😺 What comes next...

PLUS: Full FAQ on "Sora"

Welcome, humans.

AI totally popped off last Thursday. Since we like to keep things neat and tidy for you, we mostly gave you the TLDR on the two most significant developments of 2024 so far: Gemini 1.5 Pro & Sora.

Today, we’re taking a beat to let the news (and our stupidly large meals from the long weekend) digest.

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

Why Gemini’s new 1M context limits matter (a lot)…
Sora is incredible—we share what it means for you and society.
Reddit is charging an AI company for $60M to use its data.
OpenAI employees sold their shares at a valuation of $80B.

Gemini’s update is an infinite memory boost for AI chatbots.

Google’s upcoming Gemini Pro update to 1M token context limits is a game-changer—you can upload 700,000 words, compared to just 24,000 words in ChatGPT.

But tokens can feel like credit card points: big numbers that feel fake. What does it all mean?

Context limits are like your short-term memory but for AI chatbots. They cap what AI can recall.

When it comes to AI, context limits = your prompt + what AI writes back.

100x larger context limits means AI can analyze and write a lot more stuff at once. Think really big short-term memory.

Here’s what this could look like:

Compare a new project brief to your entire project archive to find similar examples → write a project plan based on those examples.
Read every department’s 2025 strategy plans → find conflict areas or areas for alignment → write a comprehensive report.
Analyze HOA meeting transcripts to figure out who was the most annoying this year (jk!).

Moreover, context isn't limited to text; it can include video, audio, code, etc.

Now, zoom out: Today, 1 to 10M token contexts. Tomorrow, 100M to 1B? That’s how you go from “read these team reports and answer my Q’s” to “read every public discussion in the company and keep me updated on the important stuff”…

Google Gemini

FROM OUR PARTNERS

Never manually update your CRM again.

"I love updating our CRM,"...said no sales rep EVER!

At The Neuron, we've automated significant portions of our sales process using AI.

Here’s how you can, too:

Attention auto-updates your CRM based on sales calls and other customer interactions—say goodbye to manually entering MEDDIC, BANT, or any other data into your Salesforce, Hubspot, or Pipedrive…

This AI wizard can:

Log crucial data in your CRM.
Write follow-up emails.
Autoscore deals with AI.

Plus, it delivers insights that have pulled in an extra $250k in ARR per rep.

Start automating your CRM with Attention today!

Your FAQ on OpenAI’s video model “Sora”:

First, how good is Sora, actually?

It's mind-blowing… for a text-to-video model. The 60-second showcases are strikingly vivid and realistic, almost as if the AI is “dreaming” up new worlds.

BUT, it’s not close to replacing filmmaking. Sora still struggles with simple elements like hands and camera movements (reminding us of the early days of DALL·E 1):

And it lacks capabilities for tasks like video editing, close-ups, or creating lifelike actors/voices.

Is there anything that compares?

So far, no.

Sora runs laps around Runway, the frontrunner (/only runner?) for most of 2023.

It’s also noticeably better than Pika, which we covered in November.

So will we be able to tell what’s real/AI?!

Yes, for now. The models only get better from here, so there will come a time when AI can produce videos indistinguishable from reality.

This demo was the mort convincing we’ve seen:

So what happens next?

Rules around how this stuff can/will be used. Then regulations. Then more rules.
Look forward to Sora-enhanced video game landscapes and narrative-driven short films (complete with soundtracks)! And inevitably, AI-generated porn, cause, well, that's often where the internet goes first.
More compute. OpenAI mentions safety as the reason Sora isn't widely available, but we reckon it's due to the massive computing resources video generation demands.

Around the Horn.

Paul is a great follow for actionable tutorials!

Reddit is licensing its data to an AI company for $60M.
The NBA launched “NB-AI”, which uses AI to make live NBA games look like your favorite movies.
Some OpenAI employees cashed out their shares at a valuation of $80B in a deal with Thrive Capital.
Masayoshi Son wants to raise $100B for a new AI chip company.

Tuesday Ticker.

Here are the results from last week’s poll:

What y’all had to say:

💨 L.S. “Mostly Github CoPilot and ChatGPT. Other tools only sporadically when needed for a task.
🔎 A.Z. “Of course perplexity.ai, then ChatGPT aaaaand, ChatPDF :)”.
✍️ S. “Github Copilot - coding ChatGPT (teams) - coding, proposal writing, personal (recipes) Bard/Gemini - extract from search”.

A detailed analysis of the survey results is coming soon!