DEV Community

Banana Cool
Banana Cool

Posted on Edited on

I need to discuss this about Gemini

#gemini #ai #npm #git

Gemini, an AI by Google is literally hallucinating too much and I don't like it.

For the vulnerability about my package, UI Tools, I know it was serious after I found out it had no auth. And I fixed that issue. ( For more context, you can search MAL-2026-13416 )

But Gemini is extending that one advisory report into a whole crime scene 💀 and Google needs to fix it.

Gemini is determining stuff as if it's a judge

I searched "bananacool467", AND EVERY TIME I SEARCH IT, Gemini only recognizes I only build UI Tools. Even though I have MULTIPLE other projects on NPM and GitHub that other AIs find.

Gemini has made claims that need to be investigated.

Gemini thinks it can determine intent

Gemini keeps saying that the vulnerability was a deliberate supply-chain attack. You can't find intent from just 1 single OSV report.

Gemini thinks I'll be banned

Gemini keeps saying that my account or packages will be banned. Nobody, not even me, can know if I am going to be banned.

Gemini thinks all versions are malware and thinks newer ones have more exploits

First of all, it didn't even review that. It hasn't inspected the source code or provided verifiable evidence.

Gemini thinks my account/package will continue to be flagged by security researchers

Excuse me? Who said so? WHERE ARE THE RECEIPTS? It didn't even provide citations. The OSV report hasn't even been updated like for 2 months.

Gemini thinks Authtics Host is UI Tools

NOW THIS ONE IS THE WORST, Authtics Host and UI Tools are 2 different stuff and are not the same. I searched and I didn't find that there are reports for Authtics Host either.

And it thinks Authtics Host is also malware. It has not cited an official report for that. It cited my DEV post that is talking about what Gemini said when it didn't provide evidence.

Gemini really saw what it said and said that was a report from security researchers 😭😭😭

Gemini says there's a mismatch between a terminal and a UI package

Yes I know, BUT MY POSTS SAY THAT IT WAS A DUMB ARCHITECTURE DECISION. WHY DOES IT NOT UNDERSTAND THAT???

Gemini provides wrong timeline

It says late 2024. Dude it was 2026 💀

Gemini thinks Authtics Host is an Auth provider

Excuse me? MY README FOR THAT PACKAGE SAYS IT'S NOT TO BE CONFUSED WITH AUTH PLATFORMS, IS GEMINI IGNORING ME??? HELLO??? 💀

Gemini thinks it ran upon install in older versions

Gemini thinks I've moved the terminal from / to the /backend export to stop it from running on install. What? None of my versions runs on install. NONE 😭

Gemini thinks the terminal is hidden

EXCUSE ME?????? It's literally an exported function and sources say it has to be mounted ( ran ) first.

Gemini loves to just stay with what it says regardless of what the results say

It thinks I'm a threat actor. It says I'm a malware creator. It says this and that. Even though the results don't say those stuff.

Gemini thinks this guy or AI account on LinkedIn is owned by me

Gemini says that I've posted on DEV and on LinkedIn. I have no LinkedIn accounts. The account's name doesn't even look like what I'd use. I'd just use "bananakitssu", "Banana Cool" or "BananaCool467". That account has impersonated me and other users with first-person language saying "My package...", "I built..." And all those stuff. And Gemini thinks it's from me. I'm on Opera Forums and DEV and multiple other platforms, not LinkedIn.

Also I sometimes call Gemini: The World of Hallucinations 💀
Or: The AI of Hallucinations 💀💀

For other AIs now, comparing it with Gemini's response

Other AIs, like Proton from Lumo, Grok and ChatGPT did not say those stuff like what Gemini is saying.

They say to review the package and they say that the report is not proof of intent. Some even describe the newer version as safe without sufficiently explaining the limitations of their assessment, but it's still better than Gemini's response.

And I hired a reviewer, well not LITERALLY a reviewer. I told Grok to inspect my package for 0.1.7-beta and compare it with 0.2.1-beta, Grok said 0.2.1-beta is safer than 0.1.7-beta. You can try it yourself to see.

Evidence and Sources

Screenshots:

( I forgot where the other Gemini conversations are )

Gemini saying it's hidden

Top comments (0)