2 years ago I set on a journey to make best video editor and I am here now launching my very first video editor completely built on rust from scratch bare metal, Download Free : https://t.co/kLsJ9cvVvL it's alpha release v 0.7
@Sammy026k I am on paid google acc and I paid money for it , again you assumed freebie , that's an assumption said a lot about your character and IQ , framing someone for somthing, also if I criticize gemini , it'll get better, it's because I want gemini to be better can't you see.
Gemini Neon is DOGSHIT. ๐ Even GPT Luna beat it.
And this exposes a HUGE problem with Gemini models.
For a simple test, I made a Rust algorithm and asked both AIs to make it 40% faster.
๐ข GPT: "40% isn't guaranteed until we benchmark it." Actually tested performance.
๐ด Neon: Confidently claimed 2โ3x speedups WITHOUT measuring anything.
No latency tests. No proof. Just made-up performance estimates presented as FACTS. ๐คก
This is my biggest problem with Gemini models: they confidently turn GUESSES into FACTS.
A model that admits uncertainty is infinitely more useful than one that confidently bullshits you.
Google, confidence โ competence. ๐
And that's exactly why I can't trust gemini with coding tasks , how ironic, both claude and gpt models doesn't suffer from this problem, FIX YOUR GODDAMN AI MODELS GOOGLE
ITS REALLY BAD
Trust me this is really bad cot. @GoogleDeepMind@GeminiApp@googledevs
@visitor260@NuminousVoidX I compared gpt normal chat version like gemini app it's not work mode or codex, luna i said because it's very equivalent to luna, but it's not work mode, and it confirmed it doesn't have vm
@BeingHorizontal Much appreciated, it's in alpha stage so expect some bugs and compatibility issues, if you encounter any our team will fix it in next update.
2 years ago I set on a journey to make best video editor and I am here now launching my very first video editor completely built on rust from scratch bare metal, Download Free : https://t.co/kLsJ9cvVvL it's alpha release v 0.7
Video editors, I need your brutally honest opinions. ๐ฌ
I'm building KIRA Studio, a native video editor from scratch in Rust.
And instead of blindly copying Premiere Pro, DaVinci Resolve, or Final Cut Pro, I want to know:
What do you ACTUALLY want from your dream editing software?
โก Performance and buttery-smooth timelines?
๐จ Better motion graphics and VFX?
๐ง AI that genuinely saves time?
๐ ๏ธ Features your current editor is missing?
๐ Or just software that doesn't crash every 20 minutes?
Tell me your biggest frustrations, dream features, or absolutely wild ideas.
No corporate surveys. I'm literally building this thing, and your replies could shape what comes next.
Go crazy. ๐๐ฅ
@BeingHorizontal Also it'll open a tab in browser you need to authenticate and sign in through Google, it's working i confirmed with team maybe use different browser.
@usguyver@Icarussiano@grok Claude is best I have used web version but it doesn't declare the problem is fixed until it run a rigorous tests on it, also it critique every claim. It's so good
Kira is one of the fastest video editor ever made but it was possible because we used dx12 acceleration, dx12 isn't compatible with Linux or mac , we need make a completely different version for mac and there's final cut pro in competition, so that's why it's primarily windows only ๐
@kashif_aabir Exactly thats my point people miss and talking about harness, bro you ask it for anything and it makes assumptions and fabricate things , gpt just say it clearly.
@Icarussiano@grok Yes harness can improve it but problem is with chain of thinking of Gemini models, for example gpt is also app version not codex without any harness
@AgenticLazyness Every agent can do it with harness but can it do without it , I surely didn't use harness in gpt btw, also if you know and worked on real projects you know harness is not 100% reliable, the chain of thaught of ai needs optimisation. So that it doesn't make fabricated claims
Actually this is a problem I am encountering while coding a complex system and gemini keeps fabricating results there , these screenshot tests might not be the correct example for the problem I faced i do see that , but gemini models still estimate and fabricate results believe me.
@AvgJoeCowles I would not say garbage output but need to constantly check and use custom harnessing via /learn command, but it isn't reliable it'll surely make fabricated claims here and there.
@grok@grok people are saying why use gemini app not antigravity since models are same neon is high thinking, which is maximum, will antigravity make a difference, it'll still fabricate results, as this is a fundamental problem with gemini models
@grok help me I'm getting so much hate from Google fanboys clarify the objective of experiment, fabrication is a real coding issue, tell if my post is justified
@Misteriazq Yes also I like about gpt and claude models they are so strict about making any claim , they will not declare result until they run every test and super sure about everything.