• The Neuron
  • Posts
  • 😺 What you should know about RAG

😺 What you should know about RAG

PLUS: ChatGPT one-liner hacks

Welcome, humans.

Brazil passed a piece of legislation written entirely by ChatGPT, the first of its kind. We usually don’t get political on this newsletter, but let the record show: we would vote for ChatGPT-4v if it ran for office in 2024. #CommanderInCode

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

  • A friendly explainer on AI that “learns” your business.

  • One-liner hacks that make ChatGPT do what you want.

  • Google faked its big AI demo video from yesterday.

  • Deeper reads into what went down at OpenAI.

Here’s one thing holding up AI for businesses (for non-technical readers!).

Earlier this week, OpenAI COO Brad Lightcap said “we’re early” for AI in business.

One of the big problems is a deceptively simple task: getting an AI to “learn” your business.

Get out your pens and papers, class - time for a mini breakdown.

ChatGPT has learned from the entire Internet, but it doesn’t have access to your company’s internal systems, wikis, documents, etc.

One fix is a total mouthful called “retrieval-augmented generation” (RAG). Here’s the high-schooler version of how it works:

  • Break up all of the stuff you want the AI to reference to small snippets and put them in a big database

  • When you ask the AI a question, a system will guess which small snippet has the “best” info

  • The system tells the AI to use the snippet to answer the question

It’s like telling the AI to hit “cmd+F” before answering you.

Again, sounds simple! “Chat with PDF” tools like PDF.ai and ChatPDF do a mini version of this. But there are two big devils in them details:

  1. It can be tricky to know what’s the “best” info and pull it 100% of the time. #1 question from legal, guaranteed: what happens when it’s wrong/inaccurate?

  2. Giving an AI snippets means it can’t answer questions about the whole thing:

    1. Works: “What was our marketing team’s goal in Q2 2021?” (only needs a snippet relating to Marketing Q2 2021)

    2. Doesn’t work: “What were our biggest successes as a marketing team in the history of the company?” (needs everything the marketing team has ever done!)

So it’s good, but doesn’t fix the problem for many use cases.

The alternative to RAG? Training the AI to actually learn the info. That’s a whole Pandora’s box for next lesson - model access, training data, running the model, etc.

That’s all kiddos—class dismissed!


Turn text into videos in minutes with this AI video tool.

You know those AI avatars that can say anything and sound indistinguishable from a real person?

DeepBrain’s AI Studios is your go-to for this. 

Select from 100+ avatars, 80+ languages, and 100+ video templates → create endless customized videos for your marketing campaigns, sales outreach, or onboarding videos. 

The best part: you create a free video today—no credit card required!

Make sure to join their upcoming AI Studios Feature Update webinar on 12/14 to maximize your use of this awesome platform.

How to hack ChatGPT’s psyche.


Look at the image above: are they inside jokes or very useful ChatGPT one-liners? Surprisingly, these all work to make ChatGPT respond better to you:

(Turns out we couldn’t find a reference for the “grandmothers” one…we’ll keep searching)

The snarky but true takeaway: If you notice (as we have) that ChatGPT is getting worse or refusing to answer, start with calming encouragements, then progress towards lying, bribery and threats! It works!

Try putting one in your ChatGPT prompts or custom instructions settings.

Around the Horn.

  • Google faked a lot of its demo video for Gemini. (We didn’t highlight the video cause we suspected this. Now it’s confirmed.)

  • AI pranks that we fell for in 2023.

  • McDonald’s says they’re “doing AI”. Other fast food chains say their AI solutions actually rely heavily on people.

  • Trevor Noah interviews Sam Altman about the whole CEO drama.

  • The EU AI Act is still under debate, and it’s getting intense.

  • Twitter’s “grok” chatbot is rolling out to Premium+ subscribers ($16/mth) over the next week.

Intelligent Insights.

Here are some pieces that got us thinking this week:

  • AI’s Spicy-Mayo Problem (link).

  • From King to Exile to King Again: The Inside Story of Sam Altman’s Whiplash Week (link).

  • How a Fervent Belief Split Silicon Valley—and Fueled the Blowup at OpenAI (link).

  • Inside U.S. Efforts to Untangle an A.I. Giant’s Ties to China (link).

  • How Jensen Huang’s Nvidia Is Powering the A.I. Revolution (link).

Friday Faculty.

  1. Software Engineer, Product at Adept AI (SF).

  2. 123 AI/ML jobs at Apple (everywhere).

  3. People Manager at Mistral AI (Paris).

  4. 8 open positions at Perplexity (SF).

  5. AI Engineer at Notion (SF).

A Cat's Commentary.

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

Get your brand in front of 220,000+ 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.