Google Bard is Now Gemini, EU Approves AI Regulations, Hugging Face Competes with Custom GPTs, & ChatGPT's System Prompt
Google launched new AI offerings under the Gemini brand, the EU endorsed their AI regulations, Hugging Face launched assistants, and ChatGPT's system prompt could be degrading the product.
Google Bard is Now Gemini, New AI App & Paid Subscription Announced
Google is streamlining their AI offerings under a single brand: Gemini. Here's the gist:
Announced Bard's chatbot is now Gemini.
Introduced a Gemini-specific app for Android users.
Integrated Duet AI features from Google Workspace into the Gemini ecosystem.
The launch of Gemini Ultra 1.0 marks a new phase for Google's AI capabilities, positioning Gemini as a versatile tool that combines aspects of an assistant, chatbot, and search engine. With the potential to replace Google Assistant, Gemini aims to provide a more conversational, multimodal, and helpful AI experience. While there's no Gemini app for iOS, Android users can now set Gemini as their default assistant.
The initiative reflects Google's investment in AI, suggesting that Gemini could become as integral to Google as their search engine. They also announced a paid plan, priced similarly to OpenAI's ChatGPT. This AI Premium Plan provides access to the most advanced features of Gemini, including the state-of-the-art Gemini Ultra.
European Union Members Unanimously Move Forward with Regulating AI
In a decisive move, the European Union unanimously endorsed the Artificial Intelligence Act, marking a significant step in regulating AI technologies.
The AI Act bans certain AI applications (such as social scoring), has tight restrictions on high-risk uses (such as health, safety, and democracy), and mandates for transparency and stress-testing of cutting-edge models.
The legislation still requires European Parliament approval, but most are confident that the law will pass with no changes.
Hugging Face Launches Open Source Competitor to OpenAI's Custom GPTs
Hugging Face has launched a new, free tool for creating customized AI chatbots, positioning itself against OpenAI's subscription-based custom GPTs.
Unlike OpenAI's model, which relies on proprietary large language models and requires a paid subscription, Hugging Chat Assistants offer a free, versatile platform where users can select from a variety of open source large language models for their AI's backend, including Mistral's Mixtral and Meta's Llama 2.
This launch underscores Hugging Face's commitment to open source innovation, challenging OpenAI's approach with a model that emphasizes accessibility and user customization.
ChatGPT's Degrading Performance Linked to Expanding System Prompt
Dylan Patel discovered that ChatGPT's system prompt is likely 1,700 tokens, and if OpenAI has been consistently adding to this prompt, that could explain why ChatGPT has been degrading over time.
The prompt has been verified by several users, and users confirmed that OpenAI's playground does not disclose this system prompt. If the playground provided the same response, this would more likely be a hallucination.
This could support OpenAI's claims that the model hasn't changed. Instead, the system prompt could be getting bloated, which would degrade ChatGPT if true.
Ultimately, the question is when the system prompt gets so bloated that the product becomes inferior to competitors. GPT-4 is still ahead, but many competitors are coming after them. Those competitors may not need something better, they may just need something that's more efficient.