Watershed moments in AI...

Wow - what a big day yesterday was for AI! Huge announcements within a span of 3 hours meant there was a lot to unravel. So that’s what I did. Here’s the juice for you:

Google teases Gemini 1.5 

The first big announcement came courtesy Google, with a teaser on Gemini 1.5, with a context window of a whopping wait-for-it 10 million tokens (multimodal).

The first obvious question is how how does it fare on the needle in the haystack eval as it scales on the context window? As per Jeff Dean, Chief Scientist at Google DeepMind, "For text, Gemini 1.5 Pro achieves 100 percent recall up to 530k tokens, 99.7 percent up to 1M tokens, and 99.2 percent accuracy up to 10M tokens." That is impressive - still to be evaluated by the community, but impressive none the less. Check out Jeff’s full twitter thread here

As for its reasoning capabilities, the following demo by Demis Hassabis shows how good it can be at reasoning, huge if true!

In summary, why is Gemini 1.5 a huge deal? Because the jaw dropping 10M token context:
a) is excellent at retrieval unlike the other SOTA models before
b) Promises real world use case application by generalizing zero-shot to extremely long context data sources
(3) Truly multimodal, shown to work well across text, audio, and video.

A good summary here, as to why this launch is such a huge deal despite OpenAI doing its thing with SORA (more on that below).

So, when can we use it? You may have to wait a while, as this seems like a targeted rollout to some developers and enterprise customers via AI Studio and Vertex AI.

But then, OpenAI did what is does best, outdo! If you ever wondered whether you were living in a simulation, well…wonder harder!

OpenAI releases SORA:

Read more here as the examples shown are truly exceptional, not perfect, but great: https://openai.com/sora

Here are a few to blow your mind:

sama has sharing a ton on his profile too, take a look and wonder…harder:

The progress in one year has been astounding, not Will Smith eating pasta today, but you get the point:

Lots of speculation on what is powering SORA, following is a good take and a good starting point for what the future could look like, as this evolves over time:

How about this for a prediction, is this how it ends:

All in all, if you ever doubted what AI can and will do, this is the watershed moment. It is time to rethink how you want to be positioned for the future. There are multiple opportunities and threats (and insane alpha) in this post. You just need to find that needle in the haystack.

Happy weekend everyone!