AITiny integrates generative AI into Joomla's TinyMCE editor, the JCE editor, and media fields to generate new content, rewrite or refine existing content, and generate context-aware image alt attributes.
Generative AI (Large Language Models – LLMs) have got a bad reputation for all the wrong reasons. LLMs are a powerful tool. Just like with any tool, you can use it the wrong way, creating more problems for yourself and everyone around you; or you can use them the right way, to improve your productivity and expand your possibilities. Using LLMs judiciously lets you process your written content to better fit the purpose and tone of your site, reducing the laborious copy-editing between rounds of revision, as well as creating meaningful, context-aware image alt tags for visitors with disabilities.
Generating context-aware image alt attributes to correctly describe visual elements of your site to visitors with disabilities.
Expanding sparse content provided by the client.
Distilling long content into shorter text appropriate for use on the web.
Proposing meta descriptions, lists of keywords, and tags for your articles.
Creating abstracts.
Extracting key points (“executive summary” / “bullet points”).
Filling in missing factual information.
Changing the tone of voice.
Correcting spelling, grammar, and text quality errors.
Machine translation.
Answering questions based on the content you are editing.
Generative AI is not meant to completely eliminate human copywriters and proofreading. It is meant to be used as a tool which improves your productivity, but requires a modicum of human oversight. Generative AI can make mistakes, or make up (“hallucinate”) information. Always read the generated content, using common sense and your knowledge of the subject matter to eliminate any problems which may have been introduced by generative AI.
You can limit the plugin, or specific features, to certain user groups. By default, only Super Users can use it. You connect to the AI service by proxying through your server, meaning that users of the plugin do not see your API keys. The plugin fully supports streaming replies (you see the LLMs answer as it is formulated), with the possibility to abort the query mid-generation.
You can use a different model –but always the same service– for each feature. You can fine-tune the system prompts and other parameters (such as temperature, top_k, and maximum tokens) used for each feature.
You can use the Custom feature to run a custom prompt against your text, or even define your own features if you find yourself using the same custom prompt frequently.
Joomla's policy is not to add features which require a third party service. This is, for example, why Joomla removed the reCAPTCHA integration from the core in the middle of the Joomla 5 release cycle. You cannot have AI features without a third party service, making such an implementation moot.
Further to that, Joomla's conservative approach to development is great for stability, but does not work very well with a fast changing field like generative AI.