Robert Gützkow

@HonkHase@chaos.social das bedeutet nach meinem Verständnis, dass alle Meldungen an das CERT-Bund an den BND weitergereicht und ggf. ausgenutzt werden würden. Im Kontext vom Cyber Resilience Act würden alle Meldungen die über die Single Reporting an die ENISA und das BSI übermittelt werden, ebenso betroffen sein.

@HonkHase@chaos.social ich bin ein wenig erstaunt dass das nicht nur impliziert sondern sogar direkt so im Referentenentwurf steht. "Bis eine 0-Day-Schwachstelle nach Veröffentlichung bei den Herstellern durch das Bundesamt für Sicherheit in der Informationstechnik in Systemen durch Patches behoben wird, vergeht Zeit, die der Bundesnachrichtendienst für wichtige Arbeiten nutzen kann, welche die Grundlage für die Auftragserfüllung darstellen. Wird das Bundesamt für Sicherheit in der Informationstechnik aufgefordert, unmittelbar, beispielsweise über eine Austauschplattform, Erkenntnisse zu teilen, kann die Zeit zwischen interner Bearbeitung, Meldung an die Hersteller, Behebung des Softwarefehlers, Bereitstellung eines Patches und finaler Installation des selbigen, ausreichend lang sein, um die Schwachstellen in Wert bringen zu können"

Qwen 3.5 is an open weight model licensed under Apache 2.0. Like other LLM models, Alibaba does not publicly share their training data corpus. It does not honor the licenses and copyrights of its source material.

@fsiddi ok, sounds good to me. For reference: The original POC by Dalai used ollama with qwen3.5 https://projects.blender.org/dfelinto/research_and_development/src/commit/3084d78edefefd5d5e2a4c1204f170e4952b2261/mcp/ollama.md The current MCP server uses llama.cpp and suggests to use HeYujie/Qwen3.5-35B-A3B-abliterated-GGUF https://projects.blender.org/lab/blender_mcp/src/commit/03004fd0216bfe5e0a3d9ac9b47d5efadc3d78c4/readme_local_llm.rst

@fsiddi @dfelinto I can completely understand if you wish to not answer these publicly for now, but do please consider them in your discussions about how you're moving forward regarding AI and Blender.

@fsiddi @dfelinto Furthermore, did you at any point during the development of the MCP server add-on discuss the potential risks of creating an interface for LLM-based automation (made officially by Blender, not a community contribution) that may in the future be used to automate artists' work? Were there any safeguards or limitations built-in as part of the design to protect against such a scenario?

@fsiddi @dfelinto Additionally, what locally run open source / open weight LLM did you use for these tests? I'm not aware of any that both align with Blender's values and are capable of delivering results comparable to their commercial counterparts. Seems like models that honor the licenses of their training materials are quite rare.

@fsiddi @dfelinto That much is obvious from the lab page and associated repositories, but that is not where I was aiming with that question. Natural language processing is not the starting point, it's a potential technical solution for problems you're trying to solve. Improvements in the mentioned areas could be driven by other approaches to make them more accessible and easier to use. Again, apologies for being persistent on this, but I think this matters. Could you elaborate on the motivation that led you to exploring LLM integrations? Is it because the technology was there and you thought: "That's neat, let's see what it can do"? Did people approach you from the community, industry, or research, perhaps at previous Siggraphs, that this is something they would be interested to see? Was this done with the intention of attracting the attention of corporations in that space, increasing the likelihood of donations?

@Umbreon Glad I donated to them yesterday!

@dfelinto @fsiddi Hi Dalai, it's been a while! I'm glad that you didn't develop this in collaboration with Anthropic. Sorry if I'm grilling both of you with uncomfortable questions, but could you please explain what the motivation was for exploring this feature?

@fsiddi Thank you for taking the time to respond Francesco! I'll keep an eye out for the decisions made and will resume my contribution to the development fund.

@fsiddi I'm glad that you have considered the feedback from the community and are taking actions. Following the new announcement and your statements here, I do still have some questions. You state that "Blender does not need to take any action towards those companies, or support their goals in any way" and that you will have to tackle how generative AI will be treated in the context of Blender in general. Prior to the announcement of Anthropic as a corporate patron, Campbell implemented an MCP server addon (https://projects.blender.org/lab/blender_mcp) which was also referenced in Anthropics press release (https://www.anthropic.com/news/claude-for-creative-work) Am I correct to assume then that this was done on the own volition of the team and not as some form of collaboration with Anthropic? Is the continuations of projects related to generative AI like this being reconsidered?