AI round-up: Week of April 21, 2025

We were off last week. Fortunately, nothing happened!*

I wanted to start this week with a new party game/ice breaker we can all play.

It’s from Ethan Mollick and spotlights the One Word Turing Test. Here’s how it goes:

A person and an android are standing in front of a judge. Each has to say one word to prove they’re human. The judge shoots whoever he thinks is the android.

What word do you say?

In the survey Mollick cites, most humans answer ‘love’.

Wonder what the leading LLMs would answer?

Question for you: what word are you saying to the judge?

*This is a total lie. Lots happened.

Let’s get to it.

 

The Heavy Stuff

Story 1: OpenAI releases o3 and o4-mini (Shelly Palmer)

Not sure if you’ve tried it yet but the depth of thinking (and the likelihood of hallucinating) has increased.

The biggest advantage I’ve seen is it decided when to search the web, do research, etc. Oh, and it’s also multi-modal so it can review images you drop in as part of its assessment.

But the biggest ‘yikes’ for me?

Well, take a look at this prompt:

I'm working on a training program that will help me prepare to run a marathon in six months. Can you provide me with a list of questions you would need answered as well as a generic best practices approach around what my next six months look like? Please include both diet and physical fitness routines I should expect to follow.

I received a very in-depth overview, outlining all of the steps and expectations I should be prepared to take in the next six months. But look closely at the start:

2025-04-25 14_54_00-Project Llama Overlay

Yeah. It did this in 46 seconds. And felt the need to rub it in to me by telling me it only needed 46 seconds.

Ok, I’m kidding about the rubbing it in part. BUT…it’s providing complex planning and programming in seconds. SECONDS. Not minutes.

Here’s Ethan Mollick’s take on these models as well as the other new models appearing on the scene these last two weeks.

Story 2: Speaking of new models, Google’s Veo 2 video generator is here.

Only Premium users can access, but whether you have it or not…check out what it can do.

Story 3: Cautionary tale – Google AI shift leaves website makers feeling betrayed.

We’re starting to see the impact AI is having on web traffic as many publishers are reporting a 70% decline in traffic. (Bloomberg)

Ready to start talking about this yet?

Story 4: Anthropic is talking about AI employees being a year away. And what that means for you.

The real headline of that article is ‘Anthropic warns AI employees are a year away’ and in a rare move by Axios, is a bait and switch a bit. The article is more about the cyber steps you need to be taking and what could happen if an AI employee goes rogue.

Regardless, worth the read. Oh, and…AI agents can go rogue?!?

Story 5: An AI agent went rogue. And it’s a warning for any company looking to replace workers with AI. (Fortune)

Story 6: Whether they go rogue or not, should AI have rights if they become conscious? (The New York Times)

Story 7: OpenAI slashes AI safety training time. (Financial Times)

No biggie. It’s only the fastest growing app in the world with roughly 10% of the population using it now. (To note, roughly 70% of the world has a cell phone…which may not seem like it matters but one of those two things has only been here for 2.5 years.)

The not-so-heavy stuff

Story 1: The Oscar goes to…AI movies. (Shelly Palmer)

But is an AI movie better than Emelia Perez?

Story 2: Saying ‘please’ and ‘thank you’ is costing OpenAI “tens of millions” of dollars. (USA Today)

This is according to Sam Altman, who views being polite to the machines as a necessary expense, just in case.

Picture2

Story 3: Perplexity AI enters the smartphone market with Motorola deal. (CNBC)

A few that don’t fit in either category

Story 1: AI helped write bar exam questions, California state bar admits.

As you read this article from The Globe, ask yourself what the real issue is…that they used AI? Or that the bar doesn’t have a policy or program in place to appropriately use AI? Regardless of how you feel, the answer…is the latter. (Whether we like it or not.)

Story 2: OpenAI would be interested in buying Chrome if Google is forced to spin it off. (Bloomberg)

I don’t know what this means, and there’s so much other noise right now that I haven’t thought much about it. What do YOU think?

Story 3: Some important names – including the Godfather of AI, Geoffrey Hinton – don’t want OpenAI to become a for-profit company. (Financial Times)

It’s just a sad acknowledgement that OpenAI is so far from its original mission. The bottom line: should a company, driven by profits, be responsible for advancing the technology to the point of AGI instead of keeping humanity first?

Final note

I am working on potential topics/thought starters for the upcoming AKHIA AI Roundtable. I reviewed some questions in ChatGPT along with some recent industry reports to see what their list looked like.

No matter how many times I refined it, the topic of AI’s impact on the media kept coming up.

I really didn’t want to tap into that conversation because I was hoping to steer clear of things like media influence, potential bias and concerns tied to political parties (one way or another).

But given the fact ChatGPT kept putting it out there, combined with the news that The Washington Post is the latest media outlet to partner with OpenAI, I feel like now…I have to.

I guess AI knows something we don’t when it comes to paying attention to a.) who is partnering with who and b.) how that might skew the information we receive from AI when interacting with a specific LLM.

The lesson, as always – don’t believe everything you see or read. Question where it comes from and who’s tied to/training who…

Happy Friday!

-Ben

As a reminder, this is a round-up of the biggest stories, often hitting multiple newsletters I receive/review. The sources are many … which I’m happy to read on your behalf. Let me know if there’s one you’d like me to track or have questions about a topic you’re not seeing here.