Product Labs🔬: Decoding “Meta AI”

Spark is the quintessential social butterfly and lives on WhatsApp, so you could say we have a power user in the house.

A new Siri-like icon😋 popped up on Spark’s WhatsApp one day which led to us getting access to Meta AI. Unfortunately, neither Spark nor Trouble have access to Meta AI on the other platforms i.e. Facebook, Messenger and Instagram but that’s never stopped them from getting the latest scoop.

What’s in it for you?

Meta AI is an integration of Llama 3 on all social media & messaging platforms owned by Meta - WhatsApp, Instagram, Messenger, and Facebook. It can handle various types of conversations, fetch real-time information, generate images, translate text and can be used in group chats or as a standalone bot to ask questions and generate images.

The Meta AI bot icon is ‘easily discoverable’ on the app’s home screen, making it super-obvious for users to find & engage. This is in total contrast to Myntra’s AI chatbot Maya which is buried somewhere inside the app - poor discoverability. (Check out the takes on Maya here!)

Imagine you created a WhatsApp group to plan a vacation with your friends and need some details about the place you want to visit. Until now, you would’ve had to navigate to a search engine or, as is the norm today, to any gen AI chatbot and ask the questions. However, now you can call the bot with “@Meta AI” and ask anything. Internally, Meta AI uses Google and Bing to search the web (if needed) & generate answers to your questions. All this works in group chats as well as personal messages!

Through analysis of user behaviour across its social media apps, Meta has understood & capitalized well on the fact that all humans inherently want the easiest way out aka “The Principle of Least Effort”. By integrating Meta AI into apps, the two-hop process of users navigating to another app to find out information has been enabled. Users now have almost a one-stop platform, to ask questions, and make plans as they need on the fly.

One major letdown is Meta AI’s staged rollout of the experience - not everyone has access (ahem Trouble 😑)! So, if you invoke Meta AI in any conversation, while the other users do not have access yet, it could lead to a bad user experience.

Beyond mere conversations

Meta AI functions the same way, with the same set of functionalities across personal messages, group conversations or in the stand-alone bot. These are the few commands that Meta AI showed when asked for a list of all commands:

• /imagine: To generate images

• /reset-ai: Delete details from the current AI chat

• /reset-all-ais: Delete all AIs from meta servers

• /download-all-ai-info: Download all info

• /ai-options: Get the list of all possible commands

However, there are also a few additional commands like /joke & /story.

The autosuggest however shows only “/imagine” when you type a “/”, while the rest of the commands are all hidden on the official FAQ page (honestly, it took Spark a decent amount of time to figure this out!).

Looks like Meta AI has taken Hick’s Law a little too seriously, to help users save time in making a decision by hiding all possible options except one.

For everyone who got overwhelmed seeing the word “command”, take a sigh of relief. Meta AI works just as well, even if you don’t use these, thereby kinda forcing us to question why are they even required. 🤔

You can tell Meta AI “reset messages” or use the command “/reset-ai”, which also works out just fine. The same flies for image generation use “/imagine” or simply say “create an image of …”.

Now, to test the Davinciness of Meta AI

Since Meta AI is free, it would only be fair to compare it to other free AI image generators such as Stable Diffusion and Microsoft Copilot.

We tried the same prompt across all 3 to generate the same image. The image from Meta AI is as good as Stable Diffusion. Its ability to produce images leans towards a more photorealistic aesthetic.

However, there are a few drawbacks:

• Meta AI does NOT allow you to fine-tune the images post creation such as select off-the-shelf image style (pencil art, claymation, comic style, pixel art, etc.) which are possible in both the others. This may cause unnecessary overheads for newbies who are not “pro” in the keywords required (or even possible) for AI-powered image generation.

• Perhaps the biggest drawback lies in Meta AI's handling of subsequent image prompts based on an original creation. While it excels at the initial image generation, follow-up attempts to modify or build upon the original often fall short. This issue while is not the case always, was quite easy to run into.

If you thought that’s all you are in for a surprise. Meta AI goes beyond just generating images to blowing some life into them aka animating it. Yes, you can now create custom GIFs right away!🤩

It’s quite simple, generate an image and then ask Meta AI to animate it. Yet again, there was no way of knowing this feature exists except after reading quite a few blogs on Meta AI integrations.

What’s the intrigue?

What’s the biggest concern for all of us as AI continues to be so deeply integrated into our lives?

3…2...1... - PRIVACY

Spark is extremely protective of WhatsApp (so is of course almost everyone). It feels like our private space and you don’t want anyone sneaking a peek let alone AI running amok.

Meta AI dispels this concern to a great extent by explicitly stating that it cannot access chats or messages. It can only be used for informational intents & image generation.

You cannot ask Meta AI to summarize messages in a group or quickly catch you up on what you missed. This is quite contrary to Microsoft’s Copilot integration in its messaging app Skype. Copilot on Skype, unlike Meta AI, can skim through conversations and assist the user.

With Meta AI a new paradigm of AI-integrated superapps has opened up. The master stroke of integrating with Meta’s existing platforms has given access to a readily available large user base. With millions interacting with Meta AI across various platforms like Messenger, WhatsApp, and Instagram excitement builds organically and this acts as a springboard for growth.

However, despite Meta AI not having access to messages, Spark & Trouble still believe the question looms large of where the line between AI & privacy is.

Whatcha Got There?!🫣

Buckle up, tech fam! Every week, our dynamic duo “Spark” ✨ & “Trouble”😉 share some seriously cool learning resources we stumbled upon.

Your Wish, Our Command 🙌

You Asked 🙋‍♀️, We Answered ✔️

Question: You featured multiple AI tools for video generation in your previous editions. With such free and easy access to these tools for all, don’t you think it’s very easy to exploit them for spreading misinformation about an individual?

Answer: As we always stress while talking about any technology, it is very necessary to respect its applications. While the intent of all AI tools is to democratize content creation, the era of AI is also becoming the era of deepfakes. To mitigate risks, developers and platforms often implement guidelines and detection systems to prevent misuse. Additionally, educating users about media literacy and the ethical use of AI tools is essential in combating misinformation. It’s a collective effort that involves creators, users, platform providers, and regulatory bodies to ensure that the power of AI is harnessed for positive and constructive purposes, rather than harm.

Fortunately, there's also growing research on AI-powered detection methods to combat this. Tools & platforms like Sentinel & Sensity AI analyze content for manipulation signs. Researchers are building massive datasets to train AI models for better detection. The battle against deepfakes is an ongoing arms race, but AI offers promising solutions. We'll explore some of these in future issues.