Seems rather out of scope for core LibreOffice ODF document preparation.

While multiple extensions with AI implementations are already available:

localwriter (Ollama based)
https://github.com/balisujohn/localwriter

LibreOfficeAICopilot (chatGPT based)
https://github.com/johnwalz97/LibreOfficeAICopilot

stable-difusion (for image generation)
https://github.com/ikks/libreoffice-stable-diffusion

WritingTool w/LanguageTool support
https://writingtool.org/index.php/en/


IMHO a simple => WF

I am removing the Android aspect of this request, as there is no LO Writer version for Android at the moment; there is only a viewer, for which the request is not relevant.

Reporter, if you believe I'm mistaken or have misunderstood, please explain how Android is relevant

(In reply to V Stuart Foote from comment #1)
> IMHO a simple => WF

My bottom line is similar, but not that simple, and not because of the availability of extensions.

(In reply to Siarhei Kuchuk from comment #0)
> Currently AI changes approach to document creation.

Not really. Or rather, you can generate generic document text using LLMs, but that is the kind of document we would rather not receive nor write.

> It offers significant improvement of speed of work.

Again, only if you either generate slop you don't care about, or try generating someone more complex without bother to very carefuly read through it and consider anything that might be missing.

> LibreOffice is free, but it does not offer possibilities to work faster for
> workers.

We should not aim to increase writing speed through the generation of text; and we should not encourage people's ability to express themselves atrophying because some LLM is filling in text for them.

> Most PC has enough performance to run free dumb models through ollama.

You are undermining your argument by calling the model you're suggesting "dumb"... 

Also, regardless of whether model is dumb or not - we should definitely not rely on maxing out the performance of "most PCs": We should be frugal in the use of resources. Granted, that objection can be circumvented by making the use of LLM an optional feature, but then - that would make it even closer to what an extension offers.

> Can you please consider add following features (i ordered them from most
> important):
> 
> - Text summarization

Oh no.

> - Content generation/drafting
> - Writing style improvement
> - Paraphrasing and rewriting
> - Tone adjustment
> - Outlining and structuring
> - Autocomplete and predictive text

I argued above that we should discourage this on principle.

> - Translation

This may be useful, but - you would need to convince that it should be an integral feature rather than an extension.

> - Grammar and spelling correction

If you can argue that our current mechanism for spelling is worse than using an LLM-based one, that might be a suggestion worth considering, separately.

> - Research assistance

More like research derailment...

> - Citation and reference formatting

We already have a mechanism for that.

> - Data analysis and insights from text
> - Image generation from descriptions
> - Content extraction from images

Out of scope for Writer. 

> - SEO optimization

Out of scope for Writer, plus, mostly illegitimate and socially detrimental I would say.

> - Table and chart generation

We have decent chart generation in Calc; and we have text-to-table in Writer.

> - Fact-checking

Let's do even better than an LLM and integrate queries to the magic 8-ball online for fact checking! https://magic-8ball.com

> - Readability analysis

We should not commit to whatever some LLM thinks about readability as LO's opinion regarding whether or not some text is readable.

> - Accessibility suggestions

To the extent we can do that, we don't need an LLM to do it.

> - Template creation

How is an LLM relevant to that?

> - Document comparison

We don't need an LLM for that.

> - Transcription and speech-to-text
> - Text-to-speech narration

Either out of scope, the OS/DE's job, or a matter for an extension.

> - Formatting assistance

You're essentially suggesting we facilitate and encourage direct formatting.

> - Content extraction from PDFs

Can you show an LLM or other FOSS AI-ish system which does this robustly? Even then I would be suspicious, but this is something I would suggest filing as a separate bug.

> Performance boost of workers and writers. Currently world is not imaginable
> without writing reports in 1 minute instead of hour with great formatting.

Somehow, I am reminded of this: https://www.themoviedb.org/movie/7512-idiocracy

Thanks for your comments.

I put it as issue here, because i see it as important in general. But i don't have vision of where project is going and if and if yes then how its supposed to be. That's why its totally up to you. Its mearely a suggestion, nothing more.

Personally, i tend to create very short documents for myself, when needed. Because i don't have time to read lots of text. And 2 person from my work tend to ask copilot respond to me so they dont have to type, which i consider rude. Because i need their opinion after all.

There's a hipe around AI in my area, programming of web-sites. As I see where its heading, in 2 years it will be moved to Azure AI cloud, where a client that needs application or web-site, will pay for Azure AI resources CPU and time, and just write in chat form. Hey, I need that site! Behind the scenes, it will spawn virtual AI agents 'team' that will design, program, test, deploy web-site and boom its done. Without need of me... anymore. In half an hour, not months.

So my thoughts were that office men should show crazy performance at their job places in the office to compete against AI and it can be done by incorporating of generating content or getting summary at a fast tempo.

From the other side i see company purchasing AI subscriptions, and it has access to all companies data with vague promises that it won't be shared. Of cause. It works in this way. Application communicates with AI subscription that has a part of knowledge - all information of company, user has access to - and based on it and document content can do something using this knowledge.

To my understanding, Ollama is just a manager for various (open) LLMs. You still access, for example, OpenAI with data from unknown sources.

Open Source advocates usually hesitate to accept these non-transparent, resource hungry, and big tech related tools. On the other hand, this AI bubble is "the next big thing" and Microsoft build all tools now around its Copilot. Many users expect AI capabilities from software nowadays.

As usual we probably better start with use cases rather than hearsaying about the blessing from new technologies.

#1 Eve wants to summarize the many pages she wrote about a specific topic to create an introduction the thesis. 
#2 Eve wants to generate a couple of standard sentences on a common topic to expand later on her own.
#3 Benjamin wants to chat with a bot while doing his homework to understand the topic better. And to avoid searching the Internet manually.
#4 Benjamin wants to find a synonym for a word based specific for the whole text and in a certain style.
#5 Benjamin wants to analyze a bunch of data but has no clue how to approach. He expects the assistive function to kick him off with the work.
#6 Adrian wants to provide his colleagues with a LLM based on input in the company that later is used to support the preparation of offers.

We probably can expand this for quite a while. But I guess it's clear that we cannot reject all use cases.

My personal take is that we should not participate in the public LLM circus but allow users to do so. Extensions are great for this purpose. Or we introduce own LLMs, or something based on public data.

One more thought that for employees, company most likely will have subscription to model that its most advance or capable and has access company data  permissioned to user... and has policy explicitely forbidding to use other models. So for them it would be cool to somehow has connector to model their company allowed them to use and paid subscription for. Also in some government or military organizations usage of AI is strictly prohibited.

For personal usage i see 3 cases. 
a) user already paid for one of popular models and wants to have it features available
b) user has for example Copilot at windows pc with free quota or paid one and just want to try it within quota daily limit.
c) user has hardware capable of running LLM (which is basically videocard with 8GB+ RAM or special cheaps that are coming to the market this and next year) and want to use whatever model software can provide or he plays with at the moment.

Where its heading on personal market?

For Copilot users - as i see it - in 2 years it will be not just LLM, but as a LLM with benefits - personal assistant. PA will have access to all documents and emails and chats on user PC (which will be a reborn of Windows Recall feature). It will be major invading of privacy area which is unregulated for now. PA will provide personalized output based on all user data. User will be happy to have it access to all its data and even emotionally attached to it. PA will have subscription levels from free to basic plans with ads (where ads will be tied to all user data - a whole new market) to very expensive without ads and more proactive assistant what will be capable of doing some tasks on its own.