9 posts tagged with "PNGTuber-GPT"

The largest release of PNGTuber-GPT to date! Version 1.1 is finally here! Happy Easter!

• This release has been a complete rebuild of the solution. It has many new features, but it's clean modular codebase is the best new feature.
• This release requires you to completely remove any previous installation by deleting all commands and actions in Streamer.bot related to PNGTuber-GPT. You can then reimport the new file.
• New release format: the download now comes in a SB file, that you can simply drag and drop on top of the import window to import the solution.
• The bot can now log question and answer pairs to a Discord channel.
• Node.js is no longer required. You can uninstall it, and then delete message_log.json from your bot directory. The bot now saves the chat log directly to memory on each chat message, using the native API token in Streamer.bot.
• You can now list bot names that will not be recorded in the chat log. These are saved under the argument "ignoreBotNames" as a comma-separated list of bot names to ignore, this can be found under the SaveSettings action.
• New moderation features allow you to use !modclearchathistory and !modclearprompthistory to remove memory from the bot when an inappropriate question is asked, or someone says something inappropriate in chat.
• The bot now remembers the previous 20 chat messages and the last 5 question and answer pairs in separate queues. This means that even if 20 messages go by between asks, you can still ask the bot follow up questions on the next redeem.
• Core Directives have been removed, this was only a test feature, and isn't really necessary. The oowoo and ohwoe replacements can be done with Text Replacement in Speaker.bot and the instructions now reflect this.
• The bot should now respond in less than 500 characters in nearly all cases. This is achieved by tacking on "You must respond in less than 500 characters" onto the user prompt, as if they said it. The bot seems to listen to the user's instruction above any context directives.
• Revamped moderation settings allow individual control of moderation categories from the central SaveSettings action
• Clearly marked variables that could lead to the bot breaking in brackets [ ] check for all of these before operating the bot and make sure they have been specified. Any variables with multiples should be comma-separated. Remember to add your bot names to ignore to the SaveSettings Action. This needs to be every bot in your channel, including the bot itself. FAILURE TO LIST THE BOT ITSELF IN THE IGNORE LIST CAN LEAD TO ERRORS.
• Fixed moderation endpoint, it was returning an empty chat message and breaking the workflow causing the bot to just not respond sometimes.
• Fixed issue where pronouns were returning incorrectly due to bad pronoun matching logic
• Added ability to remove nickname with !removenick
• Added ability to remove the thing you taught the bot about yourself with !modforget
• Added the ability to get the memory of what you taught the bot with !get memory
• Added the ability to clear the log file with !modclearlogfile
• Added the ability to clear the chat log with !modclearchatlog
• Added the ability to clear the prompt log with !modclearpromptlog
• Added !modreinitialize command, which restarts the PNGTuber-GPT module and reloads global settings.
• Added the ability to retrieve all commands with !help
• Added functionality to the !version command so this can be customized to show whatever you like; for instance, version 202311. Just for an example.
• Fixed a bug where usernames would get mixed up due to race conditions during redeems
• Added better error handling for many edge cases.
• Created an all new logging engine that logs to a separate log file under 'Database Path'\logs.
• Created a logging level global settings under the Settings Action under Actions. This can be set to INFO, WARN, ERROR, and DEBUG. In DEBUG absolutely everything is logged, while in ERROR only errors are logged, with WARN, only warning and errors, and with INFO, everything that isn't debug. INFO is the default run mode for release. DEBUG is the default log mode for pre-release builds.
• Expanded the code base from 500 lines of code to over 2,600.
• Fixed issue where chat log was not sending properly
• Fixed issue where long messages were not splitting properly
• Fixed issue where GPTLog had no limit, resulting in overloading the GPT model
• Fixed issue with !modforgetthis where the variable inputRaw was incorrectly used when it should be rawInput. Thanks, Nintalia!
• Added !sayplay command, that simply makes the bot say !play in chat, and not speak it. This will allow your bot to play Marbles on Stream with your users, without being fussy about it.
• Fixed issue where moderation settings were not being applied correctly
• Fixed issue where username instead of nickname was being used for the Speak action
• Fixed issue where !modclearpromptlog was not working as intended
• Created new action "SaveSettings" this action creates a settings.json file in the database path to save all settings, so you don't need to redo them on each upgrade.
• To support the new settings.json file, a reference was added to System.Security.dll to provide encryption of the API key for storing in a plain text file
• There is also a new action Startup that executes on each run of Streamer.bot. This will re-enable all of the bots commands when run.

