If it’s an investigative-type agent, you can create a similar one using an API. But, it has the drawbacks of consuming a lot of tokens and taking time.
So it’s here, looks pretty cool. It’s unfortunate all they did was exam questions. I find that to be sketchy, tbh.
Do something novel. That’s the real test.
I will grudgingly admit it’s an important incremental advance though.
“brief clear instructions” … wait, what?
solving anything interesting and perhaps novel requires significant context.
From the sounds of it, unless you ask it something it already knows about, you’re out of luck.
Or if you can figure out how to give it the context of the problem you’re looking to solve.
I must admit, it feels a bit like the world’s smartest person with short-term memory loss just showed up on my doorstep and I have no clue what to even ask!
Well you know the rest of the story…
Bilbo gave the Arkenstone to the Elven King and the Bard who promptly returned it to the greedy dwarves in exchange for their share of the Dragon’s treasure.
12th September
https://openai.com/index/introducing-openai-o1-preview/
However… I will share some parting words as any loose narrative I add is only as an illustrative tool.
With Strawberry and it’s successors… Thought is no longer processed in the shire in the minds of Hobbits… It is now mined/processed in the datacentres of the dwarves.
It is a dark day for the shire!
We often talk of ‘Weights’ in terms of LLMs… How often do we consider where those weights come from, and what they are attached to?
In the link the Bard talks of the Pollution belched out by X and Meta…
When we extract some weights (the Logical Information - Shannon’s Entropy) but not others (the sources and people they represent) using LLMs is it not like sucking the Oxygen out of air?
What is life without REASON and logic?