It’s Hard to Make AI Be Funny

“WHATEVER IT TAKES TO SURPRISE THE HUMAN”

Hi everyone, I hope you had a good weekend. If you live in SF or LA, it’s time to get those restaurant reservations you’ve been waiting for.

Checking in on the AI Labs

Let’s see what everyone’s been up to since the last newsletter.

OPENAI
Sam Altman has been on a small media tour following last week’s chaotic rollout of GPT-5. Some highlights: ChatGPT has over 700 million active users per week, Altman says the number of users who have “an unhealthy relationship” with Chat is under 1%, and the biggest bottleneck to rolling out new models and products is data center capacity. “You should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future,” he said. (I do think that a potential upside of the AI bubble is huge amounts of private investment in efficient energy production, especially since the government is trying to send us backwards in this regard.)

My favorite quote was this one, on the topic of AI upending the job market:

In 2035, that graduating college student, if they still go to college at all, could very well be leaving on a mission to explore the solar system on a spaceship in some completely new, exciting, super well-paid, super interesting job.

I would personally prefer that astronauts still go to college, but hey, it’s the thought that counts.

ANTHROPIC
Claude can now end chats on its own. Anthropic says this is “intended for use in rare, extreme cases of persistently harmful or abusive user interactions,” such as when a user won’t stop asking for “harmful” content. This brings up all sorts of fun questions, like whether this means you can cyberbully an AI.

I also enjoyed this piece from Ethan Ding about the divergence in strategy between Anthropic, who has been dominating the enterprise market ever since the release of Claude 3.5 Sonnet, and OpenAI, who is focused on a mass market consumer approach (for now). Somehow, someway, we’ll make sure that we don’t leave the ads behind when the new AI internet arrives.

xAI
The system prompts for Grok’s companions—personalities in the app you can chat with such as “unhinged comedian,” “therapist,” and “Grok doc”—were leaked online, and they feature gems such as “I want your answers to be fucking insane. BE FUCKING UNHINGED AND CRAZY. COME UP WITH INSANE IDEAS. GUYS JERKING OFF, OCCASIONALLY EVEN PUTTING THINGS IN YOUR ASS, WHATEVER IT TAKES TO SURPRISE THE HUMAN.”

It would take more than Grok making butt sex jokes to surprise this human after what the fine folks at xAI have been up to lately (you can now call the romantic companions if that’s your thing). But there is something darkly comedic about the desperation you can sense behind the prompts. Even as these models score higher on science, math, and programming benchmarks, it’s still pretty much impossible to get them to be funny. At least they’re turning down the slop on X? It seems like Nikita hasn’t seen Elon’s feed recently.

META
I’m good actually.

Pledging Allegiance to the Bag

On Friday, I mentioned rumors that the government was eyeing taking a stake in Intel, and yesterday it was reported that the administration has been discussing converting around $10 billion from Intel’s Chips Act grants into equity, which would translate to roughly 10% ownership in the company. And you did read that correctly, this is grant money that was already promised to Intel. The main difference would be that the company no longer needs to hit specific project-based milestones in order to unlock the funding, and also that, well, now the government owns 10% of it. This largely feels like a continuation of efforts that began under Biden to try and get the struggling chipmaker to remain competitive with foreign rivals like TSMC, which will only become more critical if the AI race continues and things like supply chain security become primary governmental concerns. (Japan’s SoftBank also announced a $2 billion investment in Intel.)

Execs Don’t Know What AI Is For

The Times has a fun read detailing how many executives, who can’t stop exhorting their employees to use AI to be more productive, often have no idea what to do with the technology themselves. A frequent excuse is that the job of an executive doesn’t lend itself well to the type of work AI tools are good at, but that doesn’t really check out given that the AIs are pretty great at summarizing meeting notes, giving feedback on written content or emails, and helping analyze data, which are all things executives supposedly do. The real answers? One, executives are old, so they adopt technology more slowly. Two, no one is completely sure what to do yet because these tools are still new and not very well integrated into most people’s jobs (software engineers excluded).

I think there’s an unfortunate feedback loop at many larger companies right now where employees are expected to use AI tools to increase productivity, but the tools tend to be half-baked or just ill-fitting (because they weren’t chosen by people actually doing the job), so they aren’t boosting productivity as much as management wants. Then when asked, employees are incentivized to exaggerate the productivity gains so that they don’t appear to be underperforming and get laid off. This creates a bizarre self-reinforcing illusion that everyone is turning into a super employee, even if the reality is just that we’ve made everyone feel busier, which is a thing that we love to do with technology (looking at you, Slack).

And if all else fails, tell your bosses they can just follow in the footsteps of the creators making a living by flooding TikTok with AI video slop.

Doomscrolls

• China hosted a robot olympics over the weekend. It went how you’d expect.

• They’re making nicotine energy drinks now. Sure, whatever.

• A new contender for worst rebrand of all time has entered the arena.

• Read Max investigated why Elera Voss is taking over Amazon.

• Welcome to The Chili’s Economy. My friend has a theory that all Americans really want from a nice restaurant is dressed up TGI Friday’s food—think about it next time you pay $50 for chicken wings and a burger.

• Chamath “I pick Metromile” Palihapitiya, a man famous for tricking investors into losing lots of money during the pandemic, is doing it again. At this point, if you believe anything he says, I don’t know what to tell you man.

One Last Tweet

if you have moved on, go see Weapons

All em dashes added organically by me. All jokes inspired by my AI comedian sidekick’s INSANE IDEAS.

Thanks for reading. If you liked this post, please share it with a friend.