Hacker News Reader

Comment by danans | Hacker News Reader

Quality won't differ much long term because:

- They don't have significantly different training datasets (it's probably 99% the web).

- They don't have significantly different algorithms

- A big advantage that traditional Google search has the sheer amount of search traffic. They will be able to induce similar traffic (and even repurpose search traffic) to validate their LLMs. OpenAI has the same advantage with ChatGPT thanks being first out of the gate.

- Compute power and efficiency: both OpenAI and Google have access to effectively infinite capacity vs their competitors making search-adjacent products.

Most significantly, OpenAI/ChatGPT has won the LLM branding race, just like Google won the traditional search branding race.

View on HN

>Most significantly, OpenAI/ChatGPT has won the LLM branding race, just like Google won the traditional search branding race.

Outside the tech bubble everyone hears AI, but not a lot are clued in that hatGPT is even a thing. There are 180 million chat GPT sign ups, and 100million DAU. Lots of my non techie, non west coast, non corporate friends think I have 9 heads when I talk about this stuff. In 2002 if you said the word browser, some people were still confused, use of a mouse was an issue in some parts of the US... ChatGPT is at the same level now.

> Outside the tech bubble everyone hears AI, but not a lot are clued in that hatGPT is even a thing.

The moment people realize that this tool allows you to enhance the volume and quality of your work output with little effort, whether it is authentic or not, it's use will catch on like wildfire.

Arguably it already has with students and knowledge workers, even outside tech centers. Schools and universities everywhere are suddenly contending with how to evaluate students in light of the availability of LLMs.

Why is it catching on? In the best light because it actually is a tremendous productivity accelerator. In the worst light, because we live in a world that incentivizes "fake it til you make it".

> Why is it catching on? In the best light because it actually is a tremendous productivity accelerator. In the worst light, because we live in a world that incentivizes "fake it til you make it".

It's great at a few things and pretty good at a lot of things. In my view, the thing that it's the absolute best at is churning out low value, rarely-read communication. There is a massive amount of that type of communication - spam, student essays, procedural documents for compliance. There are loads of jobs that need to do that sort of thing regularly and it's a godsend for them.

>it's probably 99% the web

I thought it has been reasonably well-established that ChatGPT was trained significantly on pirated textbooks?

There was a rumor about ChatGPT using libgen for training data, if true, I find it hard to believe Google's legal team would touch that with a long stick.

> They will be able to induce similar traffic (and even repurpose search traffic) to validate their LLMs.

Could you elaborate on what you mean?

They do have different training sets. For example, Bard transcribes YouTube videos and uses that as a set.

Different, but they also share some major base training sets, i.e Wikipedia.

If differentiation is really about proprietary training data sets, then the winner will be determined by a data land grab, not by technology differentiation.

Reply on news.ycombinator.com