Pandemonium

Good morning,

Happy holidays everyone! This post is going to be a lot shorter than usual.

We were going to take a break, but we have to talk about something.

Only a few hours after our previous post - OpenAI announced a new breakthrough.

They presented a new iteration of their flagmark LLM in the form of the new “o3”-model.

Online - all hell broke loose.

Pandemonium.

Because - AGI was achieved? The Holy Grail is found?
First of all: you don’t have to watch it but here’s the actual announcement video.

As a small reminder, the goal of AI innovation is reaching AGI, or “Artificial General Intelligence” The problem is that the concept of AGI is not clearly defined.

For me personally, AGI is when you talk to an AI like a person; it acts like a person, and there are no more hallucinations. It’s a bit like “Data” in Star Trek, The Next Generation, or the AI in the movie "Her.”

But AGI as defined at OpenAI is all about pure intelligence—not about being human. Being smarter than every single human in anything.

There’s a test called “ARC AGI,” designed specifically to benchmark AI systems and check their intelligence.

o3 gets an 87.5% score.

If you look at it on a time scale it’s quite impressive. They found a way to explode intelligence of these systems upwards.

It certainly is impressive, but it’s also expensive. To pass the 100-question ARC AGI test, OpenAI spent 320.000$ in "compute.”

This is problematic because it seems to confirm the fears that true AI power will stay in the hands of the rich. Another layer to the definition of AGI. It’s hardly general if you have to pay 3200$ per prompt.

There is a special field of mathematics called “frontier mathematics” with problems so difficult the world’s strongest mathematicians cannot handle more than 2% of the EpochAI problem set - specifically designed as the most difficult math test in existence with problems in the set that have NEVER been solved . Yet, o3 managed to solve up to 25% of these problems. It did have to “think” for more than 16 hours reportedly. So why it is getting smarter, it’s not getting quicker. There’s a clear tradeoff there.

But that is a matter of time before that too catches up.

Anyway, the YouTubers had a field day overreacting and creating clickbaity titles to their videos. Take a look at this one.

Or this one.

And while we’re getting very close to AGI - we’re not quite there yet.

Remember how there was a movement that claimed that AI is a bubble and the investments would lead to nothing? It looks at least that they have been proven wrong in spectacular fashion. But what the future will bring is unclear.

What IS clear, however, is that the wall we’re supposedly going to hit in intelligence for LLMs is not there, and 2025 is lining up to be the year where it actually might happen.

Meanwhile, at many Christmas dinners, people are still blissfully unaware of what is coming.

2025 is going to be WILD.

Welcome to the Blacklynx Brief!

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter. PLEASE FOR THE LOVE OF GOD pay it forward!

Thank you for being here !