Something Altman and OpenAI seem to recognize

Whassup @knickfilmskool.bsky.social fam?? Let’s go Knicks!!!!

Interesting context from @fchollet.bsky.social the creator of ARC-AGI regarding the “saturation” of the current ARC benchmark. It will be very revealing to see how o3 performs on the new ARC2 benchmark in 2025

TRS-80 Model I. With the max 16K RAM option 😂. It was a year before I upgraded to 48K and from cassette tape to a floppy drive. It was the bomb!

March 2023 comparison of GPT-4 and GPT-3.5 performance on academic exams:

openai.com/index/gpt-4-...

While we’ll have to wait to test real-world performance, to put in perspective o3 performance announced by OpenAI today, GPT-4 (March 2023) got ~40% of AP calc right (GPT-3.5 = 0%). Now o3 performs better than humans on PhD-level science. Just incredible progress.

openai.com/12-days/

So, I was expecting great things from 2.0 Flash, the newest consumer model. Here’s what I got from the same prompt. This highlights the challenge foundation model providers have balancing usefulness and “risk” - especially as the public becomes more familiar with LLM’s true capabilities

Here’s the prompt I gave Gemini Advanced 1.5 and its response. You can see it breaks down the complex task into meaningful and manageable chunks. And, amazingly, goes off and happily does the work, mining the web and social media sites, and synthesizing its findings. 2/3

So many choices, but I’d have to with…