6400 Followers
677 Following
AI Architect | North Carolina | AI/ML, IoT, science
WARNING: I talk about kids sometimes
162
33
20
7
24.2
5.2
2.8
1
the progress of OpenAI models
the jump from GPT-1 to GPT-2 is mind blowing
i hate to admit that i wasn’t paying attention to LLMs back then, must’ve been a wild ride to see that
AI has been evolving at an incredible rate. This piece aims to highlight the progress made so far.
Prophet Arena: an unhackable benchmark
this tests models’ ability to predict the future. You can’t train for it, because the data simply didn’t exist prior to test time
plus it’s hugely useful
my prompts to claude code are >60 words, often >300
when people talk about the limits of AI coding, i'm pretty sure they're talking about the limits of writing 5 words and assuming gpt-4o will figure it out
it’s tempting to look at this and be disappointed, but they effectively found a whole other architecture that’s equivalent to transformers
presumably this arch has its own strengths & weaknesses. so now we have transformers, diffusion & HRM
i still want to see it scaled up
the trust you give a model is its moat. it takes time to learn how & when to trust a new model. that’s the switching cost