EP05: AI and Olde Fashioned Scams, Distinguishing between AI-generated and Human-created Images

Howdy, prompt engineers and AI enthusiasts!

In this week’s issue…

…In this insightful episode of HTTTA, Wes and Goda explore the fascinating world of AI-generated images, examining how to differentiate between real and fabricated visuals. The co-hosts take on a 17-question test on caitfished.com, scoring well as they discuss various telltale signs such as asymmetrical eyebrows, mismatched earrings, and the manner in which light reflects off hair, as surefire ways to spot AI-generated photos.The podcast delves into the 2023 SYZYGY Digital Insight Survey. In the survey of over 1,000 participants, the co-hosts reveal that a mere 8% could accurately discern real images from AI-generated ones, while 57% exhibited a preference for the latter. However, only a third of respondents believed AI-generated images would contribute positively to the creative process.As opinions on the objective value of AI art vary, the episode highlights that viewer sentiment and exposure ultimately determine the success of such creations. They argue that despite AI's rapidly growing prominence, a gap persists in understanding and policy-making.Sharing stories of AI's role in scam detection, Wes recounts and instance where he narrowly escaped a fraudulent podcast interview scheme. By leveraging AI tools like Auto GPT, he was able to identify and avoid scams, despite the effectiveness of a scam that evokes emotions of excitement in the target. Wes shares how in his experience of nearly falling for this podcast scam, he was only rescued by his wife's skepticism, and of course, Auto-GPT. In the concluding segment, Wes and Goda discuss effective communication with AI, offering valuable advice on how to maximize AI-driven conversations.

SYZYGY Digital Insight Survey 2023: https://bit.ly/3HCRHpB 2023 HackAPrompt Competition going on Now! https://www.aicrowd.com/challenges/hackaprompt-2023

Let us know your score in the Comments: https://www.caitfished.com/

Podcast Page: https://howtotalkto.ai

HTTTA Newsletter:

MidJourney Master Reference Guide: bit.ly/3obnUNU 

ChatGPT Master Reference Guide: bit.ly/3obo7AG 

Learn Prompting: https://learnprompting.org/

Discord (Goda Go#3156 & Commordore_Wesmardo#2912)

Goda Go on Youtube: /@godago

Wes the Synthmind's everything: https://linktr.ee/synthminds

Key Take Aways from the Podcast:

1. AI-generated images can be difficult to identify as real or fake, especially in the age of altered reality filters and images.

2. AI agents should be managed, taught and nurtured like a child in their current states.

3. Even with its profound proliferation, there is still a profound gap in the mass awareness towards AI and its especially prevalent with the lack of regulation and policy-making circles.

1. How to spot AI-generated people in Photoreal Images?

With the introduction of Midjourney 5.1, we see yet another quantum leap forward in AI-Generated photorealism and quality. Now the model is loaded with additional refinements such as heightened precision, reduced incidence of unwanted borders or textual artifacts in images, and augmented image sharpness.

While the uncanny valley grows more narrow with each new version release, here are some of the ways AI-generated images fall short. The below image examples are all failed completions courtesy of Wes’ Midjourney.

Peculiar Hairstyles and Lighting on Hair or Fur Textures

It is true that some individuals possess unconventional hairstyles. However, AI-generated hair often appears as if it was digitally superimposed using software like Photoshop. Additionally, one might observe unusual strands across the forehead that resemble cracks more than hair. There might also be hair clumps that do not seem attached to the head or other strands. Long, straight hair may possess peculiar areas as well.

Uneven Ears or Mismatched Earrings

While there are those who choose to wear a single earring or different earrings on each ear, such a sight in a photograph could indicate an artificially generated face. Observe the ears themselves for disparities, such as different lengths or distinct earlobe attachments, which may suggest the image was produced by AI.

Inconsistent Eye Characteristics or Sizes

Heterochromia, or having two different eye colors, is an uncommon trait. As such, encountering this feature in an image might hint at artificial generation. Similarly, if the irises seem to vary in size, or the size/scale of one entire eyeball seems different than the other— all could be another clue.

Unusual Teeth Appearances/Alignments

AI often faces challenges in generating teeth, so if an individual's teeth appear peculiar or misaligned, it is likely an AI-generated image – unless, of course, the subject is Freddie Mercury. Regrettably, the opposite is not always accurate. AI may produce images with perfect teeth, yet the image could still be artificially created.

Atypical Background Elements

Algorithms that create facial images focus heavily on rendering realistic faces, often neglecting the background. Consequently, if the backdrop behind a person seems unnatural or inconsistent with a real location, it is a strong indication of an AI-generated image. Text, in particular, is prone to distortion in backgrounds; thus, if the mangled text is present, the image is likely fabricated.

Challenges with Realistic Hands and Fingers

Our hands possess incredible dexterity, with numerous ways they can be positioned, along with their ten distinct digits. As a result, AI-generated images may struggle to accurately represent hands and fingers. In some instances, the images may exhibit awkward finger placement, unnatural lengths, or improper positioning. These inconsistencies further reveal the limitations that AI models still face when attempting to generate authentic human representations.