To install this version, after importing, navigate to the SaveSettings action. Enter in all of the required information, and then right click on the trigger at the top and hit Test Trigger. Verify that the settings.json file has been created in the bot directory. Update the database path variable under the Startup action. Restart Streamer.bot. Upon restart, everything should be working. For future upgrades, all you will need to do is change your database path under the Startup action, and then restart Streamer.bot, ignoring the SaveSettings action.

Download Release v1.1

-Fixed issue where moderation settings were not being applied correctly -Fixed issue where username instead of nickname was being used for the Speak action -Fixed issue where !modclearpromptlog was not working as intended -Created new action "SaveSettings" this action creates a settings.json file in the database path to save all settings, so you don't need to redo them on each upgrade. -To support the new settings.json file, a reference was added to System.Security.dll to provide encryption of the API key for storing in a plain text file -There is also a new action Startup that executes on each run of Streamer.bot. This will re-enable all of the bots commands when run.

To install this version, after importing, navigate to the SaveSettings action. Enter in all of the required information, and then right click on the trigger at the top and hit Test Trigger. Verify that the settings.json file has been created in the bot directory. Update the database path variable under the Startup action. Restart Streamer.bot. Upon restart, everything should be working. For future upgrades, all you will need to do is change your database path under the Startup action, and then restart Streamer.bot, ignoring the SaveSettings action.

Download Release 1.1-rc6

As the Lead Developer of PNGTuber-GPT, I find it imperative to discuss the recent developments at OpenAI and the significant ramifications these changes have on the nature and scope of our creations.

The dramatic events at OpenAI, encompassing the brief dismissal of CEO Sam Altman, have been a startling reflection of the tumultuous landscape in which we operate. The concerns surrounding OpenAI's advancements towards AGI, and the subsequent lack of transparency with its board, have cast a long shadow on the organization. While this corporate drama unfolds, our immediate concern lies with the implications of OpenAI's stringent content moderation policies, which resonate with the contentious approaches to governance seen at Twitter and the friction between Alphabet and ad-blockers on YouTube.

In these analogies lies a common thread relevant to our discourse: the delicate balance between safeguarding digital spaces and nurturing the authentic human experience within them. The update from OpenAI on November 2nd, 2023, aimed at tightening content safety, has directly impacted our ability to craft narratives and characters with the depth and realism that mature themes require. Imagine the constraints on video game developers if characters were barred from expressing themselves with the raw vernacular of profanity, or if the narratives could not touch upon mature themes such as sexuality or substance use. The impact is a diluted form of storytelling that fails to reflect the complexity of adult human interactions and experiences.

In our community, we see a divide: some users prefer PG content from our bots, while others advocate for a representation of the full spectrum of adult conversation. The latter is crucial for authenticity in various creative contexts, where characters must reflect the diversity of human expression, including the gritty, the raw, and the real.

Yet, we must also acknowledge the importance of safety, particularly in preventing hate speech and self-harm. The challenge lies in finding a middle ground where our AI can understand and adapt to the nuances of adult conversation without crossing into the realms of harm or abuse. The current moderation policies, though well-intentioned, may not fully appreciate the subtleties of context and intent, leading to a conservative application that hinders this balance.

Our current predicament echoes the broader societal debates over content moderation – a platform's responsibility to protect its users while also fostering an environment where freedom of expression thrives. This tension is palpable in our development community, as we strive to create characters that are as multifaceted and nuanced as the humans they emulate.

As we navigate this complex landscape, alternative platforms like Cloudflare's AI Workers beckon with the promise of more flexible and varied models that can be trained and tuned to our specific needs, reflecting the entire human experience without sacrificing our commitment to safety.

I invite you to join our discussions on our new Discord. There, we can deliberate on our collective future, seeking a path that respects our creative freedoms, acknowledges the spectrum of our community's desires, and upholds the necessary standards of safety.

