Meta’s Llama 2 License is Not Open Source : Chris

Reader
August 04, 2023
Meta’s Llama 2 License is Not Open Source
by: Chris
blow post content copied from  Be on the Right Side of Change
click here to view original post

5/5 - (1 vote)

I love that Facebook (aka. Meta) released its large language model to researchers. The release of the first version has created a Cambrian explosion of open-source LLMs.

I’ve written extensively about these tools in many Finxter blogs:

But wait. Is Llama 2 open source in the first place? Let’s examine this important question!

Is Llama 2 Open Source?

Llama 2, contrary to popular belief, isn’t open-source in its purest form. Its licensing terms fall under the proprietary Llama Community License, a variant without approval from the Open Source Initiative (OSI). According to the OSI, the Llama Community License fails to fulfill the stipulations outlined by the Open Source Definition (OSD).

A crucial deviation of the Llama Community License from the OSD lies in its lack of mandate for source code accessibility. In practical terms, this inhibits users of Llama 2 from either modifying or redistributing the code, thereby putting constraints on the openness of the platform.

Meta, the driving force behind Llama 2, has countered this narrative, asserting that the Llama Community License maintains the spirit of open-source. They argue that by facilitating the execution, analysis, and sharing of the model, it adheres to open-source principles. However, this perspective isn’t universally accepted within the open-source community. Critics underscore the absence of source code availability as an insurmountable obstacle to Llama 2 being authentically open-source.

Meta has expressed an openness to revisit Llama 2’s licensing terms amid this ongoing debate. However, the specifics and timeline of this potential shift remain uncertain.

If you’re in search of a genuinely open-source large language model, several alternatives exist, including the likes of Jurassic-1 Jumbo or MPT-7B:

✅ Recommended: MPT-7B: A Free Open-Source Large Language Model (LLM)

Licensed under the OSI-approved Apache 2.0, these models offer users full access to the source code, adhering more strictly to the principles of open-source software.

✅ Recommended: 6 Easiest Ways to Get Started with Llama2: Meta’s Open AI Model

The Llama 2 License Like I’m Five

Okay, kiddo! 👶

Let’s imagine we’re talking about a brand-new, super-special robot toy called Llama 2. The company that made it, called Meta, has rules about how you can play with it.

These rules are called the “Llama 2 Community License Agreement.“

It’s like the rulebook that comes with a board game, telling you what you can and can’t do.

THE RULEBOOK:

  1. First, if you want to play with Llama 2 or share it with your friends, you need to say “I Accept” to these rules.
  2. In these rules, Meta says you can play with Llama 2, share it with your friends, or even use it to create new toys! But, if you share it with your friends, you have to give them a copy of these rules, and your play should follow all the general rules and laws, just like when you’re playing in a park.
  3. Also, there’s one very special rule: You can’t use Llama 2 to make other robot toys (aka. LLMs) better unless they’re related to Llama 2. This is for Google Bard and ChatGPT researchers that are not allowed to retrain and improve their models using Llama 2 output, for instance.
  4. If you are a big company with over 700 million people using your products every month (which is a lot, like more than all the people in the U.S.), you have to ask Meta first before you can use Llama 2. As I write this, these companies have over 700 million users: Alphabet, Tencent, TikTok, Microsoft, Snap, and Telegram.
  5. Meta also reminds you that they’re just sharing Llama 2 with you to play with – it’s like borrowing a toy. So, they’re not promising it will be the best toy ever or won’t break. If it breaks, it’s not Meta’s fault.
  6. Meta also says they won’t be responsible if you hurt yourself or someone else while playing with Llama 2.
  7. As for the name and the design of Llama 2, they belong to Meta. You can’t use their name or design to make your own toys unless you’re just saying that your toy was made from Llama 2.
  8. If you use Llama 2 to create a new toy, you own that new toy. But remember, the rules for the original Llama 2 still apply to it!
  9. If you tell Meta that Llama 2 is breaking some of your own toy-making rules (like it’s too similar to a toy you’ve created), then Meta can say you can’t play with Llama 2 anymore.
  10. If you don’t follow these rules, Meta can take Llama 2 away from you. 😢 And, if you or Meta ever disagree about the rules, a judge in California, where Meta lives, will decide who is right.

Let’s dig deeper into the discussion on whether Meta’s Llama 2 model is open-source.

Meta’s LLaMa 2 license is not Open Source

A recent article on opensource.org takes the stand that the license doesn’t adhere to open source standards.

Meta’s claim that its LLaMa 2 AI system is “open source” has come under scrutiny. The Open Source Initiative (OSI) argues that the tech giant has misconstrued the term, pointing out that “open source” implies a license in line with the Open Source Definition (OSD) requirements. The OSD mandates no discrimination against users or use-cases.

However, Meta’s LLaMa 2 license fails to meet these standards. It imposes constraints on commercial use for some users and limits the software’s application for certain purposes. OSI emphasizes the importance of true “open source” licensing, as it offers developers and users the freedom to utilize technology as they see fit, ensuring they retain sovereignty over their tech use.

Despite understanding Meta’s intentions to limit LLaMa 2’s use for competitive reasons, OSI reiterates that such restrictions contradict the core concept of open source. The inability to foresee all possible future applications of a technology underpins the OSD’s stand against such constraints.

In essence, Meta’s limitations could potentially prevent LLaMa 2 from contributing significantly to diverse sectors, including controlled substances regulation and critical infrastructure.

OSI highlights that the LLaMa 2 license isn’t “open source” in its true sense and urges Meta to address this confusion. Recognizing the pressing need to define “open” in the AI context, OSI is hosting events to formulate a common understanding of the term and encourages the submission of ideas.

A Few Words on Llama 2 Acceptable Use

Meta’s rules for using LLaMa 2, their AI system, are all about making sure it’s used in a safe and fair way.

Here are the main things you’re not allowed to do:

  1. Break the law or harm people’s rights: This means you can’t use LLaMa 2 to support anything illegal, like violence, terrorism, or child exploitation. You can’t use it to harass people, discriminate against them, or gather sensitive personal information without permission. It’s also not allowed to use it professionally if you’re not licensed to do so or to interfere with computer systems.
  2. Do anything dangerous: This includes using LLaMa 2 for military purposes, working with illegal weapons or drugs, running vital infrastructure, or promoting harmful activities, like self-harm or violence.
  3. Mislead or trick others: You’re not allowed to use LLaMa 2 to create or spread lies, defame people, spam others, pretend to be someone else without permission, misrepresent AI outputs as human-made, or fake online interactions.
  4. Keep dangers hidden: If you know any risks of using your AI system, you must tell the users about them.

In simple terms, you have to use LLaMa 2 in a responsible and honest way without causing harm or risk.

No $h!t.

If you love learning about LLMs, keep reading this recommended article next: 👇

✅ Recommended: Top 5 LLM Python Libraries Like OpenAI, LangChain, Pinecone


August 04, 2023 at 04:49PM
Click here for more details...

=============================
The original post is available in Be on the Right Side of Change by Chris
this post has been published as it is through automation. Automation script brings all the top bloggers post under a single umbrella.
The purpose of this blog, Follow the top Salesforce bloggers and collect all blogs in a single place through automation.
============================

Salesforce