- TodayAI
- Posts
- AI Interviewers Are Here
AI Interviewers Are Here
Plus, Should You Create an AI Version of Yourself on Instagram?
TodayAI
August 06, 2024
In this edition: AI Interviewers • Instagram AI Avatar • Kamala Harris Deepfake • $100B Models • Guardrails Easily Bypassed • Research Analyst Tasks • Energy Consumption
restofworld.org · 7 minute read
Companies are increasingly adopting AI interview systems, but doubts linger about whether these tools can fairly assess candidates.
“They don’t get angry or have a bad mood…it doesn’t matter what skin color you have, where you’re from, or what your accent is.”
theverge.com · 2 minute read
Meta’s new AI Studio lets users create AI versions of themselves for Instagram, but concerns about privacy and the potential for misuse remain.
The AI characters created can interact with followers, respond to comments, and even initiate conversations on behalf of the user.
Meta assures that all AI profiles are clearly labeled and that creators have control over what their AI can and cannot engage with.
ibtimes.co.uk · 3 minute read
The rapid spread of a deepfake video of Kamala Harris, shared by Elon Musk, has highlighted the growing threat of AI in manipulating public opinion during elections.
The video, which mimics a recent campaign ad by Harris, falsely portrays her making controversial statements she never actually made.
The deepfake was initially posted by a user who labeled it as "parody," but its realistic nature has raised concerns about its impact on viewers.
tomshardware.com · 4 minute read
The cost of training AI models is expected to skyrocket to $100 billion within the next few years, sparking debate about the sustainability and implications of such investments.
“If data center power requirements continue to grow exponentially, then it’s possible that we could run out of enough economically-priced electricity.”
theregister.com · 3 minute read
A recent test revealed that Meta’s new AI guardrails can be circumvented using a basic space bar hack, questioning the robustness of AI safety protocols.
Prompt injection attacks are designed to make AI models ignore their internal safeguards, leading to inappropriate or harmful outputs.
Meta’s system was fine-tuned to catch high-risk prompts, but this fine-tuning failed to account for basic manipulations of English language characters.
reuters.com · 2 minute read
JPMorgan has launched an in-house AI chatbot, designed to assist research analysts by handling tasks like writing and summarizing documents, with 50,000 employees already using the tool.
The AI chatbot aims to boost the efficiency of JPMorgan’s research analysts, allowing them to focus on higher-level tasks while the AI handles routine document-related work.
JPMorgan’s AI tool follows a similar rollout by Morgan Stanley, which introduced an AI-powered chatbot last year for its financial advisors.
wired.com · 7 minute read
Rising energy consumption is not only affecting data centers but also posing challenges to local power grids and water resources, with potential repercussions for communities.
Google and Microsoft, among others, are seeing their energy consumption spike as they race to develop more powerful AI models, with Google recently ceasing to claim carbon neutrality.
Researchers are concerned about AI’s growing environmental footprint, noting that energy consumption in AI could soon outpace many traditional sectors.
➡️ Follow on LinkedIn for additional highlights, pause-worthy quotes, and occasional sad attempts at comic relief 😂