You Can Now Use ChatGPT Openly, But Good Luck Running It!

It’s now possible to run the first open-source alternative to OpenAI’s ChatGPT on your laptop, but good luck with that.

PaLM + RLHF was made available this week by Philip Wang, the developer in charge of deconstructing closed-source AI systems like Meta’s Make-A-Video. This text-generating model acts in a manner resembling ChatGPT.

The system combines PaLM, a sizable language model from Google, with a technique called Reinforcement Learning with Human Feedback, or RLHF, to build a system that can practically execute every function that ChatGPT can, including email authoring and code recommendations.

PaLM + RLHF, however, needs to be pre-trained. In other words, the system lacks the requisite training on sample data from the web. That would involve generating gigabytes of text from which the model can learn and finding hardware robust enough to manage the training burden. Downloading PaLM + RLHF won’t automatically install a ChatGPT-like experience.

RLHF involves optimizing a language model using a dataset with prompts (such as Explain machine learning to a six-year-old) and what human volunteers anticipate the model will say (such as Machine learning is a form of AI). PaLM is the language model used in PaLM + RLHF.

Both ChatGPT and PaLM + RLHF share a secret sauce called Reinforcement Learning with Human Feedback, which aims to align better language models with what users want them to do.

Several responses are generated by the refined model based on the prompts mentioned above. The volunteers then rank each response from best to worst. Based on the rankings, a ‘reward model’ is trained to find the best responses to a given prompt using the replies from the initial model.

The nonprofit LAION, which contributed the first dataset used to train Stable Diffusion, is in charge of spearheading an attempt to replicate ChatGPT using the most up-to-date machine-learning techniques.

LAION aims to create a future assistant that conducts useful work, accesses APIs, dynamically evaluates content, and much more in addition to writing emails and cover letters. It just got started. But a few weeks ago, a GitHub page with project resources got going.

Related Articles

Leave a Reply

Your email address will not be published.

Back to top button