Meta and Microsoft Collaborate To Introduce Llama 2

Big Tech firms Meta and Microsoft join forces to launch Llama 2, an open-source large language model. It'll be featured on Microsoft’s Windows and Azure cloud platform

[{"selector":"#anim-e9ff3c77-7e42-499b-8892-972b51a08cc4","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-5fe203a2-c5bd-46f2-b3d3-1e482006b3f6","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-8d24daa8-0117-41b6-8a50-ec15bcca71e6 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Llama 2 is free for research and commercial use, optimized to run on Windows

[{"selector":"#anim-d037d12d-3876-4ed7-8df1-5d5770c197e9 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-682b546a-8949-4f19-bb4c-1cfeb416e38f","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-c4cf670b-ab29-4dea-98d6-250fde00b821","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Meta’s AI tech stack powers Llama 2, which was trained on 40% more online data sources compared to Llama 1

[{"selector":"#anim-87d12765-c6f7-4035-a5d1-409cfe890608 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-45fbafba-b58c-4369-b388-12f23f862073","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-d0704cc3-8b7c-47c2-b89e-34f30b9e68b5","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

It can process twice as much context, making it great for businesses and researchers

[{"selector":"#anim-f39305e4-24b3-4460-9be6-0b2ca7bb2ae0 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-60f0c76d-17c6-4d4e-bd52-b414290d5321","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-81e97f50-05e3-45f8-ad3e-9ba430c4eae1","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Llama 2 outperforms many open-source LLMs in coding, proficiency, reasoning, and knowledge tests

[{"selector":"#anim-0579e1cc-9828-4f35-982b-68a8926230f1 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-3cbeb41e-ff41-476d-9abc-c1ca1edfa849","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-10e69f9f-b528-457f-ae63-a9ae205d6a46","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

However, it's not as efficient as closed-source competitors like OpenAI’s GPT-4, as stated in a research paper by Meta

[{"selector":"#anim-bd8d1ca7-5930-49ba-a49d-68fb59c9f552 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-b4dfd4c0-e3a0-4aab-82c0-60ddd21fb877","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-6b88ee92-fc7f-4d14-be3b-d7450703ab9f","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Meta CEO Mark Zuckerberg, on July 18, called Llama 2 the foundation for researchers and businesses to build with their next-gen large language model

[{"selector":"#anim-21d2a38f-d8b3-4a7a-9417-ebd0822c1e6b [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-860b689b-5fdd-4b8a-b930-f77a373c76b7","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-d27388e1-16d6-46a7-8851-9baf5ee8b4f1","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Llama 1's demand was impressive, with over 100,000 access requests, but ChatGPT surpassed it with an estimated 100 million or more users in the first three months

[{"selector":"#anim-3b924bad-6250-408c-bc14-7d052ffa818c [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-a09ca303-79ae-414e-90ff-12639accd1f7","keyframes":{"opacity":[1,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-10298f7c-b9d6-4f70-9707-7a575513369e","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Meta faced criticism for open-sourcing Llama, with concerns about potential misuse due to "minimal" protections, according to two U.S. senators in June