Hacker News Reader

Comment by hmottestad | Hacker News Reader

Is GPT-2 instruction tuned so that it can actually be used for chat? Otherwise I feel it’s more than just a stretch to call this a ChatGPT clone.

View on HN

From the article:

> You can then use this to create something like Chat GPT---just so long as you don't care about the quality of the output. (It's actually pretty terrible output, objectively speaking... But it does run.)

It's unusable and bears no relationship other than the name-dropping. But it's a program that compiles and runs.

By the looks of those in this discussion who are giving high praises about the capabilities of a project whose author admittedly states it doesn't really work, I guess the point is using buzzwords to bait the bandwagon types.

the point seems to be making a small binary to run language model. How useful is that? From a functional perspective, I guess not very but the model can be improved. From a performance / cost perspective, also not very because most of the cost is in training the model and small binary isn't necessarily indicative of quick. I guess it's just kind of interesting that it doesn't take that much code to run the model.

People "demake" games all the time. Nobody complains if those games are missing most of the features of the original because the core gameplay loop (the essence) is the same. That's the point here -- you can make a small-scale, low-resolution version of a major program in a very small space.

> By the looks of those in this discussion who are giving high praises about the capabilities of a project whose author admittedly states it doesn't really work, I guess the point is using buzzwords to bait the bandwagon types.

Instead of assuming everyone else is naive, perhaps consider another perspective?

My thinking was more that ChatGPT is called "Chat" because it allows users to chat with a Generative Pre-trained Transformer.

Reply on news.ycombinator.com