😺 Censor AI?

PLUS: Well, almost all of them....

Welcome, humans.

For many, the term "AI" has morphed into a blanket term for virtually any piece of modern tech. Just recently, a buddy of mine was berating his smart toaster for burning his toast, attributing it to "AI mischief."

Firstly, smart toasters aren't AI, and secondly, perhaps avoid getting lost in TikTok while your multigrain is on the brink of combustion…

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

  • Meta is prepping Llama 3 to answer more controversial questions.

  • Stack Overflow is charging Google to access coding answers.

  • Cybercriminals are exploiting AI to send phishing attacks.

  • Microsoft announced an AI assistant to aid with financial data.

Companies decide how much to censor AI.

What types of questions should a chatbot answer?

  • Inquiries on poisoning methods? Skip.

  • How to build a nuclear bomb? Definitely skip.

  • Pictures of racially diverse Nazis? Just ask Google (jk).

It’s an important question that AI companies have faced for 15 months, often landing them in hot water.

Initially, the strategy was to dodge any topic that might spark a PR nightmare. The following is satire, but you get the point:


Heavy “censorship” pissed people off (looking at you, Google), so things swung in the opposite direction.

The Information reported that Meta researchers are training the next version of its AI (Llama 3) to answer more controversial questions than the “safe” Llama 2.

Here’s our take: This is a similar debacle as social media/Section 230, except the difference between AI and Twitter is that while Twitter might display content about bomb-making, AI could theoretically teach you exactly how to build one.

And there’s a balance between absent and helicopter mom.

This balance is especially difficult with open-source models, which anyone can use and build with. For instance, there’s been a massive uptick in cybercriminals leveraging Llama and Mistral for email phishing by impersonating people.

Assuming you're not in the bomb-making business, here are some tips to encourage AI to more eagerly respond to your questions (including the classic "I'll tip you $20").


Build AI-powered apps in hours, not days. 

Can you imagine it?

You…building custom AI applications that skyrocket your business…without having to worry about being an AI genius? #promotion

webAI’s Navigator helps businesses build AI apps in hours, not weeks. 

Navigator’s IDE gives you the flexibility to train models with full code OR drag-and-drop in minutes. You're always in the driver's seat—fully controlling your models, with options to deploy them on local devices or cloud platforms.

AI dominates Q&A platforms, only to circle back when it needs new insights.

Ahhh, nothing like the good old days when coding meant scouring Stack Overflow in your parents’ basement at the first sign of a bug?

Now it’s just ChatGPT, and Q&A sites like Stack Overflow are feeling the heat:

search interest for Stack Overflow since ChatGPT emerged

It makes sense: why wade through Stack Overflow for coding answers when ChatGPT can answer all of them in a snap? Well, almost all of them.

Stack Overflow is proving there still is value in really good answers from humans—it just struck a partnership to provide Google’s AI with access to its vast Q&A repository via an API.

Why it matters: There’s a limit to how smart AI chatbots can get from the open internet, and holders of unique data are wisely choosing to monetize access instead of offering it gratis.

Around the Horn.


  • Microsoft announced a “Copilot for Finance” that can pull/reconcile/analyze financial data across Office apps like Excel.

  • The SEC is investigating if Sam Altman misled investors before his ouster as CEO.

  • *Advertising spots in The Neuron are very quickly. Fill out this form ASAP to secure your spot in front of 400,000+ readers.

  • Humanoid robot startup Figure raised $675M from OpenAI, Nvidia and others at a $2.6B valuation.

Intelligent Insights.

“Photo of a single banana, just one banana, not many, ONE" in Midjourney V6

  • The Lone Banana Prompt Problem (link).

  • I Spent a Week With Gemini Pro 1.5—It’s Fantastic (link).

  • The Humane AI Pin worked better than I expected — until it didn’t (link).

  • AI and Knowledge Work—early insights on the impact of AI on knowledge work (link).

  • 10 AI terms everyone should know (link).

  • AI deepfakes are cheap, easy, and coming for the 2024 election (link).

Friday Faculty.

  1. Community Manager at Ideogram (Toronto/remote).

  2. Dozens of open roles at Glean (Bangalore/SF/NYC).

  3. GPU Programming Expert at Mistral (London/SF).

  4. 551 AI companies hiring for all roles.

  5. Frontend Engineer at Phind (SF).

Search from over 10,000 AI roles on The Neuron’s Jobs Board.

A Cat's Commentary.

That’s all for today, for more AI treats, check out our website.

Get your brand in front of 400,000+ professionals here.

See you cool cats on Twitter: @nonmayorpete & @noahedelman02

What'd you think of today's email?

Login or Subscribe to participate in polls.