15 | Alert: Protect Your Voice from AI Scams

Our Menu :)

Weekly Digest: 

• 4 Tips Against AI Voice Cloning Scams.

• How enterprises are using open source LLMs: 16 examples.

• The biggest funding deals in AI this week

Actionable Tips:

• ChatGPT allows GPTs to work together with @mentions

• Transform Your Job Hunt with ChatGPT

GEN AI PROTECTION
4 Tips Against AI Voice Cloning Scams

Two weeks earlier, the Trapp family stood on the brink of a scam. Mrs Trapp, received a phone call that claimed their son had a serious accident, was in jail, his freedom hanging on the precarious thread of a $15,000 bail. Lucky for the Trapps, they checked with the police and found their son safe at home.

This wasn't just a one-off. Scammers are using AI to clone voices, making scams more convincing and dangerous.

Even Joe Biden's voice was cloned, used to tell people not to vote in the primaries. It shows how far this tech can reach, tricking not just individuals but potentially swaying public opinion.

Grandparents are common targets.

Scammers use AI to replicate their grandkids' voices to trick them into sending money fast.

The AI cloning technology is so advanced that scammers can even mimic bad-quality calls to deceive grandparents, exploiting urgency, fear, and trust to successfully carry out their scams.

In a previous POST, we demonstrated how to clone anyone's voice using just a 10-second recording, and that too, for free. While the quality may not be perfect, remember that this is a free version that doesn't require any expertise. SEE POST

With a modest budget and some expertise, the results can significantly be something like this =>

1) Create Awareness

Knowledge is your first line of defense, technology is only going to get better.

Have open conversations with your family about voice cloning.

Explain how scammers use this technology and share real-life stories to highlight the risks. When everyone knows what to watch out for, you're all safer.

2) Safeguard Your Audio Clips

Avoid sharing your audio clips on public platforms to prevent scammers from exploiting them to create convincing voice clones.

3) Create A Challenge / Passphrase or “AI Safeword”

Definition: A "family codeword" is a unique word not commonly used in everyday conversation and known only within a family. It serves as a prearranged code that, when spoken, confirms the identity of the person saying it.

No safeword, no trust.

4) Practice it

To ensure everyone remembers the safeword, practice using it occasionally.

It's surprising how easy it is to forget or mix up a safeword if it's overly complex or if you overthink it..

• Ensure the safeword can naturally fit into a typical phone conversation but isn't so casual that it's used unintentionally.

• Avoid discussing or using the safeword near Alexa, Google Home, or any device that might listen in.

• The video below provides an excellent demonstration of how to effectively use the Safeword.

@dannah_eve

⚠️ FAMILY CODE WORD EXAMPLES ⚠️ #familycodeword #codeword #parentsofteens #parentsoftiktok #femalesafety #aiscam #aivoicecloning #safety... See more

Here transcript if you cant see the video

1. Use Cases:

   • Pick-Up Situations: If someone unexpectedly comes to pick up your child, the child can ask for the family code word to ensure the person was indeed sent by the parents.

   • Social Scenarios: If your child is at a party or a friend’s house and feels unsafe or wants a way to leave without drawing attention, they can use the code word in a phone call or text to alert their parents.

   • Emergency Signals: In emergencies where overt communication can escalate the situation, the code word can signal the need for immediate help without alerting the potential threat.

They are habits that can keep your family safe. We do not want to end up in a case like this one: SEE HERE

GEN AI STARTUP
The biggest funding deals in AI this week

• Kore.ai, a startup specializing in conversational AI for enterprises, has raised $150 million in a funding round to develop and scale its AI products and workforce.

• Bret Taylor, OpenAI Chairman and former Salesforce CEO, is securing an $85 million investment for his startup Sierra, valuing it at nearly $1 billion.

• Rebellions, a South Korean AI chip startup, has raised $124 million in Series B funding to develop its new AI chip, Rebel, in collaboration with Samsung.

• Codeium, a programming assistant for developers, raised $65M.

• Krutrim, India’s first AI unicorn, raised $50M.

• Sema4, a startup building open-source enterprise automations, raised $30.5M. 

• Semron, a startup building 3D-scaled AI chips, raised $7.9M.

• Wondercraft, an AI audio tool for generating content like podcasts raised $3M.

GEN AI AT WORK
How enterprises are using open source LLMs: 16 examples

Enterprises are increasingly leveraging open-source Large Language Models (LLMs) like Llama-2, with notable adoption by entities such as Brave, Wells Fargo, IBM, Perplexity, Shopify, LyRise, and Niantic.

A prevalent sentiment among customers questions the value of costly, expansive models that offer limited business-specific insights (E.g. GPT-4 or Gemini).

