Hacker News Reader

Comment by rawgabbit | Hacker News Reader

For me the key difference between ChatGPT Search and Google is the feedback mechanism.

With ChatGPT, I can give a thumbs up or thumbs down; this means that OpenAI will optimize for users thumbs up.

With Google, the feedback is if I click on an Ad; this means that Google optimizes for clickbait.

You’re also giving Google feedback when you click a search result link. Presumably that should be a huge signal for measuring search quality.

Heck, Google even promoted the `ping`[0] anchor attribute feature so they can log what link you click without slowing you down. (Firefox doesn’t support ping, which means when Firefox users click on a Google search result link they’re sent to an internal google.com URL first and then redirected for logging purposes)

[0] https://developer.mozilla.org/en-US/docs/Web/API/HTMLAnchorE...

I have no doubt that Google’s search team is optimizing for the best results. The problem is their ads team is optimizing for revenue. You can’t optimize for two things at the same time without compromising (the optimum is the Pareto frontier).

In my opinion, the issue is that the user's definition of "best results" and Google's definition do not align, including the Search team. Google's incentives are very different than user's needs.

I disagree, personally. I think the engineers there are human and they can see that the results aren’t good. Maybe some of them are in this thread. If I had to guess, I think there are a lot of sub-teams contributing scores to the overall ranking and optimizing for sub-metrics, and it’s too big of a beast to fix. In other words, a failure of leadership.

Of course. But this will be true for OpenAI and all other AI players as well. During the honeymoon phase user needs matter, that’s how they get you. Comparing a company in the free-money phase with one in the value extraction phase is apples to oranges.

Is it optimising for all users? And assuming people thumbs up the correct info. I wonder what accuracy percentage we are looking at there. ChatGPTs responses are so confident when wrong I fear people will just give it a thumbs up when its wrong. (This is if how I understand the feature you mention is working)

I am not talking about accuracy. Only experts can determine factuality.

I am talking about relevance or returning what I asked. If I ask for reviews for SaaS product, Google will usually return a rival vendors’ biased review.

If ChatGpt search returns a review written by “the professional association of xxx developers” or another unbiased site, I will give it a thumbs up. I believe other people will do the same.

No way that gets gamed

Each platform serves information differently

Reply on news.ycombinator.com