VENDORiQ: Meta AI Will Release Open Source AI Model

Meta AI is set to release Llama 2, an open source AI model in a bid to disrupt the AI industry. Should enterprises trust the model?

The Latest

8 August 2023: Meta AI is launching a commercial version of its open-source AI model, Llama, providing a free alternative to expensive proprietary models from OpenAI and Google. The new version, Llama 2, will be available through Microsoft’s Azure and will run on Windows. Llama 2 was trained with two trillion tokens or raw text, and has double the context length than Llama 1, according to Meta AI’s whitepaper. The company will also release Llama 2-Chat which will leverage publicly available instruction datasets and over 1 million human annotations.

Why It’s Important

This move could challenge the dominance of closed source models in the generative AI market and could have an impact on companies like OpenAI, whose models are already available on Azure. Meta AI acknowledged that its tests are not yet comprehensive enough and that the benchmarks may not include enough diversity, which could lead to biases.

Similar to any other generative AI model for critical applications, enterprises should exercise caution when adopting Llama 2 to mitigate potential risks. Thorough evaluation and testing of the model’s performance in diverse real-world scenarios are essential. Organisations must also consider employing advanced filtering mechanisms and moderation tools to ensure the generated outputs meet acceptable standards. It is crucial to recognise the Western bias in the model’s results. To address this, it is necessary to make an effort to promote data representation and incorporate localised training datasets that are specific to different regions and cultures. 

Who’s Impacted

  • CEO
  • AI developers
  • IT teams

What’s Next?

  • To ensure responsible AI practices, enterprises should collaborate with AI ethics experts, use explainable AI techniques, and establish strong monitoring and feedback systems. This will help assess and adapt the model’s behaviour, with a focus on fairness and transparency.

Related IBRS Advisory

Trouble viewing this article?

Search