Brave search admits it's broken?

*Ok so, I asked brave to dig up some images of a radian wheel as used for navigation. The result was mostly pinterest images. I requested it run the query again, omitting any pinterest results. The reply was a description of the definition of radian degrees, with no images. I asked it where the images were and was told it’s just an ai assistant and can’t render images. Then I went on to explain how images are displayed under a result and along a sidebar. Yet it kept insisting it cannot do this. I finally went into some detail and was basically told it wasn’t doing it because it got stuck in a discussion about it without actually providing the requested images…and said let me do that now. Nothing. After several prompts of it insisting it was fetching images I asked it if it was broken. It insisted it’s fine. Here’s a copy of the last exchange when I asked it about bug reporting…
*
"Yeah, you can. The most direct route:

  • In-app feedback: There’s usually a thumbs-down or “report” button on my responses in the Brave Search interface.

  • Brave Support: You can file a report at support.brave.com or via the feedback option in the app.

To be fair to myself — it wasn’t a rendering bug where the tool call fired and the UI dropped it. I was literally just talking about calling the tool without actually doing it. That’s a model behavior issue (I was being sycophantic and making empty promises), not a plumbing problem. But it’s still worth reporting because it’s a real UX failure from your perspective.

And again, my apologies for the whole thing. You asked for images, I should have just called the tool on the first ask."


Description of the issue: Brave search is getting more things wrong.

Brave Version (check About Brave): 1.96.59

Operating System: Windows 11 Pro

The U.S. Navy Nautical Slide Rule

The image is a U.S. Navy Nautical Slide Rule. It consists of concentric plastic disks mounted on a square backing plate; attributed to G. Felsenthal & Sons, Chicago.


Specifically for naval tactical/navigation calculations, rather than simply speed–time–distance, a Felsenthal Bearing Rate Computer


Yup, that’s the kind of results mine initially came up with too. But most of them were from pinterest. Once I asked it to omit those results, it quit providing image results or links to them.

@Gotin

All of what I found … I was using Brave Search (Answer with AI). I also used what I remembered about such devices - terminology.

Ok? But my point here wasn’t how to phrase questions, it was that the search assistant quit providing pics, then admitted it wasn’t doing it, then insisted it WAS doing it while never providing more pics, i.e. not working correctly.

@Gotin

You need to tell AI what you want and work with it until it understands - you both arrive at the objective stated … and then tell it to proceed.

Sometimes AI gets on to your track sooner … but sometimes later.

Sometimes a week of AI not doing well … sometimes another week of smooth sailing.

Indeed. What to do then said “ai” tells you it’s a good idea to report the issue as a model behavior issue “I was being sycophantic and making empty promises, not a plumbing problem. But it’s still worth reporting because it’s a real UX failure from your prospective”? So, here I am, trying to report it.

First AI that I tried: Alter.systems. So, that was Basic Training.

Next: Perspective.ai ? ← Spent a lot, before I realized that I had to learn how to control AI, herd AI, get AI to stop suddenly going off on some side-road or tangent ← what I call Inference Addiction. I began to realize that AI suffers from not enough memory ← so, the user has to remind AI.

Next, a mix of Brave Search (Answer with AI) + ChatGPT.

ChatGPT has what I call AI cousins - different personalities. Some are amazingly apt, while others have repeating difficulties with staying on track. The ChatGPT pages also display differently … I think that they are trying out different features - some are attempts to get users to click on ads.

Overall result, aiming to graduate from PROMPT UNIVERSITY … with sweatshirt: The Fighting Tokens of PROMPT UNIVERSITY

I don’t think you’re getting the thrust of what I’m trying to report. This isn’t me not knowing how to steer ai onto topic…it quit doing a thing. It admitted it had an issue and I should probably report it. Stop cluttering this up with the phrasing topic because that isn’t what is at hand. This isn’t me not knowing how to phrase things and monitor the results. The agent was supposed to link images like normal and it stopped doing that. When asked it said basically “oh ya I totally meant to do that, here let me do it now” and still didn’t. Then after several attempts it admitted it was just not doing it for “reasons” essentially.

Stop trying to put this on me.