"Andrej Karpathy, a former OpenAI researcher and Tesla's former director of AI, calls his latest project the "best ChatGPT $100 can buy." Called "nanochat," the open-source project, released yesterday for his AI education startup EurekaAI, shows how anyone with a single GPU server and about $100 can build their own mini-ChatGPT that can answer simple questions and write stories and poems."
"Karpathy, who called nanochat a "micro model," wrote on X that models like his should be thought of as "very young children" that "don't have the raw intelligence of their larger cousins." Scale up your spending to $1,000, however, and such a model "quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests.""
"But it's also an example of what has become a growing trend: smaller, cheaper and more specialized models that have fewer parameters, or the "knobs" inside a model that get fine-tuned during training to help it make sense of language, images, or data. Massive large language models (LLMs) may have trillions of parameters, requiring access to GPUs in the cloud and enormous computational power, while the latest small models may have just a few billion parameters."
Nanochat is an open-source, low-cost mini-ChatGPT that can be built on a single GPU for about $100 and perform simple question answering, storytelling, and poetry. The model is a micro model with limited raw intelligence compared with larger models, but increasing compute and spending to about $1,000 yields noticeably better coherence and the ability to solve simple math and coding problems and take multiple-choice tests. The release attracted millions of views and praise from industry figures. The trend toward smaller, cheaper, specialized models emphasizes fewer parameters, lower compute requirements, and suitability for phones, laptops, researchers, startups, and hobbyists. Samsung AI Lab released a Tiny Recursive Model showing impressive efficiency on complex reasoning.
Read at Fortune
Unable to calculate read time
Collection
[
|
...
]