Together, we can shape the trajectory of our platform, ensuring that it remains a space where creativity is nurtured, and all voices can find their authentic expression within safe boundaries.

-Fixed issue where chat log was not sending properly -Fixed issue where long messages were not splitting properly -Fixed issue where GPTLog had no limit, resulting in overloading the GPT model

Download Release 1.1-rc5

-Fixed issue where chat log was not sending properly -Fixed issue where long messages were not splitting properly -Fixed issue where GPTLog had no limit, resulting in overloading the GPT model

Download Release 1.1-rc4

Fixed issue where pronouns were returning incorrectly due to bad pronoun matching logic Added ability to remove nickname with !removenick Added ability to remove the thing you taught the bot about yourself with !modforget Added the ability to get the memory of what you taught the bot with !get memory Added the ability to clear the log file with !modclearlogfile Added the ability to clear the chat log with !modclearchatlog Added the ability to clear the prompt log with !modclearpromptlog Added !modreinitialize command, which restarts the PNGTuber-GPT module and reloads global settings. Added the ability to retrieve all commands with !help Added functionality to the !version command so this can be customized to show whatever you like; for instance, version 202311. Just for an example. Fixed a bug where usernames would get mixed up due to race conditions during redeems Added better error handling for many edge cases. Created an all new logging engine that logs to a separate log file under 'Database Path'\logs. Created a logging level global settings under the Settings Action under Actions. This can be set to INFO, WARN, ERROR, and DEBUG. In DEBUG absolutely everything is logged, while in ERROR only errors are logged, with WARN, only warning and errors, and with INFO, everything that isn't debug. INFO is the default run mode for release. DEBUG is the default log mode for pre-release builds. Expanded the code base from 500 lines of code to 2,135 an increase of 327% in a week. The code is now much better commented, summarized, and human-readable.

Download Release 1.1-rc3

This release fixes an issue where the custom nickname would fail to set when the user did not have pronouns configured.

Download Release 1.1-rc2

This is a massive release, that ended up being almost a complete rebuild. All of the settings have now been centralized under a Settings Action. Pronouns are now automatically supported when recording chat messages and prompting, trigger !setpronouns to learn more. You can now set a custom nickname that the bot will call you instead of your Twitch username. Set it with !setnick 'nickname'. You can also check how your current appearance looks to the bot with !currentnick. Mods can now add words to the channel context, by using '!rememberthis' 'keyword' definition. These definitions will automatically be added to the context anytime the keyword is used in a prompt. A new Speak option was added for mods and VIPs which allows the use of !speak 'phrase' and the bot will say "'User Nickname' said 'phrase'." Two new commands were added for !clearchathistory and !clearprompthistory, and both now have error handling and confirmation messages. This will help deal with naughty prompters that get the bot stuck misbehaving.

Long messages will now split across multiple chat messages, instead of just not logging to the chat. You can now log questions and answers directly to a Discord channel for your viewers to see later. Moderation categories are now individual settings you can toggle on and off at a whim. Keyword_Contexts and preferred_userNames will both be automatically created if the files don't exist. The only required file in your database path is the context.txt file. Anything else will get created as people redeem. Speaker.bot is now internally integrated through coding, instead of using actions that lead to weird variable handling like the infamous %response%. Errors are now more rich, and handled in dozens of new possible cases. Logging has been improved, and made more consistent. This one will probably be buggy, due to the nature of the changes. If you don't want bugs, you might want to wait for the official 1.1 release.

Moderation is now provided for ALL the things. Want to set a custom nickname, better pass moderation checks. Want to speak something out loud? Better pass moderation checks. The same moderation used for processing prompts, is now deployed to help you with the new features. You can take solace knowing that someone can't just make a ridiculous custom name if that's not your thing.

Download Release 1.1-rc1

We did it. Here is the first full release of PNGTuber-GPT. Thank you to all of the testers who made this possible. Development is going to continue. New features are planned and coming soon. Those on the latest release candidate do not need to upgrade. However, if you don't have the solution currently installed, this is the one that you want.

Download Release 1.0