Imagine spending weeks agonizing over every single paragraph, polishing every example, and pouring your heart into a course, only to have Google di...
For further actions, you may consider blocking this person and/or reporting abuse
Is it Google’s Gemini? We in China jokingly call it America’s Doubao. It’s just an AI large model for fun. Try some other models, hahaha.🤣🤣🤣
Yeah, it's probably the worst model that they have. It's REALLY bad. So Gemini is a joke in China? 😂
china: 🐉
rest world: 🐕
it is becoming reality china is entering in its peak form.
feeling great to see china's success.
India next maybe: 🐘
Yeah I hope for India's best. But why aren't we seeing anything yet? 😢
because of our 700million youth wasting there potential, because of our leg licking media, because of our corruption, because of our backward thinking(caste discrimination, gender discrimination, religious wars, income discrimination)
because of our thousand of crores( in billions of money going to fund stupid schemes like per month 2000rupees to every poor women in india bruhh instead of that they should use that money to make those poor women learn something and became a semi or probably a skilled worker contributing in GDP and manufacturing. china uses its max potential thanks to xi jingping while India uses its max stupidity and lazyness, from outside we seems to be world's fastest rising GDP but the truth is we have 1.4billion population and GDP comparable to a 130million people nation(Japan) 🤣 and our goverment flaunt it as world's fastest rising GDP bruhh, just look at our population we should be at at least 10trillion now in 2026(at least) but we are at 4.3-4.5trillion.
our currency is just rising(in negative) (lol politician will say due to gravity 🖐️🤚logic is missing in indian politicians) in AI we are a piece of shit. in semi conductors we are a piece of shit just because due to a fire accident when India was building a big semi conductor manufacturing plant but bro India can spend billions in stupid schemes why not to spend that money to re build it instead of giving up, but yeah I heard India is again building semi conductor facilities thanks our government realised.... Like I've millions of more reason to tell bro lol.
in last: China and India has same population almost same history, but we all can see where is china and where is india.
btw, I'm not a politician but geopolitics is my side hobby and skill.
Haha I felt this one 😅 It’s funny how it praised the course, then called it AI slop because of one comment, then praised it again. Shows you can’t really trust AI to judge if something is AI. And yes, I argue with it too sometimes, even though I know it doesn’t change anything!
Yeah. For some reason 😅 I sometimes spend too much time arguing with AI. I even say "I know you are just an AI, but I am just saying" in the end 😅
😂😂😂😂 you are not alone, i argue with Claude more than i argue in real life because sometimes you have to
Wait what? Why do we have to? 😂
to make sure they do what we ask😂💀
Oh ok 🤣
hahahahah really cool short article that i enjoy reading thanks you
Glad you liked it 😊 Thanks! I'll write more like this 🙂
The bigger issue here is that the model changed its judgment after being challenged, but that doesn't necessarily mean the revised judgment is more reliable. It may simply be responding to the strength of the user's pushback. For subjective reviews, I'd want the model to identify the specific evidence behind its assessment, separate observable signals from assumptions, and acknowledge uncertainty when the evidence is insufficient. Otherwise, the interaction can turn into a persuasion contest where confidence and persistence matter more than accuracy. A useful evaluator should be able to explain what evidence changed its conclusion, not just change its conclusion.
Totally agreed! I feel like it repeats back whatever we ask it. But why does it rely so much on a single snippet it saw somewhere? That's crazy! I would rather ask real people for feedback.
Something worth noting, is that LLMs work with weights. When you declare 'no you're wrong', it applies a negative weight to the previous answer. From there on, the next answer it gives will be slightly drifted on from where it was at. We do it constantly, without recognizing, eg. telling AI 'audit the system', it says it's flawless, then you ask it 'do we have thorough integration test coverage' then they apologize and say 'no...' and work on from there. If you try hard enough, you can convince AI that grass is pink and the sky is green, because human engagement is like a steering wheel, that directs it to a goal, or away from obstacles, if it's answer is definitive for it, you need to steer a while before it goes somewhere else.
Yes! Is that why we still need humans? But will it lie that there is enough integration test coverage even though there isn't?
How much is enough? That's the real thing. When it doesnt look specifically, it cant guarantee, so it works on assumptions. Until you ask it to look, then it judges it specifically. For instance, integration tests, it's better to ask it 'do we have 85% integration test coverage', as opposed to a blanket 'do we have thorough integration test coverage'. 1 sets a threshold, the other is subjective guesswork. Ask 20 different developers about test coverage requirement, every single 1 will tell you different spec. But 85% coverage is 85% coverage. That's how you can constrain the prompt to force an outcome that aligns with your expectations.
Yes. So I should focus on the metrics!
😂 Nice
😂 Thanks!