Two Steps Forward, No Steps Back. Ever.

[00:00:01] It is Tuesday, May 14th. Today, battle of the AI. Both Google and OpenAI come out swinging

[00:00:08] with their latest technology. What does it all mean for marketers? We have full coverage

[00:00:13] in today's special episode.

[00:00:17] What a difference 24 hours makes. In the last day, both OpenAI and Google revealed

[00:00:22] their plans for their respective AI technologies. And it's either exciting or terrifying,

[00:00:28] depending on which side of the debate you set. We're devoting our whole episode

[00:00:31] to those two announcements today. OpenAI was first, so let's start there. First, despite

[00:00:36] a ton of speculation, which included double-sourced reports from Reuters, OpenAI did not reveal

[00:00:42] a search engine. That's obviously good news for Google, which seemed a little trepidatious.

[00:00:48] Just 45 minutes before OpenAI's big announcement yesterday, Google tried to steal a little

[00:00:52] thunder by posting a short video tease of their upcoming tech. So no search engine.

[00:00:57] But OpenAI did show off three things. An upgraded model, an upgraded mobile app,

[00:01:03] and a new desktop app. First, GPT-4 will be getting an upgrade to GPT-4-O. That's

[00:01:09] the letter O, apparently short for Omni Model. OpenAI says it's smarter, it can

[00:01:14] detect things in images better, it can listen to things. But most importantly

[00:01:18] for marketers, I think, is it'll be free. GPT-4 right now costs 20 bucks a month

[00:01:22] in the US, and it's solid. It's definitely better than three. I spent a couple of hours

[00:01:27] today testing this new 4.0 model, and I'm underwhelmed. It still gets confused about

[00:01:33] word count. It starts to ignore commands and lengthy prompts. I wasn't able to detect

[00:01:38] much of an upgrade for some common copywriting-related marketing tasks. I uploaded a photo of a

[00:01:43] woman in a green dress and asked it for a product description. It wrote,

[00:01:47] Turn heads in this vibrant green button-up mini dress, the playful ruffled hem adds a

[00:01:52] flirty touch while the short sleeves and belted waist create a flattering silhouette.

[00:01:56] So basically the same level of intelligence as its GPT-4 predecessor. One area of marketing

[00:02:02] where it did seem to get smarter was in analysis. I uploaded a CSV of every response

[00:02:08] we've received from our newsletter survey. It's about 900 rows with data about demographics,

[00:02:13] job titles, and so on. I removed the email addresses and other identifying information

[00:02:17] beforehand. It actually did a pretty good job of summarizing the data from, remember,

[00:02:22] just rows and cells in a CSV. I told it to visualize the data and I got a bunch of charts.

[00:02:27] This is something the GPT-4 model can do. But then I said to make the design nicer,

[00:02:31] and it did. It replaced the colors with more muted tones and added color

[00:02:36] differences between the bars. Then I gave it what I thought would be impossible.

[00:02:40] Turn that into a PowerPoint presentation with one chart on each slide. Make the slide

[00:02:45] background blue, use a sans-serif font and make sure each chart's background is transparent.

[00:02:50] And in about 10 seconds it did exactly that in the form of a downloadable PowerPoint file.

[00:02:57] So if your work involves data analysis and what marketing job does in these days,

[00:03:01] this new model is pretty solid in that department.

[00:03:04] Most of the announcement time was dedicated to showing off its new mobile app which comes

[00:03:08] with a disturbingly realistic voice, one that replies without a delay and even has a personality.

[00:03:16] Maybe too much personality. Here's one of the demos. And it's hard to think that this is not

[00:03:21] a script that they've put into their bot. A slightly disheveled software engineer is holding

[00:03:25] his phone up and the app is showing him on the front camera. The bot walks right into

[00:03:31] uncanny valley.

[00:04:01] I don't have a lot of time so I'm just gonna throw this on. What do you think?

[00:04:19] He puts on kind of a ridiculous looking fisherman's hat.

[00:04:24] Rocky that's quite a statement piece. I mean you'll definitely stand out,

[00:04:31] though maybe not in the way you're hoping for an interview.

[00:04:34] Okay I got it, I got it.

