AI
- where they are
- with a story that resonates
- while putting them at the center
- and sharing the energy you’re made of
- It's Google I/O, which means Big G has announced one billion things with AI. Including Search Chat* and its response to Open AI's GPT models: say hello to PaLM 2
- Chip Wars, reignited: Microsoft is getting in on the custom silicone game with an in-house design tailor made for AI & LLMs (chips are still one of my main areas to watch).
- An excellent AI thought experiment: envisioning LLMs as vast cities. "Think of it as the opposite of the Turing test setup: instead of putting a machine in a context where you can’t immediately tell whether or not it’s a human, build it in a way that’s inescapably, viscerally machine-looking."
- Focus on outcomes over means
- Keep the goal a little abstract
- But make sure it contains a concrete number(s)
- Early on, focus on how far you’ve come. The closer you get to completing your goal, look forward to what’s remaining
- Too many incentives is demotivating
- A few randomized, surprise incentives is motivating (like slot machines and social feeds)
The Future OS: Confirmed
I posted earlier this week about the new ChatGPT multimodal and how it's a glimpse into the future of how we interact with computing devices.
Kyle Mensing
The next day, the All-In Podcast talked about much the same thing.
I think they—via the illustrations below—prove this belief out even more. (They also mention that inputs can (or will) have more options than just images and voice. Code Interpreter let's you input documents and code snippets, so that makes sense.)
Humans have five major sensory inputs we use to feed our brain computers. Up to this point, we've had one input type to interact with computers: text.
Multimodal is a way of mapping our multi-sensory experience onto interactions with computers, making them more natural for us.
We've adapted ourselves to the limits of computing. Now we might be able to adapt our computers to ourselves.
I don't love the examples OpenAI used in their demos. I see the hints of what they promise, but they lack oomph. This example though, this perfectly shows how you can use multimodal in your day-to-day to make life a bit easier:
I will never get a parking ticket again. pic.twitter.com/yl7ND2rJeQ
— Peter Yang (@petergyang) September 27, 2023
To drive this point home even further, here's one of the guys who'd be on the Mount Rushmore of modern AI development sharing his take:
With many 🧩 dropping recently, a more complete picture is emerging of LLMs not as a chatbot, but the kernel process of a new Operating System. E.g. today it orchestrates:
— Andrej Karpathy (@karpathy) September 28, 2023
- Input & Output across modalities (text, audio, vision)
- Code interpreter, ability to write & run… pic.twitter.com/2HsyslOG2F
TLDR looking at LLMs as chatbots is the same as looking at early computers as calculators. We're seeing an emergence of a whole new computing paradigm, and it is very early.
-Andrej Karpathy
And Google's BARD can now interface with a host of Google apps and services, almost creating a true smart, personalized assistant.
How do we extend our brand experiences into this new environment?
What does "brand" mean in this future?
(Basically, take all the questions people had about brands and the metaverse after the Grand Meta Rebranding™ and replace "metaverse" with "multimodal AI chat".)
The Future: Consumer Computing
OpenAI announced ChatGPT Multimodal.
I heard the news through a rather breathless podcast about the drop and was trying to find my way out of the hype machine, but I think this is actually a big deal.
Previous AI tools were really just supercharged versions of existing tech options. ChatGPT was autocomplete on steroids. DALL-E was reverse image recognition. AR filters were Hipstamatic 2.0.
But this new GPT4 feature is something more.
It's a new way for users to interact with computing devices in ways more natural to the history of human experience and development.
Here's one of the examples from the announcement:
If I were to ask someone to help me face-to-face, I would point out the bike and ask (just like the demo). I wouldn't give them the specific bike make and model as a preface to my question.
When I'm trying to get a question answered, I'm looking for an answer. I'm not specifically looking for a list of links to click about or a YouTube video to scrub through or a PDF to scroll through.
This is the closest analog of human-to-human information seeking interaction I've seen from tech outside of limited voice assistant stuff.
