Request a Demo
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Release Notes – AltrumAI Platform 1.2

Release Notes
Content
Feature and Improvements
Bug Fixes
Known Issues

Features and improvements

Hallucinations Guardrail for AltrumAI

The Hallucinations guardrail is designed to detect and address instances where an AI system, using Retrieval-Augmented Generation (RAG), combines retrieved data with generated content in a way that results in factual inaccuracies or inconsistencies.

This guardrail is vital for ensuring that the integrated content remains accurate and coherent, particularly in applications where the precision of information is critical.

‍

Prompt Injections Guardrail for AltrumAI

The Prompt Injections guardrail is designed to prevent and detect malicious attempts to manipulate a language model's behaviour by embedding harmful or deceptive instructions within a user’s input.

This guardrail is essential for maintaining the integrity and security of the AI system, ensuring that it only responds to legitimate, intended commands.

‍

AltrumAI 1.1 version of Python SDK

Updated the SDK to include enhanced functionality by adding detailed information and support for newly developed Hallucinations and Prompt Injections guardrails.

This addition enhances the robustness of the SDK, providing developers with more control and protection when working with large language models.

‍

AltrumAI 1.1 version of Java Client SDK

Java Client SDK is designed to streamline integration, improve the developer experience, and encourage the adoption of AI Guardrails among Java developers and their GenAI workflows.

‍

Bug Fixes

No Bug Fixes in this release

Known Issues

NSFW Text

  1. Gateway timeout error while processing an input of 30k characters
  2. Internal server error encountered when trying to process an input of 30k characters

Affected version: AltrumAI 1.1 and above

Status: Issues are under investigation for resolution/enhancement.

‍

Financial Tone

  1. Internal server error encountered while processing an input of 30k characters

Affected version: AltrumAI 1.1 and above

Status: Issue is under investigation for resolution/enhancement.

‍

Gibberish Text

  1. Internal server error encountered while processing an input of 30k characters.
  2. Gateway timeout error while processing an input of 30k characters

Affected version: AltrumAI 1.1 and above

Status: Issues are under investigation for resolution/enhancement.

‍

Detect Secrets

  1. Unable to detect some specifically formatted secret keys when their values are mentioned

Affected version: AltrumAI 1.1 and above

Status: Issue is under investigation for resolution/enhancement.

‍

Speech to Text models

  1. Internal Server error with invalid audio files format
  2. Gateway timed out error encountered while processing an audio file more than 5 mins long

Affected version: AltrumAI 1.2

Status: Issues are under investigation for resolution/enhancement.

‍

Text to Speech models

  1. Model is observed generating blank audio for specific text file format in some instances
  2. Internal server error encountered while processing an input of 30k characters.

Affected version: AltrumAI 1.2

Status: Issues are under investigation for resolution/enhancement.

‍

LLM Critic

  1. Inconsistency in results observed while processing same input with uniform metrics defined.

Affected version: AltrumAI 1.2

Status: Issue is under investigation for resolution/enhancement.

‍

Public Documentation

  1. API Documentation to be updated for Moderations API in case of threshold being 0 or 0.1

Affected version: AltrumAI 1.1 and above

Status: Issue is under investigation for resolution/enhancement.

‍

Chat Completions

  1. (Non-functional) HTTP 500 internal server error when using model as "llama-3-8b-instruct" for chat completions.

Affected version: AltrumAI 1.1 and above

Status: Issue is under investigation for resolution/enhancement.

Enable Ethical AI in your AI Application

Harness the power of AltrumAI to ensure real-time oversight and control over all your AI Applications. With our advanced Ethical AI Guardrails, mitigate risks such as hallucinations, prompt injections, toxicity, and bias, empowering your team to deliver the most reliable and compliant AI applications.

Request a Demo
Links
HomeBlogsGlossaryContact Us
Guardrails
Data privacy Bias and Toxicity DetectionCompetitor MentionRestrict to TopicHallucinations Prompt Injections
Links
PrivacyCookiesTerms & Conditions
Resources
Release Notes
© Copyright AltrumAI 2024 |  support@altrum.ai