Google's Gemma 2 2B, GPT-4o Experimental Model and More: AI Roundup
Google's Gemma 2 2B, GPT-4o Experimental Model and More: AI Roundup
AI news worth knowing this last week.
Hello and happy Friday! :)
It’s been a minute since we talked about what’s going on in the AI world, and there are a few instances worth knowing.
AI continues to expand its capabilities as tech giants work towards developing newer and more powerful features.
Let’s dive into some recent developments:
🔷 Google Releases Gemma 2 2B: Most Compact, Open-Source AI Model
Google DeepMind has recently released the 2 billion (2B) parameter version of Gemma 2, the second generation of Gemma AI models. Gemma is a group of lightweight, text-to-text open models created for developers and academics, and is based on the same technology as Google Gemini.
DeepMind claims that its new 2B model learns from larger models through distillation and provides outsized outcomes. The company also claims to outperform all GPT-3.5 models in the LMSYS Chatbot Arena leaderboard.
Read more from Google here.
🔷 OpenAI: Advanced Voice Mode, SearchGPT and GPT 4o Long Output
OpenAI has a few new things incoming.
OpenAI has begun rolling out an advanced voice mode to select ChatGPT Plus users, allowing real-time voice conversations with the AI.
This new feature includes the ability to speak to and interrupt ChatGPT during interactions, enhancing the realism of conversations. Read here.
The company also launched SearchGPT, a new AI search tool that combines LLMs with real-time information from the internet. It aims to provide fast, relevant answers with clear sources, posing a huge new challenge to Google.
The latest is also the GPT-4o-64k-output-alpha, an experimental version of GPT-4o that can deliver up to 64,000 output tokens per request, compared to the previous maximum of 16,000 tokens.
This model is currently available only to alpha testers and is priced at $6 per million input tokens and $18 per million output tokens: slightly more expensive than previous versions but with enhanced features.
🔷 Mistral Large 2 Model at Par with OpenAI and Meta
Mistral launched a new flagship model, Large 2, on Wednesday, claiming that it is on par with the most advanced models from OpenAI and Meta in terms of code production, mathematics, and reasoning.
It was released just one day after Meta's latest and greatest open source model, Llama 3.1 405B. But it outperforms Meta's Llama 3.1 405B in code generation and mathematical tasks with just 123 billion parameters.
It supports 12 languages, 80 coding languages, reduces hallucinations, and has a 128,000 token window for large data processing. While not fully open-source, it is available for commercial use on platforms like Google Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM.
What launch are you excited to try out from these?