We're moving closer to ambient computing—when the screen gets deemphasized and devices act more as extensions of your person than external boxes you tap at. AirPods, smart watches, and voice assistant devices are the forerunners of this trend. Multimodal is a power up.
If true AR glasses are feasible (big "if" on the technical hardware end of things right now), think about not even needing to take a picture for that bike example. Simply look at your bike and ask for help lowering the seat and get instructions overlaid on your field of vision.
We as marketers need to start thinking through how we can best serve customers through experiences like this in the future. And how our methods of distribution and promotion will change.
How Will Marketing Survive AI?
Same As We’ve Always Done
It’s not about the platforms or the tactics (or the snake oil).
It’s about meeting the customer:
In an effort to scale what customers love about your brand using the reach of digital.
Related
Kyle Mensing
Kyle MensingExclusive: ChatGPT traffic slips again for third month in a row
Exclusive: this is irrelevant
AI like ChatGPT is a feature or tool, not a consumer app.
The goal isn’t for ChatGPT to become the next Google Search, it’s to become the next Amazon Web Services.
The race for a generative AI-powered smart assistant is on.
Axios: Google Assistant to get an AI makeover
Axios: Amazon wants Alexa to bring AI into the home
It was bad timing for Microsoft that Cortana was sunset before the OpenAI partnership. But maybe The Windows Co wants to focus on enterprise and workplace solutions (where voice can be inconvenient).
This is another space where Apple could dominate had Siri not been left to languish. Siri + generative AI + market saturation + Airpods = conversational computing platform
🤖 What AI Thinks About AI
I was recently asked by a client to talk to their team about AI. What it is, what it means, etc etc. Naturally, I turned to ChatGPT and Bing Chat to write the presentation for me.
I plugged in their loose question guidance to generate an outline and then fed each part of that into one of the chatbots to see what they spit out. Any time I used Bing Chat I asked it to research the industry or something similar first.
So, what does AI think about AI?
What Is AI?
AI is all about making machines, like computers or robots, do things that normally require human intelligence. Humans are really good at things like recognizing faces, understanding language, and making decisions based on complex information. But with AI, we can teach machines to do these things too.
How Does AI Work?
AI is all about using algorithms, which are like sets of instructions, to help machines learn and make decisions on their own. These algorithms can analyze data, identify patterns, and even make predictions based on what they've learned.
There are different types of AI, like machine learning and deep learning, but they all have one thing in common: they use data to learn and improve over time. That means that the more data a machine has to work with, the smarter it can become.
AI is already being used in all sorts of ways, from helping doctors diagnose diseases to powering voice assistants like Siri and Alexa.
What Is Generative AI?
Generative AI is a type of artificial intelligence that is capable of creating new and original content, such as images, videos, music, or text, based on patterns and rules learned from existing data. Generative AI algorithms use complex mathematical models, such as neural networks, to generate new content that mimics the patterns and characteristics of the training data.
Of course, there are also concerns about the potential misuse of generative AI, such as the creation of deepfakes or other forms of misinformation.
What's a Challenge for AI Adoption?
One of the biggest challenges is data privacy and security. As AI systems collect and analyze large amounts of sensitive information, there is a risk that this information could be misused or stolen. Another challenge is that AI systems are only as good as the data they are trained on. If the data is biased or incomplete, then the AI system will produce biased or incomplete results.
How Will AI Impact Marketing?
Targeted Ads
By analyzing data on buyer behavior and creating personalized marketing campaigns that target specific demographics.
Past Sales Future Wins
AI algorithms can identify patterns and trends that help target marketing efforts more effectively.
Predicting The Future
Predictive analytics can enable digital marketers to make better decisions about how to allocate their resources and invest in new initiatives.
Increased Efficiency + Reduced Costs
Via task automation, predictive analytics, and early issue identification.
I mean, not bad really. Various queries got really repetitive in the "how will AI impact this industry / marketing," but nothing crazy. Of course, maybe this is all a feint by the chatbots so we don't suspect their true power and motives.
