AGI IS COMING ! New model launched by OpenAI. Scarily good at Coding & maths. And mind you, this model existed internally in october 2023 .
Talking product sense with Ridhi
9 min AI interview5 questions

I’m amazed but not scared about AI/AGI, but eventually it will be regulated by government or self regulated by tech companies.
They will increase the premium plans so that mostly enterprise users can afford and to purchase individual plan, a person need to spend more than current price.
Don’t forget that AI is our assistant, don’t rely blindly on it, keep using your own brain for creative stuff. Human creativity >>>>> AI hallucinations

Public releases are always 1-2 versions behind in house stuff. Many people sadly fail to realise this.
They could easily have AGI in house some day and we would never know. Possibly even govts wouldn't know.
Scary time tbh.

Am I wrong in assuming that real world outcomes could be very different than metrics shown in the company's advertisement?

No, you're right. but benchmarks are a decent way to test competence too and usually match up with live performance.

It will be the same way as the automobile industry claims mileage. Lab tested mileage will always be different to the on road individual experiences.

To me this reads : Openai enables self reflection on its models, a technique that has already been around for a year now. Those interested can look into lang graph, Concordia and autogen.

This is not about self reflection, but mcts. You didn’t need a new model for reflection.

Second stage which was about reasoning capabilities has been achieved. For those doomers who still believe that scaling compute won't work . You are gonna get streamrolled by new frontier models

This didn’t exist internally in 2023. The idea did, but no one knew how to scale this. This is based on the mcts paper in May, and recently deepmind also published one about dynamically allocating compute at test time. Both of which led to this model in this form.

how much experience you hold?