[00:04:36] You can interrupt the bot mid-flow and it'll respond immediately.

[00:04:38] I think you can see the obvious application here for customer service phone lines.

[00:04:43] Sure this one was programmed to be informal, maybe even a little flirty,

[00:04:47] but that's the prompt they gave it. We'll be able to change that.

[00:04:50] Hell we'll probably just upload our brand guidelines and be done with it.

[00:04:54] Then there was the desktop app on both Mac and Windows. There are a ton of third

[00:04:58] party apps out there that use OpenAI's API to get you that functionality,

[00:05:02] but they almost all cost money. I used the GPT plugin on Raycast.

[00:05:06] That's $8 a month for 3.5, GPT 3.5 and it's another $8 more if you want GPT 4.

[00:05:13] OpenAI did not say what the pricing would be for the 4.0 API, but any pricing would mean

[00:05:19] that these app developers would have to charge, probably pushing a lot of users to download

[00:05:23] the free official desktop app. That official app does exist right now,

[00:05:27] though it's a test build. It's a little hard to find, but when you log in,

[00:05:31] it just tells you it's not ready yet.

[00:05:34] For its part, Google had a big IO developer conference today and here's what's interesting

[00:05:39] from that. First of course, the model itself is getting an upgrade that will let it analyze

[00:05:44] longer documents, code bases and longer videos and audio clips than before.

[00:05:48] The company's main AI, which it calls Gemini, will be able to search,

[00:05:52] summarize and draft Gmail's using the tech. It'll also be able to handle multi-step tasks

[00:05:58] like looking for an e-commerce receipt in your archives, finding the receipt,

[00:06:02] then filling out an online form. They also previewed their version of a talking chatbot

[00:06:07] called Gemini Live. It's not quite as quick to respond as OpenAI's bot and it sounds a

[00:06:13] little more robotic than its eerily human sounding competitor. Like that one though,

[00:06:18] you can interrupt it, ask clarifying questions and when used on a phone,

[00:06:22] it can also see and respond to images you show it. Quoting TechCrunch,

[00:06:26] quote, at first glance, Live doesn't seem like a drastic upgrade over existing tech,

[00:06:31] but Google claims it taps newer technologies from the generative AI field to deliver superior,

[00:06:37] less error-prone image analysis and combines these techniques with an enhanced speech engine

[00:06:42] for more consistent, emotionally expressive and realistic multi-turn dialogue, unquote.

[00:06:48] Google will also be putting a small version of the model directly into its Chrome browser.

[00:06:53] The company says this will let developers power their own simple AI features like a

[00:06:57] help me write tool and so on. More interestingly from a marketing perspective

[00:07:02] is they'll be letting users search using a video they upload. Users will then ask a

[00:07:06] question about the video or search they want. It's similar to the image search feature they

[00:07:11] have, but this kind of media search has already changed the game in SEO. How much longer before

[00:07:16] we have to consider exploiting this new tech to try to get up first in the AI's response

[00:07:21] messages? Again, TechCrunch, quote, first introduced in 2021, the ability to search

[00:07:26] using both photos and text combined has helped Google's in areas that it typically struggles

[00:07:31] with. Like when there's a visual component to what you're looking for that's hard to

[00:07:35] describe or something that can be described in different ways. For example, you could pull up

[00:07:39] a photo of a shirt you liked on Google search and then use Google Lens to find the same pattern

[00:07:45] on a pair of socks. Now with the added ability to search via video, the company's reacting

[00:07:50] to how users, particularly young users, engage with the world through their smartphones.

[00:07:55] They often take videos, not photos, and express themselves creatively using video as

[00:08:00] well. It makes sense then that they'd also want to use video to search at times, unquote.

[00:08:09] So all told, there was a lot of hype, a lot of buzz around AI in the last day. I hope that gives

[00:08:13] you a bit of a summary tomorrow back with some other stories in the industry, including

[00:08:17] a lawsuit facing Reddit over disputed clicks and Instagram's big warning about the reels

[00:08:23] your brand is uploading. Make them too long and you'll kneecap your reach.

[00:08:29] I'm Todd Maffin. See you tomorrow.