Exciting AI News: GPT 4.5 Leak, Winter Performance, and More!

 GPT 4.5: The Most Advanced Model Yet

A recent leak on Reddit suggests that OpenAI's next version of the GPT model, GPT 4.5, is in the works. While this leak has not been confirmed, the leaked screenshot reveals some exciting features of GPT 4.5, including audio, vision, video, and 3D capabilities, as well as complex reasoning and cross-modal understanding. The pricing list in the screenshot indicates different versions of GPT 4.5, such as GPT 4.5 64k and GPT 4.5 audio and speech. If this leak is accurate, it means that GPT 4.5 will bring these advanced features to chat GPT as well.

What's interesting is the pricing comparison between GPT 4 Turbo, released in November, and GPT 4.5. The leaked pricing for GPT 4.5 appears to be significantly higher, with a six-fold increase in cost per input and an 18-fold increase in cost per output. Additionally, the context window for GPT 4.5 is half the size of GPT 4 Turbo, which is quite surprising. It's important to note that this is just a leak and has not been officially confirmed by OpenAI.

AI Winter Break Hypothesis

There's an intriguing theory circulating that GPT 4 might perform worse during the winter months. The hypothesis suggests that since GPT 4 was trained on data from the internet, which shows a decrease in activity during December, the model might reflect this behavior and be "lazier" in its responses. An experiment conducted by Ethan Mullik and shared on Reddit supports this theory. By changing the system prompt to indicate it was May instead of December, GPT 4 performed better and provided longer responses. While this theory is speculative, it highlights the potential influence of training data on AI models.

Partnership with Axel Springer

OpenAI has formed a partnership with Axel Springer, a media company that owns well-known news outlets such as Politico, Business Insider, and Bild. The partnership aims to deepen the beneficial use of AI in journalism. Chat GPT will use content from Axel Springer's media properties to provide updated information on current events and news. This collaboration marks an interesting turn of events, considering that back in July, Axel Springer was planning to sue OpenAI for using their content without permission. Now, they are working together to enhance the capabilities of AI in journalism.

Chat GPT Plus Subscriptions Re-Enabled

Sam Alman and the OpenAI team have recently re-enabled Chat GPT Plus subscriptions. This means that individuals who have been waiting to become Chat GPT Plus members can now sign up. Chat GPT Plus offers additional features and benefits for subscribers. If you've been wanting to unlock the full potential of Chat GPT, this is your chance!

Google's Gemini Demo Controversy

In a previous news video, we discussed the controversy surrounding Google's Gemini demo, which was criticized for misleading viewers about the capabilities of the model. However, Greg Technology recently demonstrated that similar results can be achieved using OpenAI's GPT Vision model. By showing various prompts and receiving accurate responses, Greg's demo proves that GPT Vision can perform many of the same tasks showcased in Google's demo. Google faced backlash for their misleading video, but they responded by releasing new AI tools, such as Notebook LM.

Notebook LM: A Handy Research Tool

Google's Notebook LM is a useful tool for researchers and students. It allows users to upload files, such as PDFs and Google Docs, and use the search box to retrieve specific information from these documents. The tool also offers suggested actions, such as summarizing notes, adding quotes, suggesting related ideas, and creating outlines. While similar tools like Claude and chat GPT can perform these tasks, Notebook LM is purpose-built for research purposes and provides a more efficient experience.

Gemini Pro API and Music FX

Google has recently launched the Gemini Pro API, which allows developers to start developing with this new API. The Gemini Pro pricing appears to be more affordable compared to GPT 4 pricing, making it an attractive option for developers. Additionally, Google released Music FX, a tool that generates music based on user inputs. Users can specify the genre and other parameters to create custom music. The flexibility and quality of the generated music make Music FX a valuable tool for content creators.

Image In2, Stable AI, and Meta's Audio Box

Vertex AI customers can now access Image In2 from Google, which offers improved text rendering, logo generation, and visual question-answering capabilities. This model enhances the overall experience of working with images and provides more accurate and informative responses.

Stable AI has released its stable 0123 model, which allows users to turn 2D images into 3D representations. The model shows promise in generating detailed and realistic 3D images from 2D inputs. While there are some limitations and occasional quirks, it is a significant step forward in the field of AI-generated 3D models.

Meta has introduced its Audio Box, a powerful tool with features like voice training, voice generation, restyling voices, adding sound effects, erasing audio elements, and infilling and replacing audio segments. This tool opens up new possibilities for voice-related applications and creative audio projects.

Wirestock's Themes for Custom AI Art

Wirestock, a platform for selling stock images, has rolled out a new feature called Themes. Premium users can train custom AI styles using their own images and earn income when others use their themes. This feature allows AI art enthusiasts to monetize their creations and contribute to the artistic community. Wirestock continues to innovate in the field of stock photography and AI-generated images.

Mid Journey Update and Mixol of Experts

Mid Journey, an AI art generator, has made significant updates to its platform. They now offer a website interface for creating AI-generated images, making it more accessible to users. Additionally, Mid Journey version 6 is on the horizon, with users participating in the fine-tuning process through rating parties. This collaborative approach ensures that Mid Journey meets the expectations and preferences of its user base.

Mistol AI, a French company, has introduced Mixol of Experts, an open-source large language model that outperforms LAMA 2 and GPT 3.5 in various tasks. Mixol of Experts utilizes a unique approach, employing smaller models with specific areas of expertise. The models work together through a routing mechanism to provide accurate and efficient responses. This approach represents a potential future direction for large language models.

Updates from Instagram, Snapchat, and Tesla

Instagram has rolled out its generative AI-powered background editing feature, allowing users to remove the background from their images. Similarly, Snapchat now offers an AI Extend feature that extends the content of an image by filling in the gaps. These features provide users with more creative options for editing their photos.

Tesla has unveiled its Optimus Gen 2 humanoid robot, showcasing its improved mobility, balance, and dexterity. The robot's smooth movements and advanced capabilities make it an impressive development in the field of robotics.

These are just a few of the exciting AI developments happening in the world right now. Stay informed and stay tuned for more updates in the fascinating world of artificial intelligence!