HNHacker News
TopNewBestAskShowJobs

fnqi8ckfek

56 karma · joined December 22, 2024

submissionscomments
fnqi8ckfek··on A combo of ChatGPT and spot EC2 instances screwed me over
> Every time ChatGPT gives you a sketch recommendation, ask for downsides.

Funny I thought the take away was "don't outsource your critical thinking".

fnqi8ckfek··on I Run LLMs Locally
I know nothing about gpus. Should I be assuming that when people say "ram" in the context of gpus they always mean vram?
fnqi8ckfek··on Piracy in the UK: the failed war on illegal content
I don't know. TPB seems less active than ever. I know that's not UK specific, but does not bode well nonetheless.
fnqi8ckfek··on Missiles are now the biggest killer of airline passengers
Now this is gonna shock you, but USSR is Russia.

Russia created the USSR as a legal framework to exert power over its neighbors. This was engineered by Stalin himself.

The most important feature of Russian culture is the sentiment that Russia is great but the world is conspiring to put it down. That's 100% the same in present day Russia as it was in the USSR. They're the same.

fnqi8ckfek··on Offline Reinforcement Learning for LLM Multi-Step Reasoning
I don't buy it. LLMs can already put together long phrases without needing RL for training. And crucially those long phrases _make sense_ they're not use syntactically correct, which is what you'd expect by learning to predict the next word.

So clearly it's possible to get lond correlations Right even without RL.

fnqi8ckfek··on Ask HN: Predictions for 2025?
So they aren't confidence levels. They're odds, but he's trying to sound sophisticated.
fnqi8ckfek··on Tell HN: I just updated my wife's Chrome, and uBlock is no longer supported
Mozilla let's me use ublock origin, Google doesn't.
fnqi8ckfek··on Ask HN: Predictions for 2025?
How do those confidence levels work? It looks to me you're just assigning probabilities to outcomes.
fnqi8ckfek··on Offline Reinforcement Learning for LLM Multi-Step Reasoning
Gee I can't even understand the abstract.

Can someone explain in plain English how RL is even doable here, let alone desirable?

fnqi8ckfek··on Huge math error corrected in black plastic study; authors say it doesn't matter
That's dodging my question. I know theyre just predicting the next word, but that doesn't explain how that works so amazingly well in this case. There's a few steps missing.

If anything this seems like an ad for an AI. When I try these things they can even solve leetcode, for which they actually have the solution in the training set.

fnqi8ckfek··on How we made our AI code review bot stop leaving nitpicky comments
Ok?

So if the boss's nitpick is "you must pass your code thru an indicator that doesn't affect the output", I'm going to ignore it. I don't understand what point you're trying to make.

fnqi8ckfek··on How we made our AI code review bot stop leaving nitpicky comments
Reading from the other comments here, I'm the only one thinking that this is just busywork...? Just get rid of the thing, it's a solution without a problem.
fnqi8ckfek··on How we made our AI code review bot stop leaving nitpicky comments
My boss can't stop nitpicking, and I've started telling him that "if your comment doesn't affect the output I'm ignoring it".
fnqi8ckfek··on Huge math error corrected in black plastic study; authors say it doesn't matter
That's actually crazy. How does it do it? I thought that these machines only "predict the next word".