What Do the GPT 4.5 Possible Leak, Midjourney V6, and Opensource LLMs That Beat Open AI Reveal? | AI News

TL;DR

The latest AI news points to broader multimodal models, cheaper APIs, competitive open-source LLMs, and stronger image, video, and 3D generation tools. The possible GPT 4.5 leak describes language, audio, vision, video, and 3D capabilities, while Google offers 60 free Gemini Pro API queries per minute. Read on for the reported pricing, capabilities, comparisons, and important caveats.

Transcript

buckle up guys because we have a lot to talk about today AI has legitimately been exploding lately faster than I've ever seen it grow before and today's AI news video is no exception so let's jump right into it starting off with a leak from openai about GPT 4.5 so here we are on Reddit this is where the leak was originally posted the openai Reddit ... Read More

Key Insights

• 🎮 OpenAI's GPT 4.5 leak suggests significant advancements in multimodal capabilities, including language, audio, video, and 3D understanding.
• 🛄 Google's Gemini Pro API introduces competitive pricing, aiming to gain market share by outpricing OpenAI.
• 🛩️ Microsoft's F-2 small language model demonstrates the potential for compact models to compete with larger models in performance.
• 🔂 Mistl AI's Stable Zero provides impressive 3D object generation from single images, further advancing AI capabilities in the 3D domain.
• 🎮 Pika Labs 1.0 showcases remarkable video in painting and modification features, allowing for easy editing and manipulation of videos.
• 😯 AI advancements in text-to-image generation, music composition, and text-to-speech capabilities continue to elevate the quality and realism of AI-generated content.
• 🤗 Open-source models, such as Mixl 8X 7B, offer competitive alternatives to proprietary models, driving innovation and affordability in the AI landscape.

Questions & Answers

Q: What does the possible GPT 4.5 leak claim?

The leak describes GPT 4.5 as a multimodal model spanning language, audio, vision, video, and 3D, with complex reasoning and cross-modal understanding. The presenter repeatedly warns that OpenAI had not officially confirmed the material and that it should be treated with caution.

Q: What pricing was shown in the possible GPT 4.5 leak?

The regular GPT 4.5 chat model was shown at 6 cents per 1,000 input tokens and 18 cents per 1,000 output tokens. A 64k context version was listed at 12 cents per 1,000 input tokens and 36 cents per 1,000 output tokens.

Q: What could GPT 4.5 audio and speech capabilities mean?

The presenter suggests that separate audio and speech support could extend beyond speech-to-text. It might interpret nuanced sounds, such as a train horn or glass breaking, but this is speculation based on the unconfirmed leak.

Q: How does the Gemini Pro API compete with OpenAI?

Google's Gemini Pro API reportedly allows 60 API queries per minute for free. The presenter describes this as highly competitive pricing intended to gain market share by underpricing OpenAI.

Q: What was announced about Midjourney V6?

Midjourney announced the upcoming release of its V6 model. The existing page summary says it was expected to compete with DALL·E 3.

Q: How does Microsoft's F-2 small language model compare with larger models?

The existing page fields describe Microsoft's F-2 as a small language model that competes with substantially larger models. It is presented as evidence that compact models can deliver competitive performance.

Q: What does Mistral AI's Stable Zero generate?

Stable Zero is described as a model that generates high-quality 3D objects from single images. It represents the video's reported progress in AI-assisted 3D generation.

Q: What capabilities does Pika Labs 1.0 demonstrate?

Pika Labs 1.0 demonstrates video generation and video inpainting for modifying selected regions or elements. The existing page fields also highlight AI-powered face and object swapping.

Summary & Key Takeaways

• Definition: The possible GPT 4.5 leak describes one model spanning language, audio, vision, video, and 3D.

• Number: Regular GPT 4.5 pricing was shown as 6 cents per 1,000 input tokens and 18 cents per 1,000 output tokens.

• Number: The leaked 64k model pricing was 12 cents per 1,000 input tokens and 36 cents per 1,000 output tokens.

• Compare: The leaked 64k context window is smaller than GPT-4 Turbo's stated 128,000-token context window.

• Number: Gemini Pro reportedly provides 60 API queries per minute for free.

• Who: Google is portrayed as using competitive Gemini Pro pricing to gain market share from OpenAI.

• Tool: Google's text-to-image model generates high-quality, photorealistic images from prompts.

• Compare: Midjourney's upcoming model is expected to compete with DALL·E 3.

• Tool: Stable Zero generates high-quality 3D objects from single images.

• Compare: Mistral AI's open-source model is presented as a competitive, affordable alternative to proprietary models.

• Tool: Pika Labs supports video generation, inpainting, and AI-powered face and object swapping.

• Who: Microsoft demonstrates that a small language model can compete with substantially larger models.