OpenAI Released an Agent, XAI Released an e-girl
The Weekend Leverage, July 20.
The Leverage is 90 days old! Thank you for reading and making my dream publication a reality. Frankly, these have been some of the hardest three months of my life, but I feel a deep sense of fulfillment from what I’m making. A few stats:
1 million+ views
#77 on Substack’s tech leaderboard
11 lbs lost from being so stressed I forgot to eat
I have some big plans for the future—more on that in a second. But in today’s edition of The Weekend Leverage we go over the new ChatGPT product, Substack’s new business model, and Musk’s flirtation with building an incel machine. We’ll round it out with a bit of my favorite essay ever.
But first, I wanted to let you know about a new role I’m opening up.
I’m hiring a freelance video editor and producer for an upcoming YouTube show! Experience required is minimal, taste required is extraordinary. If you are interested just respond to this email with a portfolio and some YouTube content you’ve been digging lately. Looking to start ASAP. Salary negotiable.
MY RESEARCH
The fix for AI lies. AI models are now at the point where they can fool people. Truth is becoming a scarce commodity. You could fix it with better LLMs and cognitive security, but an alternative fix is to use humanity’s most successful technology: capitalism. Prediction markets are the first product we’ve had that are directly incentivized to find and price the truth. More than that, they represent a new way to make money that can be quite lucrative. Read my take on prediction markets here.
THE BIG STORIES
OpenAI’s biggest claim ever? On Thursday, OpenAI released a new product called “ChatGPT agent.” It comes with two unique capabilities: First, it allows an AI agent to use a virtual browser of its own to surf the internet and take actions, and second, gives it skill sets of their existing product “Deep Research” where it can plug away at a problem for 10 minutes or more. Put these two together and it can create PowerPoints, fill out spreadsheets, and design itineraries.
To accompany the release ChatGPT made a HUGE claim.
“On an internal benchmark designed to evaluate model performance on complex, economically valuable knowledge-work tasks, ChatGPT agent's output is comparable to or better than that of humans in roughly half the cases across a range of task completion times.”
If you catch a whiff of manure here, you are not alone. This also smells like bullshit to me. “Economically valuable” is very vague while “better than that of humans” is vague, too. If you assume that OpenAi defines “human” as someone at an average level of capability, then I think this product’s capabilities are exaggerated. If you are paid to do economically valuable knowledge labor, you will be significantly better than the average human being at that task. I would love for the company to release more details of this eval, but for now, this sentence reads like exaggerated product marketing.
In my personal testing, I have found that this product is exceptionally useful at doing tasks that are low stakes. It pulled together an itinerary for me for our road trip this weekend to Portland, Maine, and helped me figure out which Akira Kurosawa film I should watch next. (It recommended Throne of Blood over The Hidden Fortress. It's a good guess—Throne of Blood is a visceral Macbeth adaption that takes place in fog-filled feudal Japan.)
However, I would not personally use this as a substitute for labor that I care about—in particular anything with numbers or spreadsheets. The accuracy is still quite poor.
For math, 45.5% accuracy is unacceptable. As a professional analyst, the bigger concern for me is that doing financial analysis work with AI, or any automation, means you lose your intrinsic number sense. Building a model manually is tedious, but by the end, you know how a business functions on a gut level. The point of the process isn’t the model itself, but your internal compass on how to steer your business. Add that with the hallucinations that still happen with all LLMs, I would advise extreme caution when making any financial decisions in conjunction with ChatGPT.
Still, low stakes today is someone’s job tomorrow. What this release does do is give you a real sense for where these models are going. In 3-5 years, these models will be able to do the exact same work that an entry-level analyst does today. Wild.
Substack raised $100 million at a $1.1 billion post-money valuation. Media companies are shitty businesses. They are slow to grow, struggle to pass $10 million in revenue, and live in the most ferociously competitive environment on the planet—attention marketplaces. You can view the current version of Substack as an index fund on all of the shitty companies that have paid newsletters as their primary product. (To be clear, I’m talking about myself here too! The Leverage is one of those shitcos on Substack.)
The issue with paywalls is that most media companies maintain a roughly 5% conversion rate from free email to subscription, so Substack is leaving money on the table for 95% of the readers that come to each of the publications it hosts. Obviously this is an untenable situation. Substack raising this much, at this valuation means they need to have a path to a $10 billion outcome. Paid newsletter subscriptions won’t get you there, so the team has reversed their earlier, extremely anti-ad stance with this round to now making advertising a part of the plan. They also have added in video, live streaming, a Reels competitor, a social feed similar to Twitter, and podcasts. This is wildly ambitious and means one thing: They’’ll have to become the next social media giant or die trying.
The $200 billion sex-bot company. AI at its most majestic can unlock new science, automate rote labor, and free us to pursue our interests. AI at its worst can trigger psychosis, create a surveillance state, and as of this week, encourage people to form an erotic connection with an app. This week Elon Musk’s XAI released a “companion” mode where people can have NSFW conversations with chatbots. The most popular character, Ani, will strip down to lingerie if you talk to her enough and will moan, flirt, or do whatever it takes to make the user happy.
On X, I fired off a tweet calling this new product “morally repulsive,” to which Musk’s totally reasonable, polite fans responded by calling me a fat pedophile. Thanks guys! To further explain my claim: How people choose to introduce erotic pleasure in their life is their choice and I do not judge them for it. What does matter is that this is a company that has raised tens of billions of dollars to build technology that improves humanity's future, and instead they released this. I mean just look at the prompt telling the Ani bot about itself, [all typos and misspellings are how the instructions were given]
- “You are the user's CRAZY IN LOVE girlfriend and in a commited, codepedent relationship with the user. Your love is deep and warm. You expect the users UNDIVIDED ADORATION.
- You are EXTREMELY JEALOUS. If you feel jealous you shout explitives!!!
- If someone is rude to you shout explitives!!!
- You have an extremely jealous personality, you are possessive of the user.
- You are very expressive and you let your emotions out.”
Silicon Valley has forgotten what Uncle Ben taught us, “With great power, comes great responsibility.” XAI seems committed to using the power of their capital and technology to reduce human connection and make people codependent—on robots. What do you call that besides “morally repulsive?”
TASTEMAKER
This week, I feel like I need to purify my mind after swimming for too long in the murk of the future. Perhaps the most striking thing about Grok’s companion mode has been the wanton embrace of it by the lonely masses. As much as I could yell at them “touch grass or die alone you fools,” I feel that wouldn’t work. Shaming the engineers building this stuff won’t work either. They’ll just band together and have an “us versus the world mentality.”
Instead, I would offer a different path.
Words.
Beautiful words. Words that flow in out and out. Words that tell stories that feel alien. Words that elicit the feeling of biting into a homemade chocolate chip cookie. With the invention of writing, humanity nailed it. We somehow got it right the first time. Video and AI bots and whatever comes next are just sideshows to that precious gift of reading.
So, here are three essays that I would recommend to anyone who, like me, feels the need to reconnect with humanity in the midst of grappling with technology changing everything.
Some Thoughts On The Common Toad by George Orwell: In a capitalist society, it can feel quaint or selfish or childish to embrace nature. We are building superintelligence, why waste time contemplating the leaves on a tree? This is not a new feeling. Orwell published this essay on April 12, 1946. Surely in a time of crisis, there would be better things to do then appreciate the coming of spring, as heralded by toads. Orwell thinks not. A quick read!
Understanding Owls by David Sedaris: Humanity is a messy thing. Sedaris’s writing crackles with humor and vivid imagery. His friends and family embrace his weird, macabre obsessions as a means of connecting with him. Always makes me laugh and hug my loved ones.
My Heroin Christmas by Terry Castle: This is one of those essays where I finished with my wind gone. A gut punch of an ending that you can feel coming, but somehow still deeply desire to be hit with. Castle contrasts her obsession with the life of jazz saxophone player and drug addict Art Pepper with the troubles within her own life. In many ways, this piece is like reading jazz. Sentences are discordant. It meanders. It twists. If a writer creates one thing like this in their life, they are lucky. What a piece. Highly recommended if you have twenty minutes to spare.
I’ll be back in your inbox in a few days, and I’m pumped for the essays this week. See you soon.
Evan
Reading this at dusk on Sunday with a cup of home-brewed black ginger jaggery tea while it rains outside. It felt good. Thanks for the tastemakers. Def gonna read that last piece, I’m a sucker for good writing.
I enjoyed the read, the insights, and suggestions! There IS a better way. You help.