- Starter AI
- Posts
- GPT-4o goes mini, Researchers trick LLMs, Mistral’s NeMo model
GPT-4o goes mini, Researchers trick LLMs, Mistral’s NeMo model
The compact trend continues.
Sthefania
July 22, 2024
In partnership with
Hello, Starters!
As we've stated before, major companies in the AI league are shifting their focus to small models due to their cost-effective potential. This trend has been ongoing, and we expect it to keep growing... So, let's be on the lookout!
Here’s what you’ll find today:
OpenAI introduces GPT-4o mini
LLM safeguards are not as strong as we think
Mistral and Nvidia present NeMo
Apple released new DCLM models
Big names in the industry focus on AI safety
And more.
🤖 OpenAI introduces GPT-4o mini (4 min)
After hinting at its release in the LMSYS chatbot arena, OpenAI has finally made its new model official. We're now ready to experience the power of GPT-4o mini. There's much to be said about the model, but the focus remains on two key aspects: its size and its cost efficiency. Clearly, OpenAI acknowledges the industry trend of smaller models, as it makes AI technologies more approachable and affordable.
GPT-4o mini is now available for free to ChatGPT users as a replacement for GPT-3.5. With a context window of 128K tokens and updated knowledge up to October 2023, it stands as a strong competitor for similar models like Claude Haiku and Gemini Flash.
There's a considerable flaw that affects most language models and could potentially lead to harmful outcomes. Researchers from the École Polytechnique Fédérale de Lausanne (EPFL) reveal in a study that it is easy to bypass the safeguards of AI systems by just rewriting malicious queries in the past tense.
The paper explains that by writing in the past tense, models like GPT-4o are likely to answer queries that in other circumstances would be blocked by safety measures. This leads to the belief that AI models consider past actions less harmful than future scenarios, a vulnerability that calls for attention as LLM technology keeps developing.
💥 Mistral and Nvidia present NeMo (1 min)
Mistral and Nvidia have joined forces to introduce Mistral NeMo, a compact model with 12B parameters and a context window of 128k, that is directed at researchers and enterprises. Its size allows it to have enhanced performance without the need for fancy systems, as it can easily run on Nvidia's RTX GPUs.
Mistral NeMo is multilingual, and recent comparisons show it surpassing models like Gemma 2 9B and Llama 3 8B.
Learn AI-led Business & startup strategies, tools, & hacks worth a Million Dollars (free AI Masterclass) 🚀
This incredible 3-hour Crash Course on AI & ChatGPT (worth $399) designed for founders & entrepreneurs will help you 10x your business, revenue, team management & more.
It has been taken by 1 Million+ founders & entrepreneurs across the globe, who have been able to:
Automate 50% of their workflow & scale your business
Make quick & smarter decisions for their company using AI-led data insights
Write emails, content & more in seconds using AI
Solve complex problems, research 10x faster & save 16 hours every week
🍎Apple has already been in the small model trend for a while; however, they're stepping up to the game with new additions to its family of open DCLM models. They've recently released two versions with 7B and 1.4B parameters, which are securing great spots in most benchmarks.
🚨A group that includes familiar names like Amazon, Nvidia, Google, OpenAI, and more, is teaming up in the creation of CoSAI (Coalition for Secure AI). This project focuses on providing users with the necessary information, tools, and frameworks to increase the safety aspects of AI.
⚡️Quick links
What did you think of today's newsletter? |