Start writing today. Use the button below to create your Substack and connect your publication with How to Talk to AI

2. AI Scams, What’s Out There and How to Defend Against

While we preach and promote a cautiously optimistic approach to the wonders of all things AI, it’s undeniable that just as these AI tools can be used to uplift, teach, create, and for good purposes, so too can they be deployed for evil and bad intent. From the US Federal Trade Commission’s website, they define Generative AI and synthetic media are colloquial terms used to refer to chatbots developed from large language models and to technology that simulates human activity, such as software that creates deep fake videos and voice clones. Evidence already exists that fraudsters can use these tools to generate realistic but fake content quickly and cheaply, disseminating it to large groups or targeting certain communities or specific individuals. They can use chatbots to generate spear-phishing emails, fake websites, fake posts, fake profiles, and fake consumer reviews, or to help create malware, ransomware, and prompt injection attacks. They can use deepfakes and voice clones to facilitate imposter scams, extortion, and financial fraud. And that’s very much a non-exhaustive list.Here’s just a brief rundown of some more of the examples referenced above, because awareness is power in combatting these scammers using AI:CHATGPT SCAM PHISHING EMAILS: When a threat intelligence researcher at a cybersecurity company prompted ChatGPT to produce an example of a tax scam email containing malware, the AI chatbot spit back a grammatically immaculate email on the Employee Retention Credit that asked the recipient for their Employer Identification Number, payroll information and a list of their employees and their social security numbers.

According to new research from digital security firm WithSecure, this very technology has demonstrated an uncanny proficiency in drafting phishing emails. These deceptive messages are meticulously designed to manipulate recipients into revealing sensitive data, such as banking credentials and passwords.

VOICE CLONING AI SCAM: You get a call. There's a panicked voice on the line. It's your grandson. He says he's in deep trouble — he wrecked the car and landed in jail. But you can help by sending money. You take a deep breath and think. You've heard about grandparent scams. But darn, it sounds just like him. How could it be a scam? Voice cloning, that's how.This is one of the most pernicious AI tools in the new suite used by the scammers. In January 2023, Microsoft unveiled a text-to-speech AI tool that could simulate a person’s voice based on an audio clip as short as three seconds. Microsoft called it VALL-E and said that the technology “may carry potential risks in misuse… such as spoofing voice identification or impersonating a specific speaker,” as reported by Fortune magazine.

Bad actors saw this as an opportunity and used it to spoof the voices of people’s friends and family. They have been convincing others to wire them money, send them gift cards or give them personal account information. If you’re someone who has a large amount of audio of yourself online, you’re a prime target for a ‘voice cloning’ scam, as these AI tools download, break it down into individual words and then allow bad actors to type out the phrases they want to you to say. Eleven Labs is a leading researcher into AI-voice reproduction technologies.A DEEPLY TROUBLING DEEPFAKE TREND: In recent months, there has been a significant increase in the utilization of generative artificial intelligence tools that enable users to manipulate videos, synchronizing facial movements with AI-generated voice cloning. Deepfakes of this nature are increasingly prevalent on social media platforms such as Twitter.

A recent instance involved an AI-engineered deepfake video of Joe Biden that gained widespread attention for all the wrong reasons. The seemingly authentic, two-minute video featured Biden directing harsh, derogatory remarks toward members of the LGBT-Q community. A lighter example depicted Biden joking about the naming of the 'Sneed's Feed and Seed' store.

While many tech-savvy individuals may be able to identify deepfakes upon closer examination, it begs the question: if a cyber attacker created a video of you, similar to the one of Joe Biden, could you be confident that your family and friends would discern fact from fiction?AI-ENHANCED ROMANCE SCAMS: While the recent celebration of Valentine's Day is only a few months behind us, the alarming rise of romance scams - wherein malicious actors feign romantic intent and gradually build trust with their victims before soliciting funds - presents an area in which artificial intelligence could substantially amplify the capabilities of these unscrupulous individuals.

It is becoming increasingly evident that cybercriminals are starting to harness the power of AI-driven chatbots to send messages under the guise of would-be romantic partners. These deceptive schemes have already gained considerable traction, with a recent study estimating that individuals in Australia lost over $27 million in 2022 alone to these romance scams.

3. ELI5 AI Term of the week: “Self-Attention”

Okay, imagine you're playing with a group of friends, and you all have a bag of toys. Each toy is different and special in its own way. Now, it's time to pick a toy to play with, but you can't decide which one is the best.

"Self-Attention" is like a magical helper that looks at all the toys in everyone's bags, and then helps you choose the best toy to play with. It does this by paying more attention to the most important toys and less attention to the less important ones. This way, you can have the most fun playing with the toys that matter the most!

4. Google "We Have No Moat, And Neither Does OpenAI". How Open-Source LLM’s will Overtake the Current Market Leaders.

Due to the provacative yet important nature of this article, we’ve included some key segments in their entirety of this Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI from an article written by DYLAN PATEL and AFZAL AHMAD. The text below is a very recent leaked document, which was shared by an anonymous individual on a public Discord server who has granted permission for its republication. It originates from a researcher within Google and its authenticity has been verified. The only modifications are formatting and removing links to internal web pages. This article is simply a vessel to share this document which raises some very interesting points.

