down to fix this at Stability AI and elsewhere. He also mentioned that he is "all in on decentralized AI," but it's unclear what his plans are or what he means by decentralized AI.

Despite Ahmad's departure, Stability AI didn't slow down. The very next day after he stepped down, they announced Stable Code Instruct 3B, a new coding large language model, seemingly to show that the company will keep shipping even without Ahmad at the helm.

Open AI is now starting to monetize the GPT store. Previously, users could create custom GPT models and make them available in the GPT store, but there was no way for them to earn income from the usage of their models.

This week, Open AI announced a pilot program to partner with builders and allow them to earn based on the usage of their GPT models. Builders who have highly utilized GPT models in the store are being invited to participate in this program.

One example is the "Automated Writer" tool, whose creator Matt Schit received an email from Open AI stating that his GPT model is highly utilized and they would like to incorporate his feedback as they define the approach to GPT monetization.

Another creator, Nick Doos, who built the popular "Gamer GPT" model, was informed that he could earn a minimum of $11,000 USD monthly from the usage of his GPT, with potential for additional earnings.

This new initiative from Open AI will start on May 1, 2024, with the first payout scheduled for mid-June. It provides an opportunity for creators to monetize their custom GPT models and earn from the usage by ChatGPT users.

There's a rumor circulating that OpenAI is about to remove the quotas in GPT-4. A leaked image shows a "Smart Model Selector" that might be coming soon. This feature will essentially decide whether to use GPT-3.5 or GPT-4 when a user asks a question. If the query doesn't require the full capabilities of GPT-4, it will use the less expensive GPT-3.5 model. But if the query needs a more complex response, it will use GPT-4. This is how OpenAI seems to be planning to get rid of the usage limits - by dynamically switching between the most optimal model for each query.

We also got a peek at some new Sora videos this week. Sora was made available to a handful of creators and while I'm not going to show you all of the videos on here that were made by these creators, they are super super impressive. One of them even comes from somebody who's a buddy of mine Don Allen Stevenson, so really cool to see that he got some Early Access to Sora and was able to make videos with these crazy hybrid creatures like a flying pig and a flamingo giraffe hybrid and whale squids or whale octopuses. Really really cool stuff.

One of the standout videos that a lot of people have been circulating is this one called Airhead by shy kids about this person who has a balloon for their head and it is some actually really good storytelling. Again, I don't want to play this video here, I don't know the sort of copyright on these videos, but I will link this post up in the description below so you can check out some of the new videos that have been circulating from Sora. Again, just like last time, they are really dang impressive videos.

And while we're on the topic of Sora, apparently open AI has been talking to the film industry, talking to people in Hollywood, trying to get them on board with the Sora of software. Obviously, Hollywood is a segment of the world who has been very very fearful of AI because it threatens voice actors, it threatens actors, it threatens animators, it threatens script writers. There's a lot of roles in Hollywood that AI kind of threatens right now, and open AI is sort of on tour trying to get the people in Hollywood to come around to accepting and leveraging AI as a tool that will make their lives easier, lower budgets on videos, and ideally not replace jobs, but to help make the people that are doing the jobs much more efficient.

Hume is a new chatbot that claims to be the first to have true empathy. Unlike typical language models, Hume is built on deep insights from emotion science, allowing it to understand and respond to human expressions, not just words.

Hume is designed for voice interactions, using the user's tone of voice to detect their emotions and tailor its responses accordingly. If it detects frustration, it will try to understand and address the user's concerns. If it senses happiness or silliness, it will respond in a more playful manner.

The mission of Hume is to align AI with human well-being. By focusing on emotional intelligence, the chatbot aims to have more natural and empathetic conversations, providing a more positive experience for users.

While the current voice demo may not be entirely realistic, the team at Hume is likely prioritizing the underlying research and functionality over polished audio. As the technology matures, the voices will likely become more natural and lifelike.

Overall, Hume represents an intriguing step forward in the development of chatbots that can truly understand and respond to human emotions, rather than just processing language. It will be interesting to see how this technology evolves and is applied in various use cases.

Adobe announced the launch of Adobe Gen Studio, a new platform that serves as a central hub for creative design work, including the use of AI. Some key features of Gen Studio include:

• Brands Tab: Allows you to manage your brand assets like logos, colors, fonts, images, and copy in one place.
• Firefly Integration: Enables you to generate images directly within the platform using Adobe's Firefly AI model.
• Campaign Management: Provides tools to build and manage campaigns across various social media platforms.
• Analytics and Reporting: Includes data and stats to track the performance of your creative campaigns.

The platform aims to offer a comprehensive solution for managing all your creative design needs, seamlessly integrating AI capabilities like Firefly for image generation. This centralizes your brand assets, creative workflows, and performance tracking in a single interface.

