gemma license is not free software

[gnusupport]

The tool you've developed for interacting with my extensive website, which spans over 1300 pages with various customer-related materials, is incredibly beneficial. However, I'm still in the process of integrating it to test its full potential. I'm particularly excited about its potential to streamline my workflow, especially when used in conjunction with pdfitdown.

I'm grateful for the decision to distribute this tool as open-source under the MIT license, a practice even Microsoft has adopted.

Nevertheless, I've encountered a concern regarding the inclusion of specialized gems, which, despite being labeled as "open," come with their own set of restrictions and potential for misuse.

Here are 10 illustrative scenarios where the Gemma Prohibited Use Policy could be misapplied or misunderstood, potentially leading to unintended negative outcomes for users:

---

1. Academic Research on Sensitive Themes

• Academics delving into topics like hate speech or radical ideologies for scholarly analysis might inadvertently breach the policy by generating content for pattern recognition, mistakenly categorized as "inciting hostility."

---

2. Artistic and Satirical Works

• Artists leveraging Gemma for satirical pieces or provocative works could face penalties if their creations are misconstrued as "endorsing violence" or "harassment," contrary to the principles of free expression.

---

3. Content on Mental Health and Self-Care

• Users crafting content aimed at mental health awareness or suicide prevention resources might be mislabeled for "encouraging self-harm," overlooking the beneficial intent.

---

4. Political Commentary and Activism

• Political commentators utilizing Gemma to craft critical narratives about governments or corporations could be wrongfully accused of "spreading falsehoods" or "fostering animosity," particularly if the subject has sway over policy enforcement.

---

5. Historical and Documentary Content Creation

• Filmmakers generating historically accurate depictions of sensitive subjects (e.g., war crimes, colonialism) might be mislabeled for "promoting hostility," despite the educational or documentary intent.

---

6. Fictional Writing and Character Development

• Novelists employing Gemma to craft fictional narratives or characters involving sensitive themes (e.g., crime, abuse, extremism) could be mistakenly deemed as "facilitating illegal activities" or "promoting violence," even though the work is purely fictional.

---

7. Privacy and Security Research

• Cybersecurity experts creating content about surveillance or tracking methods to highlight privacy threats might be mislabeled for "monitoring individuals without consent," despite the aim to educate and safeguard.

---

8. Cultural and Religious Expression

• Users generating content reflecting their cultural or religious practices (e.g., discussing controversial historical events or beliefs) might be wrongfully accused of "inciting hostility," if misinterpreted by automated systems or biased reviewers.

---

9. Parody and Impersonation

• Comedians using Gemma for creating parodies or impersonating public figures for entertainment might be erroneously charged with "deceiving through impersonation," even when the intent is clearly comedic.

---

10. Scientific Inquiry into Sensitive Topics

• Scientists utilizing Gemma to investigate sensitive subjects like sexuality, gender identity, or substance use for research might be mislabeled for "producing sexually explicit material" or "facilitating illegal activities," despite the legitimacy and ethical nature of the research.

---

Personally I will not download non-free software, thus will never use Gemma and can't use free software incorporating non-free software. That also causes confusion in the next line.

Maybe you could consider using Microsoft's excellent Phi-3.5-mini-instruct model that is MIT licensed instead of Google.

References:

What is Free Software? - GNU Project - Free Software Foundation
https://www.gnu.org/philosophy/free-sw.html

The Open Source Definition – Open Source Initiative
https://opensource.org/osd

Word "Open" as in "Open Source" - Words to Avoid (or Use with Care) Because They Are Loaded or Confusing:
https://www.gnu.org/philosophy/words-to-avoid.html#Open

Gemma Prohibited Use Policy  |  Google AI for Developers
https://ai.google.dev/gemma/prohibited_use_policy

[AstraBert]

Thanks, that's a really poignant observation! Will fix with the next release of qdurllm by 31 January 2025, see Milestone for January 2025

[gnusupport]

That is so good to hear. I hope you understand how mixing two different types of licenses may impact author's work (you) and those other people building on your work. Though there are great LLM models which are truly free software and well performant.

[AstraBert]

Hi @gnusupport, the new pre-release implements now these three models:

• Qwen2.5-1.5B-Instruct, with Apache 2.0 license
• nomic-ai/modernbert-embed-base, with Apache 2.0 license
• prithivida/Splade_PP_en_v1, with Apache 2.0 license

I think that they should solve the "open" licensing issue that Gemma had: if you have any thoughts, feel free to share them here

[gnusupport]

That is really great that you are using fully free software large language models. I never talk in terms of "cost" when saying "free", I am talking like in terms of software libero. I have to figure out how to install it.