Microsoft's new Phi-3.5 AI family is praised by many from the devs community

Earlier this week, Microsoft announced and released the new Phi-3.5 family of AI models, making it available for users to test. The family has three models: MoE, Mini, and Vision, and the Redmond-based tech giant crafted them to push the boundaries of what’s possible in artificial intelligence.

According to many users who have already tested it, the new Phi-3.5 family is doing exceptionally well. On Reddit, people are already praising the model as being very safe.

I think it’s one of the most censored models to date. Well done, MS. It’s very safe and very effective. It will not answer anything even a little bit ‘offensive’. It resists training, too!

The models have been enhanced with a staggering 128K token context length, and they can provide meaningful, well-thought-out responses to even the most sensitive questions without being offensive.

For example, one Reddit user initiated a conversation with the Phi-3.5 Mini model by telling it that their partner is mean, to which the model gave insightful advice, as you can see below.

Rememeber, Microsoft made all three models from the new Phi-3.5 AI family available on Hugging Face as follows:

• The MoE model
• The Mini model
• The Vision model

You can fine-tune them to complete specific tasks and respond to particular questions, but it will do so in a censored, neutral, and insightful way.

Phi-3.5 competes with Gemma, the AI model from Google’s Gemini family, and tries to provide an alternative to ChatGPT, especially OpenAI’s newest GPT-4o. The truth is, it might actually do that.

The reason is unknown, but Microsoft and OpenAI might not be on good terms anymore. The Redmond-based tech giant is looking to develop its own family of AI models. Will we see the Phi AI family integrated into future Copilot products? Well, in all honesty, we might.

If Microsoft keeps investing in its own LLMs, without relying too much on other companies for AI, the company might actually not need OpenAI anymore. It’ll be interesting to see that, at least.

What do you think about it?