Meta recently unleashed their latest creation, Llama 2, upon the AI community - and it's making some serious waves. This new large language model (LLM) from Meta is more than just a research project, it's a bold move towards open source and a direct challenge to closed models from big tech.
An Impressive Open Source Model
So what exactly is Llama 2 bringing to the table? It's a major upgrade over Meta's previous Llama model with double the parameters at 70B (a 40% increase over the first model). It was trained on a huge dataset of 2 trillion tokens, giving the model stronger performance on benchmarks for helpfulness, safety, and reduced toxicity according to Meta.
But the real kicker is that unlike key competitors like Google's LaMDA and Anthropic's Claude, Meta is openly licensing Llama 2 for non-commercial research AND commercial use. This gives developers way more flexibility to build on top of Llama 2 and integrate it into their own applications and services.
Meta is even providing pretrained versions of Llama 2 ready for fine-tuning on platforms like HuggingFace. There's the base Llama 2 model, as well as Llama 2-Chat finetuned specifically for conversational tasks.
Sparking a Generative AI Showdown
By releasing such a high-capability open source LLM, Meta makes a bold statement - they are ready to compete with the likes of Apple, Google and others on equal footing in generative AI research.
Previously, these tech giants guarded their latest models like LaMDA, GPT-4, and Apple's secret project closely. Llama 2 challenges them to stop hoarding their technology and put their money where their mouth is on open source development. We're now witnessing a generative model arms race. With Meta's move, the pressure is on for companies like Apple to unveil what they've been brewing behind closed doors. Llama 2 just upped the stakes in this AI battle among the tech titans.
"Open Source" With Caveats
However, some industry observers dispute Meta's "open source" characterization, pointing out the license places notable restrictions:
- Companies with over 700 million active daily users (looking at you Apple) are excluded from using Llama 2 without explicit permission. This aims to block tech giants like Apple from leveraging the model.
- License prohibits using Llama 2's outputs to improve proprietary LLMs. Again targeting competitors benefiting from Meta's work.
So while more open than private models from tech giants, Meta still imposed limitations to maintain competitive edge.
Pushing The Boundaries of Open Source AI
The big question now is how will other tech giants respond? Will we see them release their own "open" models without the restrictive licensing? Or do even larger proprietary models get unveiled to eclipse what Llama 2 makes possible?
One thing's for certain - Meta just accelerated the evolution of generative AI with Llama 2's arrival. Their open source challenger is pushing boundaries, forcing Big Tech to rethink their closed-door approach. Exciting times are ahead at the intersection of language models and open collaboration! It's agreat time to be a an AI developer and user as we see more and more of these fine tuned and competitive models being release. I for one and really excited to see what some of the 'less censored' offerings bring us!
Reference: You can aread more about Llama 2 in the release paper here.
Image Created on MidJourney
Prompt: "funky llama, blues brothers suit, fantastic mr fox style, cool bright sunglasses, standing in village, wes anderson style, funny, --ar 4:3".
If you'd like to know more about AI and making it work for your business checkout my startup Erudii. I'm always looking for people to collaborate with!
Top comments (0)