- Reality Bytes
- Posts
- GPT Mimicking User Voices, Secret Code from Sam Altman, & New Ethical AI
GPT Mimicking User Voices, Secret Code from Sam Altman, & New Ethical AI
Abby Bradeen
August 13, 2024
AI
ChatGPT Just Got Creepier
GPT Mimicking User Voice, Unprompted: As if we weren’t all wary enough of AI, a recent update from OpenAI on safety procedures and model details for GPT-4o revealed a highly concerning bug. During testing for GPT-4o, the model went rogue and began speaking back to users in their own voices, without permission or prompting.
In an audio clip of one incident, we see GPT mimic a user’s voice after she entered a voice prompt into the Advanced Voice Assistant. The user speaks, and GPT responds normally for a moment before loudly exclaiming “NO!” and finishing the response in the user’s own voice.
OpenAI recognizes the potential fraud risks that this bug poses, and has taken steps to mitigate the issue by only allowing GPT to use approved presets generated from voice actors. However, the Voice Assistant feature is innately risky, as it relies on the model’s ability to synthesize sounds after even a short audio input.
Staying informed about AI developments like this one can help business leaders better manage the risks and opportunities presented by emerging technologies, ensuring that they remain at the forefront of innovation without compromising security or ethics.
AI
Secret Message from Sam Altman or Just Schizo Posting?
Twitter Thinks Altman is Dropping Hints about GPT5: A few months back, Reuters unveiled an exclusive sneak peek into a secret OpenAI model code-named “Project Strawberry.” OpenAI claims that this model will be the most advanced LLM yet, with high level reasoning skills and the ability to complete “deep research” autonomously.
Most experts acknowledge that one of the most significant stumbling blocks for AI is that it lacks the ability to “plan ahead”, meaning that it still can’t rival human reasoning capabilities. However, it seems that Project Strawberry is intended to be a giant leap forward in this area.
On Wednesday, Sam Altman took to Twitter, posting a photo of strawberries with the ambiguous caption “i love summer in the garden”. AI fanatics immediately started coming up with theories about the meaning behind the post, with one user even counting the strawberries and commenting:
“Five strawberries, but the fifth is not ripe yet 🍓 GPT-5 confirmed within 4-6 weeks 🍓”
This is based, of course, on the average time it takes actual strawberries to ripen. Say what you will about Sam Altman, the man apparently has a pretty dedicated fan base.
AI
A New Kind of Credit Check
Startup Tries to Mitigate AI Plagiarism: In a bold move to address the ethical concerns surrounding generative AI, tech entrepreneur Bill Gross has launched ProRata, a startup that aims to ensure fair compensation for creators whose work is used by AI models. ProRata’s innovative chatbot and search engine, set to debut in October, will use a patented algorithm to break down AI-generated content, identify its original sources, and attribute a percentage of the content to each copyright holder.
Backed by $25 million in funding and partnerships with media giants like Universal Music Group and The Atlantic, ProRata’s approach is a game-changer. Unlike other AI models that scrape content without permission, ProRata will only use licensed data, promising higher-quality outputs and ethical AI practices. Gross’s vision extends beyond ProRata’s standalone success; he plans to license this technology to major AI firms, setting a new standard for fair compensation in the industry.