While Gen Studio is not yet widely available, the announcement suggests Adobe's continued focus on empowering creatives with AI-powered tools and workflows. As the platform rolls out more broadly, it will be interesting to see how it is adopted and utilized by designers and marketers.

Jen rolled out a awesome new feature called Habitar In Motion 1.0. This feature allows you to move your hands, make gestures, and use unique tones of voice, and it will flawlessly track, translate, and lip sync your video with any input text. The demo videos look impressive, showing a person walking around, moving their hands, while the audio is dubbed over and the lips still sync to the new audio.

LTX Studio also had their launch party this week, and they are now starting to give access to people on the waitlist to try out the beta version. LTX Studio is a tool where you can enter a single prompt, and it builds out the entire storyboard from that prompt. It can maintain consistent characters throughout the scenes, making it a powerful all-in-one video creation platform. The creator is excited to dive in and put LTX Studio through its paces, and plans to release a video about it on Monday.

Researchers at MIT have introduced a new image generation method called "distribution matching distillation" that can generate images 30 times faster than stable diffusion while maintaining or surpassing the quality level of outputs.

The example shows how quickly the new model generates the same four images compared to stable diffusion. The new model appears to be able to rapidly generate high-quality images, which could be leveraged by various AI image platforms in the near future.

This advancement in image generation speed, while maintaining quality, is an exciting development that could significantly impact the field of AI-powered image creation.

This week, Leonardo AI introduced a new feature that allows users to generate images much faster using the new "Leonardo Lightning XL" model. This model can generate images 30 times faster than Stable Diffusion, while maintaining similar quality levels.

To access this feature, users can go to the "Image Generation" section, click the "Fine-tune Model" dropdown, and select the "Leonardo Lightning XL" option. This will allow them to generate images quickly, with the example prompt "lightning" taking only around 10 seconds to produce four images.

In addition to the new lightning-fast generation model, Leonardo AI also rolled out a background removal feature last week. Users can now click the "Remove Background" button on any image, and the tool will generate a version of the image with the background removed, which can then be downloaded as a PNG file.

These new features further enhance the capabilities of the Leonardo AI platform, providing users with faster image generation and the ability to easily remove backgrounds from their generated images.

There's a cool new 3D design app called Bezi AI that has been circulating. It appears to be another text-to-3D object generator that creates game-ready and movie-ready assets.

When you sign up, you get a free asset bundle made available in your library. The app has an asset gallery of existing 3D assets that it has created.

To create a new file, you get a 3D plane to work with. The interface feels similar to working in tools like Blender, with a node-based system where you can add "States". It seems like something that would require some time to learn and get familiar with.

The key feature is the "Bezi AI" button, which allows you to generate 3D objects from text prompts. For example, when I entered "X-Wing spaceship", it created an X-Wing model that I could then drag and manipulate in the 3D scene.

The app is free to use right now, and you can find it at bezi.com. It's an interesting new tool that combines text-to-3D generation with a 3D design interface, allowing creators to quickly generate and work with 3D assets.

Quellcop, a popular gaming YouTuber, has created what he calls "Quellcop AI" - an AI agent playing Minecraft with an AI version of Quellcop providing the voice commentary. While the jokes and commentary are still a bit "cheesy," as Quellcop himself has acknowledged, the progress made in this AI-powered content creation is impressive.

The video shows the AI agent navigating through the Minecraft world, interacting with the environment, and the AI-generated Quellcop voice providing commentary. Despite the room for improvement, the concept of an AI agent playing a game while an AI version of the creator provides the commentary is a fascinating development in the world of automated content creation.

Quellcop has been working on this project for some time, and the advancements made in the past few months are noteworthy. As the technology continues to evolve, the quality and seamlessness of the AI-generated content are expected to improve, making it an intriguing area to watch in the future.

The world of AI is rapidly evolving, with new developments and advancements happening at a breakneck pace. From the rise of new language models like Claude 3 Opus and dbrx, to the shakeups at Stability AI, the open-source AI landscape is in a constant state of flux.

Meanwhile, OpenAI is making strides in monetizing the GPT ecosystem, with plans to share earnings with creators of custom GPT models. The rumored removal of quotas in GPT-4 and the introduction of a smart model selector also point to OpenAI's efforts to make their language models more accessible and efficient.

The emergence of tools like Hume, which aims to bring emotional intelligence to chatbots, and Adobe's new Gen Studio, a centralized hub for creative work with AI, further showcase the diverse range of AI-powered applications being developed.

Additionally, advancements in image generation, such as the distribution matching distillation technique from MIT and the lightning-fast image generation in Leonardo AI, are pushing the boundaries of what's possible with AI-powered visuals.

As the AI landscape continues to evolve, it's clear that the pace of innovation shows no signs of slowing down. Staying informed and keeping up with the latest developments in this rapidly changing field is crucial for anyone interested in or working with AI.