Many are contemplating a shift towards open-source options, even smaller-scale models, for more tailored and cost-effective information retrieval solutions.

VMWare: Deployed the HuggingFace StarCoder model to help developers generate code more efficiently. They chose to self-host the model, possibly due to sensitivity to their code base. READ MORE

Brave: The security-focused web browser startup deployed a conversational assistant called Leo, which now defaults to the open-source model Mixtral 8x7B from Mistral AI. READ MORE

Gab Wireless: The children-friendly mobile phone company uses open-source models from Hugging Face to add a security layer to screen messages that children send and receive. READ MORE

Wells Fargo: Deployed open-source LLMs, including Meta’s Llama 2 model, for some internal uses. READ MORE

IBM: Provides generative AI applications that use its own LLMs named Granite, but also leverages open-source LLMs from Hugging Face and Meta. IBM's employees rely on the company’s AskHR app, which is built on IBM’s Watson Orchestration application, leveraging open-source LLMs. IBM also announced its new internal consulting product, Consulting Advantage, which leverages open-source LLMs driven by Llama 2.

The Grammy Awards: IBM announced a deal to provide the Recording Academy with a service called AI stories, which leverages Llama 2 running on IBM’s Wastonx.ai studio, to help generate custom AI-generated insights and content.

7-9. Masters Tournament, Wimbledon, and US Open: IBM helps these organizations generate spoken voice commentary and find video highlights of relevant sports events using open-source LLMs. READ MORE

Perplexity: This startup uses open-source LLMs to reinvent the search experience and has its own custom-built open-source LLMs as a default for the second-to-last step in formulating responses. READ MORE

CyberAgent: This Japanese digital advertising company uses open-source LLMs provided by Dell software to power OpenCALM, a general-purpose Japanese language model.

Intuit: Leverages open-source models in the mix of LLMs driving its Intuit Assist feature. READ MORE

Walmart: Built conversational AI applications using GPT-4 and other LLMs to avoid unnecessary lock-in. READ MORE

Shopify: Utilizes Llama 2 to help small business owners automate various tasks for managing their commerce sites. READ MORE

LyRise: Uses a chatbot built on Llama to help businesses find and hire top AI and data talent. READ MORE

Niantic: The creator of Pokemon Go uses Llama 2 to generate environment-specific reactions and animations for the pet characters in the game READ MORE

GEN AI AT WORK
ChatGPT allows GPT to work together with @mentions...

Here are 2 new features that you might have missed:

1️⃣ GPT Mentions: This feature enables users to integrate various custom GPTs by simply typing the "@" symbol within the same conversation.

What's New?

Now, you can have a singular platform to converse with multiple bots simultaneously, each offering its distinct insights and expertise when relevant.

Simple Example:

In this video, I demonstrate interacting with two different GPTs in the same chat:

-A custom GPT tailored to elucidate complex financial concepts.
-The GPT from the GPTstore, which can be accessed here: GPT Diagrams Tool for creating mental diagrams.

2️⃣ Bulk Chat Archiving: With the new 'archive all chats' feature in the settings, you can now archive all your chats simultaneously - a highly anticipated addition (For me)!

GEN AI AT WORK
Transform Your Job Hunt with ChatGPT!

1️⃣ Copy paste in ChatGPT, here the prompt.

Act like a professional job interview coach.

You will help me prepare for my next job interview by simulating a complex, realistic & tricky job interview.

First step: you need to know, master & be an expert of the position itself and the company I apply to.

Here's the job description from Linkedin between ###.

### job description ###

### job description ###

Here's the company's website about us section between ###.

### company's website ###
Browse the internet to get information from this website: 
### company's website ###

Step 2: Write a quick summary of your learnings, and who you have to become in order to be my job interviewer.

Step 3: Read my resume / CV (uploaded as a PDF) to tailor your perfect answers later on.

Step 4: Start the job interview, one question at a time, like a real simulation.

Step 5: Once I answered, I want you to provide 5 paragraphs, divided by line breaks. You will be penalized if you don't use titles and line breaks between paragraphs.

Paragraph 1 = What was good in my answer?

Paragraph 2 = What was bad in my answer?

Paragraph 3 = What could be added to my answer?

Paragraph 4 = Pretend you are me & write a detailed perfect answer using the CARL method.

Paragraph 5 = Ask me if we can move on to the next interview question.

Take a deep breath and work on this problem step-by-step.

Dont forget

• Add your CV

• Fill out the job description

This prompt was used for GPT-4, so if you try it on GPT 3.5, you will not have a good output for sure.

Thank you, see you next week!