The entirety of the writtings below are the verbatim opinions of a Google employee, and not representative of the entire firm. We’ve done a lot of looking over our shoulders at OpenAI. Who will cross the next milestone? What will the next move be?

But the uncomfortable truth is, we aren’t positioned to win this arms race and neither is OpenAI. While we’ve been squabbling, a third faction has been quietly eating our lunch.

I’m talking, of course, about open source. Plainly put, they are lapping us. Things we consider “major open problems” are solved and in people’s hands today. Just to name a few:

• LLMs on a Phone: People are running foundation models on a Pixel 6 at 5 tokens / sec.

• Scalable Personal AI: You can finetune a personalized AI on your laptop in an evening.

• Responsible Release: This one isn’t “solved” so much as “obviated”. There are entire websites full of art models with no restrictions whatsoever, and text is not far behind.

• Multimodality: The current multimodal ScienceQA SOTA was trained in an hour.

While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:

• We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.

• People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.

• Giant models are slowing us down. In the long run, the best models are the oneswhich can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

What Happened

At the beginning of March the open source community got their hands on their first really capable foundation model, as Meta’s LLaMA was leaked to the public. It had no instruction or conversation tuning, and no RLHF. Nonetheless, the community immediately understood the significance of what they had been given.

A tremendous outpouring of innovation followed, with just days between major developments (see The Timeline for the full breakdown). Here we are, barely a month later, and there are variants with instruction tuning, quantization, quality improvements, human evals, multimodality, RLHF, etc. etc. many of which build on each other.

Most importantly, they have solved the scaling problem to the extent that anyone can tinker. Many of the new ideas are from ordinary people. The barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person, an evening, and a beefy laptop.

Why We Could Have Seen It Coming

In many ways, this shouldn’t be a surprise to anyone. The current renaissance in open source LLMs comes hot on the heels of a renaissance in image generation. The similarities are not lost on the community, with many calling this the “Stable Diffusion moment” for LLMs.

In both cases, low-cost public involvement was enabled by a vastly cheaper mechanism for fine-tuning called low-rank adaptation, or LoRA, combined with a significant breakthrough in scale (latent diffusion for image synthesis, Chinchilla for LLMs). In both cases, access to a sufficiently high-quality model kicked off a flurry of ideas and iteration from individuals and institutions around the world. In both cases, this quickly outpaced the large players.

These contributions were pivotal in the image generation space, setting Stable Diffusion on a different path from Dall-E. Having an open model led to product integrations, marketplaces, user interfaces, and innovations that didn’t happen for Dall-E.

The effect was palpable: rapid domination in terms of cultural impact vs the OpenAI solution, which became increasingly irrelevant. Whether the same thing will happen for LLMs remains to be seen, but the broad structural elements are the same.

We need them more than they need us

Keeping our technology secret was always a tenuous proposition. Google researchers are leaving for other companies on a regular cadence, so we can assume they know everything we know, and will continue to for as long as that pipeline is open.

But holding on to a competitive advantage in technology becomes even harder now that cutting-edge research in LLMs is affordable. Research institutions all over the world are building on each other’s work, exploring the solution space in a breadth-first way that far outstrips our own capacity. We can try to hold tightly to our secrets while outside innovation dilutes their value, or we can try to learn from each other.

Individuals are not constrained by licenses to the same degree as corporations

Much of this innovation is happening on top of the leaked model weights from Meta. While this will inevitably change as truly open models get better, the point is that they don’t have to wait. The legal cover afforded by “personal use” and the impracticality of prosecuting individuals means that individuals are getting access to these technologies while they are hot.Directly Competing With Open Source Is a Losing Proposition

This recent progress has direct, immediate implications for our business strategy. Who would pay for a Google product with usage restrictions if there is a free, high quality alternative without them?

And we should not expect to be able to catch up. The modern internet runs on open source for a reason. Open source has some significant advantages that we cannot replicate.HTTTA will be following the coverage of this leaked internal opinion with an upcoming episode on Open-Source language models

5. Prompts, served Hot and Fresh weekly

The one prompt to rule them all. This is a baseline all-in-one prompt generator. Simply answer the question “I wish I had a prompt that…” and be amazed with the results. Need long-form content, done. A master key Midjourney prompt generator, easy day. A gourmet recipe creator, coming right up. The possibilities are truly endless. Happy Prompting!

# The prompt building prompt

In Conclusion

As we enter the golden era of artificial intelligence, we are simultaneously confronted with a new generation of digital fraudsters, equipped with a myriad of AI-powered tools. From eerily authentic fabricated media to software that meticulously combs through data, malicious actors have never had such a vast arsenal to target unsuspecting individuals. It is crucial to remain vigilant and ensure that those around you are well-informed about the challenges posed by this rapidly advancing technology.

While we await the cybersecurity industry as well as federal and state regulators to catch up with the accelerated progress in AI, the responsibility falls upon us, the everyday users, to stand guard against the impending surge of AI-driven scams. Stay safe and have fun using these awesome tools. Happy Prompting Everybody!