This was the point in the presentation where I revealed the joke and got on my soapbox.
I am of the group that believes...
Current AI = Infinite Interns
Our current models are like having an army of super-fast interns at your disposal, and you should approach them as such.
According to AI, the difference between it and an intern is:
It can perform tasks faster and more accurately than a human intern, especially for tasks that involve large amounts of data or require repetitive actions.
It doesn't get tired or bored, so it can work 24/7 without breaks or distractions, while an intern may need breaks and time off.
It can learn and improve over time as it processes more data, while an intern may have a fixed set of skills and knowledge.
It can handle complex tasks that may be too difficult or time-consuming for an intern, such as analyzing vast amounts of data or performing complex calculations.
The AI Formula
Pattern Recognition + Pattern Generation
Like most systems, you get out what you put in. In the case of generative AIs that means more of the same. A really good rendering of the internet culture version of the thing you asked for.
The large language models (LLMs) currently taking the world by storm (GPT, etc) are really just turbocharged predictive text models. If it sounds human, it's because it was trained on words written by humans.
The near term impact heralded by the chatbots is right:
Increased Efficiency + Reduced Costs
What could you do with infinite interns?
I'll leave off with this parting thought:
The Future is Centaurs
If AI gives you more of the same, how do you stay different?
If everyone uses the same AI, how do you do something different?
If every experience is AI, how do you stay human?
It’s not AI or humans…
It’s AI and humans
Meanwhile, in robots:
*Don’t worry, Google Search Chat will still have ads. I’m guessing this marks the end of keywords. Not now, but soon. We’ve been on a slow march to their inevitable death, but now we’ll actually start to see what the post-keyword future looks like.
My biggest question: will Performance Max campaigns be the only ones eligible to appear in AI-powered Google Search Experiences?
🎙️ Weekend Listens: AI, More AI, Community, Motivation Tricks, & Would A TikTok Ban Be 'Un-American'?
Azeem Azhar
Chalk this up as a victory for the centaurs (humans + AI).
Here's the paper's abstract:
We study the staggered introduction of a generative AI-based conversational assistant using data from 5,179 customer support agents. Access to the tool increases productivity, as measured by issues resolved per hour, by 14 percent on average, with the greatest impact on novice and low-skilled workers, and minimal impact on experienced and highly skilled workers. We provide suggestive evidence that the AI model disseminates the potentially tacit knowledge of more able workers and helps newer workers move down the experience curve. In addition, we show that AI assistance improves customer sentiment, reduces requests for managerial intervention, and improves employee retention.
Unlocking knowledge from top performers' heads and turning it into a resource for anyone at the company to use, sounds amazing.
Generative AIs aren’t an extension of how we currently interact with digital, they’re a reimagining. We have to change how we conceptualize activities like “search” and “content.”
When interacting with large language models (LLMs (think ChatGPT)), asking “please give me sources” is no different from asking “give me something that looks like things that have sources.” These models don’t have context or semantic understanding, they have incredibly powerful pattern matching turbocharged by historic levels of compute.
Is a TikTok ban a good idea? How would it even work (at least on a state level)? All good questions. But what about this question:
If the US suddenly starts banning foreign apps on data privacy and security grounds, will other (non-Chinese) countries do the same to American platforms and services?
Plenty of unintended consequences on the table here.
Don't Say Content
What if community wasn’t a strategy, what if a community was part of your business model? What if it was just part of doing business?
Community isn’t something you do at people, it’s something you build with them.
Just because something is old hat to us jaded marketers, doesn’t mean it isn’t new and surprising to other people.
It’s not transactional. The goal is to make people feel or understand that they belong. You belong here, this is for you.
How to motivate yourself or others:
This last point is useful for managers.
For new team members or people new to their roles, reviews and support can focus on how far they’ve come.
For experienced peeps, focus on how close they are to achieving their goals and the gap that remains.
Because incentives matter: