😺 OpenAI's secret math scam

PLUS: Super-agents got us shook + how to properly prompt o1...

Welcome, humans.

Here’s something silly to start your Monday… a thread of people using AI to turn their screen-names into images. Moist Leggings had us ROFL…

The Ritzler options were also very good.

Now for some housekeeping: If you want to reach 500K engaged decision-makers actively implementing AI in their companies, fill out this form.

We're talking VPs, Directors, and tech leaders at both Fortune 500s and SMBs, across operations, sales, and engineering—all of whom read The Neuron (just like you!).

We have only a few sponsorship spots left in Q1... so click the button below and fill out the form (it takes less than 1 minute) to book your spot while you still can.

Advertise in The Neuron here.

Here’s what you need to know about AI today:

OpenAI promoted o3-mini and super-agents amid “Benchmarkgate.”
OpenAI and Microsoft made AI science breakthroughs.
Perplexity made bid to merge with TikTok U.S.
Study says Millennials most at risk for AI job displacement.

OpenAI hype season is in full swing with secret fundings, closed-government meetings, and “super-agents”…

It's only January and the OpenAI hype machine is spinning on all cylinders. Last Friday, Sam Altman announced the new o3-mini is shipping in “a couple weeks.” He says “it’s very good” and also “worse than o1 Pro at most things (but FAST).”

At the same time, he's heading to DC for a closed-door meeting with government officials on Jan 30th about supposed “Ph.D.-level super-agents” OpenAI staff are reportedly telling friends they're “jazzed and spooked” about what comes next.

That’s on top of the ominous posts from last week. To believe, or not to believe…

Here’s what we know for sure (via Axios):

These super-agents are designed to do complex human tasks.
They “pursue goals” so you can delegate entire projects to an AI.
They can “synthesize massive amounts of information, analyze options and deliver products.”

Axios says these super-agents could theoretically build entire software products from scratch, analyze investment opportunities across thousands of sources, or plan complex events down to the tiny details.

Meanwhile, some of OpenAI's top researchers are trying to temper expectations. Noam Brown just tweeted a warning about “vague AI hype” and even felt the need to clarify, “we have not yet achieved superintelligence” (Oh, good!).

Then there’s the drama happening behind the scenes… Remember when OpenAI announced o3 crushed a new high-level math benchmark? Well... turns out they were secretly funding EpochAI, the org who runs that benchmark.

And it gets more interesting: EpochAI's associate director admitted the lab was contractually “restricted” from disclosing OpenAI's involvement until the o3 launch.

Think about that timing for a second. OpenAI funds a benchmark in secret, gets access to all the problems, announces amazing results, and only then allows the creators to disclose the connection.

Here’s why “Benchmarkgate” matters: In an era where Apple and Samsung are getting dragged for overhyped AI features, OpenAI has to play a different game entirely. With no phones or computers to sell, it has to sell something bigger: a vision of the future of intelligence itself, with every revelation calibrated between “don't worry, we don't have superintelligence yet” and “but we're this close.”

We know SamA and OpenAI aren't just blowing smoke. But pushing a narrative about super-agents and superintelligence is probably an easier way to make back Microsoft's billions than selling chat completions… just saying.

FROM OUR PARTNERS

Build AI that works offline!

Join LatentAI CEO Jags Kandasamy and Dell's Logan Lawler on Feb 6th to learn how Dell Precision workstations powered by NVIDIA RTX™ let you create lightning-fast AI applications that run directly on devices—no internet required.

Prompt Tip of the Day

Great framework for how to prompt o1 from this amazing piece by Ben Hylak in the Latent space newsletter

Ben Hylak says you need to stop treating o1 like ChatGPT. In a new piece in Latent Space, he reveals o1 works best when you flood it with context and treat it like a report generator instead of a chatbot.

His key insight: Record a quick voice memo explaining your full problem, transcribe it, and dump it in. The more context, the better the output. Plus, focus on telling o1 WHAT you want, not HOW you want it.

Go read his full breakdown here for more game-changing tips that'll help justify that $200 monthly subscription. There’s also a podcast discussion on all this if you prefer.

Treats To Try.

*Your sensitive data is being traded by data brokers daily. Incogni scrubs it from the web. Get 55% off with code NEURON here.
Raycast Focus helps you block distracting apps and websites while you work; you can set specific goals, choose custom durations, and easily snooze blocks when needed—great launch video.
AIVLOG selects and edits the best moments from your raw footage into polished vlogs (Apple only rn).
Humva turns your photos and scripts into a lifelike avatar (20 free credits when you sign up).
Typelink lets you build a website by turning your social links into clickable portfolios, landing pages, and bio links (free to try).
Recap turns any webpage (or Youtube video w/ the extension) into visual mind maps that help you remember key points.
Eververse lets you collaborate w/ your team to organize and plan your product's future features and timelines—completely free to try + open-source.
Left visualizes your “time remaining” as dots on your home screen, making every day / week/ month / year left mean more as it counts down (Apple only rn).

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

OpenAI developed a protein engineering model that's 50x more effective at turning regular cells into stem cells, its first foray into scientific discovery (“level 4” of AGI), while Microsoft released MatterGen, an AI that directly generates novel materials w/ specific properties.
Perplexity submitted a merger bid to ByteDance to combine with TikTok U.S., as Trump promised to give TikTok 90 more days to work out a deal.
Microsoft launched semantic search for Windows 11 on Copilot+ PCs, letting users find files and settings by typing “bridge at sunset” even when offline.
A new study found Millennials (38%) face the highest risk of AI job displacement, followed by Gen Z (25%), Gen X (20%), and Baby Boomers (10%), with roles rather than age being the primary “displacement factor.”
Despite 48% of legal pros ranking “overwhelming work volume” as their biggest challenge, only 32% of law firms plan to use AI to help—meanwhile, “Big Law” is already using AI for everything from deposition summaries to associate training.
Today is U.S. President Trump’s inauguration day…here’s a reminder of what we said his Presidency could mean for AI going forward.

FROM OUR PARTNERS

PromptQL: RAG with 100% Accuracy.

While traditional RAG delivers only 40-70% accuracy, PromptQL's agentic AI planning achieves perfect scores through dynamic planning, limitless data processing, and full decision transparency.

Don't take our word for it—try it out today!