Welcome, humans.

ByteDance’s AI game is strong—especially its TikTok content algorithm. Arguably, it’s the company’s money-maker.

Well, over the weekend, an intern “maliciously interfered” in one of the company’s other AI projects, reportedly causing millions in damages.

At first, we thought maybe the intern slipped some spicy content into the dataset as a prank. But this wasn't your typical intern mixup, like accidentally CC’ing the entire company your Chipotle order (true story!).

According to internal reports (original post here), the intern exploited vulnerabilities to deliberately sabotage training runs by inserting delays, killing processes, and even reversing training progress—and then, they had the audacity to sit in on troubleshooting meetings so they could adjust their attacks to avoid detection.

Why? All because GPU resources were diverted from their research. Somewhere, back at HQ, an NVIDIA salesperson is taking notes…

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

• We break down the bear case against AI.
• Meta announced facial recognition to stop “celeb bait.”
• Microsoft launched 10 new agents ahead of Salesforce.
• Teachers’ widespread use of AI detectors hurt vulnerable students.

Advertise in The Neuron here.

Here’s the “Bear Case” against the AI Boom…or should we say “Bubble Case”?

Two finance wise-guys, Jeffry Funk and Gary Smith, just dropped an op-ed claiming the AI bubble will be much bigger than the dot-com bust.

This was an op-ed in Dow Jones & Co, so this was directed at investors. The message? Brace for impact.

Here’s their argument:

1. AI bulls compare rapid AI adoption to slower internet uptake as proof against a bubble, but that's flawed:
   1. In 1999, a computer cost $5K+ and internet was $113/month in 2024 dollars. Big costs = big payback hopes.
   2. Today, GenAI is mostly free to try (thanks to big tech + VCs subsidizing it). No cost = no friction = no stakes.
2. AI’s creation costs (>$100M for training + loads of water and energy) outweigh the benefits.
   1. OpenAI expects losses of $14B in 2026, $44B in 2028, and won’t profit til 2029…when its revenue hits $100B.
   2. OpenAI’s $157B valuation (~42x earnings on $3.7B revenue) tops 87% of S&P 500 companies.
3. AI needs $600B annual revenue to justify investments; current revenue = ~1% of that.
   1. Internet revenue in 2000 was $1.5T (2024 dollars)—and it still burst.
   2. AI revenue today < $10B, suggesting a way bigger pop.

Source: MorningStar

Why this matters: Funk just wrote a book about exploiting tech investment bubbles, and Smith is an expert at “debunking” data misuse. Smith argues that AI’s danger isn’t computers outsmarting us, but us trusting them too much (because we think they’re smarter).

Our take: It’s definitely a bubble, but even the dotcom bust saw winners. Right now, GenAI providers are targeting enterprise use-cases pretty heavily, which indicates a new stage in the industry’s growth. This shift from consumer to enterprise follows previous tech waves like cloud storage, social media, and most recently, blockchain.

FROM OUR PARTNERS

DIVACS is the secret to turning skeptical prospects into eager buyers.

Do you know the true value of your business? Like, have you ever sat down and actually asked your customers why they choose you over the competition?

With DIVACS, you don’t have to.

DIVACS is like your AI wingman for B2B sales success—it helps you quantify and communicate the REAL benefits your company provides, whether thats:

• Cost savings.
• Efficiency gains.
• Revenue increases.

DIVACS then helps you showcase this value to any prospect, customer, or vendor.

In short: You can close deals faster by proving your real financial impact to your buyers.

Request a demo and try it out yourself here.

Around the Horn.

Just for fun 😛

• Meta will use facial recognition to combat celebrity scam ads (known as “celeb-bait”) by running face scans over ads that get flagged as suspicious.
• Newscorp’s Dow Jones and NY Post sued Perplexity for orchestrating what they call a “content kleptocracy” and want up to $150K per infringement.
• Microsoft announced 10 new agents for Dynamics 365 (launch video), including bots for sales qualification, supplier communication, and customer service—some think this was a strategic dig at Salesfroce Agentforce, which launches this Friday.
• Over two-thirds of teachers report using AI content detectors, with testing showing even highly accurate systems had a 1-2% false positive rate—this particularly impacts neurodivergent + ESL students, who write in more formulaic styles (as we explained here).

Treats To Try.

1. *Metronome powers billing and pricing for the top names in AI. Define and iterate on your usage-based pricing seamlessly. Request a demo (or check out this OpenAI case study).
2. IBM’s Granite 3.0 models (with 8B and 2B parameter versions) are now available for commercial use on the Watsonx platform—code here, read more here, or try Granite in the playground here.
3. Llama Workspace is a new tool that gives you access to most major chat models (GPT-4o, Claude, Llama) in one team workspace for a fraction of ChatGPT's cost ($9/user/month)—there’s also LibreChat, which is another tool you can try as well.
4. LLamaLife helps you complete your daily tasks one at a time by showing estimated finishing times and task timers (to combat time blindness).
5. Elipses News creates personalized daily podcasts on the topics that matter to you—use promo code ElLIPSISOCT for 1 month free (app store only).
6. Talkstack gives you a fleet of agents for calls, texts, and emails—new launch, so read more here. You can mess around for free, but gotta buy credits to test.
7. Duonut drops surveys into your apps, asks follow-up questions as users respond, and gives you quick summaries—free to try, pretty neat.

See our top 51 AI Tools for Business here!

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

Tuesday Ticker

• J.M. chose Web searches: “I've pretty much stopped using Google for searching and fact checking. I also use Perplexity as very factual AI for document checking for style, spelling, tone, etc. I find it's another valuable angle to add to my tool box.”
• T.J. chose Web searches: “Perplexity was helpful this year for getting up to speed on and comparing local political candidates up for election in November.”
• E.J. chose Haven’t tried it yet: “It came highly recommended by someone that also pointed me to your newsletter a few weeks ago. Great article! Thanks!”

Check out the full story here if you missed it.

A Cat's Commentary.