Andrej Karpathy is an AI researcher and engineer known for work on deep learning and autonomous driving, who now often posts publicly about model capability and training practice. His recent posts focus on the widening gap in how people understand LLMs, noting that roughly 6 billion people have barely encountered them, and share evaluation and interpretation tips, such as a land-or-water coordinate test and asking models to explain things in the aerospace controlled language ASD-STE100. He also amplified Dario Amodei's essay on slowing the AI frontier.
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population…
Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet. [Quoting @celestepoasts]: results for all claudes
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originall…
I love this and really hope we can come together as an industry and make it happen. [Quoting @DarioAmodei]: We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilater…
3.8B LLM Trained to 0.384 CORE for $998An individual trained a 3.8B-parameter model scoring 0.384 on CORE using 65B tokens in 43 hours for $998, built a config-driven little-lm framework, and found B200s offered better value than H100s for this work.
The A.I. Daily