Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Trae: An AI-powered IDE by ByteDance

trae.ai

Post reply on HN
Print view View original

161–170 of 428 posts

Re: Trae: An AI-powered IDE by ByteDance

#161

Post by guappa »

nikkwong wrote: post #154
ioulaum wrote: post #133

Earlier quoted context omitted.

It was always BS though. AOC did an interview where she explains (what should be obvious to anyone), that US Gov had no proof whatsoever beyond trying to point at that Chinese law (much like Section 702 in the US), that makes TikTok seem risky. Except that TikTok isn't legally a Chinese company, and isn't really under their jurisdiction. Chinese companies like to be under the control of the Chinese government no more…

Absolutely and verifiably false, again! There are numerous data points that prove that the CCP exploits the TikTok algorithm to influence western views towards a pro-eastern bend. See my comment history for references. The close-doored national intelligence briefing that the senate got about the way the CCP manipulates the algorithm and on how TikTok is a national security threat resulted in a 100-0 vote in favor of…

And the USA wants only fb and google to be able to tell them what propaganda to push. It's all a propaganda war, it's not about propaganda-vs-real news.

Re: Trae: An AI-powered IDE by ByteDance

#162

Post by guappa »

msoad wrote: post #67

Can someone please do some security research on this to see what it sends when it calls home? I'm terrified to install a binary like that

VScode phones home… what's the difference?

Use something like emacs/vim/kate/eclipse/qtcreator if you want to avoid "phone home" software.

Re: Trae: An AI-powered IDE by ByteDance

#163

Post by homebrewer »

nurumaik wrote: post #145
ioulaum wrote: post #135

Apparently the target market for this is supposed to be "Chinese coders living abroad". Thus it supporting English and Chinese languages. And it may be a competitive step taken after Alibaba launched their new coding assistant.

Isn't any modern llm inherently multilingual?

Not really, there' a world of difference between how every model I've tried¹ handles English vs my own native language². It's especially noticeable with smaller models, which produce coherent English, but completely break down in other languages: spewing out incorrect grammar, mixing words from other languages where they make zero sense, etc.

¹: which is dozens of them at this point, open and not.

²: with lots of speakers and training data.

Re: Trae: An AI-powered IDE by ByteDance

#164

Post by guappa »

raincole wrote: post #89
toprerules wrote: post #26

I feel like I'm taking crazy pills, because every time I try to use AI for any of my projects - including the "better" Anthropic and OpenAI models, I get terribly written, buggy code that has nothing to do with the domain or question. Unless I'm feeding it actual softball, single function questions it adds almost no value to my dev cycle. Mind you, all of this terrible code also costs a ton in tokens. It's sad becaus…

> but I feel like we aren't allowed to be honest anymore that LLMs aren't going to replace programmers or moderate our expectations. But literally every single programmer I know has this opinion.

I know developers that love it.

Of course they are terrible at their job, so AI doesn't make their code much worse.

Re: Trae: An AI-powered IDE by ByteDance

#166

Post by guappa »

toprerules wrote: post #87
nosefurhairdo wrote: post #44

Earlier quoted context omitted.

It takes some practice to get good using them. I use it as a sophisticated autocomplete. For example, if I'm building an api client layer in a web frontend, I might have `fooApi.ts`, `mockFooApi.ts`, and `barApi.ts`. instead of writing `mockBarApi.ts`, something that might take 5 - 15 minutes, I can feed those 3 existing files into context, and a few seconds and pennies later I have a nearly perfect file ready for re…

That's just it, I try not to write the boring parts at all. That's what libraries and design pattern are for - avoiding writing boilerplate. I certainly can see it being useful for writing additional tests, although I am very worried based on what I've seen that it will introduce more bugs than it's worth, which is why it's still a tool I use for hobby projects and not for real work.

> I certainly can see it being useful for writing additional tests

If your metric is number of tests, sure. If your metric is number of useful tests… eh

Re: Trae: An AI-powered IDE by ByteDance

#168

Post by jcheng »

thesz wrote: post #100
throwaway2037 wrote: post #76

Earlier quoted context omitted.

Can you provide a specific example of where an LLM failed? If you show us your prompt, your "want/need", and the result, we can better judge your situation. My guess: Your domain is weird/rare, so LLMs are terrible because their training data is very limited.

> Your domain is weird/rare, so LLMs are terrible because their training data is very limited. And this is how knowledge collapse [1] shows its' head. [1] https://arxiv.org/abs/2404.03502 I am not a big fan of LLMs so I try them once in a while, asking to "implement blocked clause decomposition in Haskell." They can recite a paper (and several other papers, with references) almost verbatim, but they do not posess eno…

I gave Claude 3.5 Sonnet your prompt and it generated this: https://claude.site/artifacts/7aa41881-937e-4863-a407-c999ea...

With this example usage:

    -- Example usage:
    let clause1 = Set.fromList [1, 2]  -- represents (x1 ∨ x2)
    let clause2 = Set.fromList [-1, 3] -- represents (¬x1 ∨ x3)
    let formula = Set.fromList [clause1, clause2]
    
    -- Decompose the formula
    let (nonBlocked, blocked) = decompose formula
How did it do?

Re: Trae: An AI-powered IDE by ByteDance

#169

Post by guappa »

kristopolous wrote: post #108
toprerules wrote: post #26

I feel like I'm taking crazy pills, because every time I try to use AI for any of my projects - including the "better" Anthropic and OpenAI models, I get terribly written, buggy code that has nothing to do with the domain or question. Unless I'm feeding it actual softball, single function questions it adds almost no value to my dev cycle. Mind you, all of this terrible code also costs a ton in tokens. It's sad becaus…

I've got a colleague at work and when he's wrong, I've got to get Claude to agree with me before he believes me. I used to think he was competent but now I think he's a moron

If you formulate the questions right, AI will agree to anything.

Re: Trae: An AI-powered IDE by ByteDance

#170

Post by guappa »

throwaway2037 wrote: post #76
toprerules wrote: post #26

I feel like I'm taking crazy pills, because every time I try to use AI for any of my projects - including the "better" Anthropic and OpenAI models, I get terribly written, buggy code that has nothing to do with the domain or question. Unless I'm feeding it actual softball, single function questions it adds almost no value to my dev cycle. Mind you, all of this terrible code also costs a ton in tokens. It's sad becaus…

Can you provide a specific example of where an LLM failed? If you show us your prompt, your "want/need", and the result, we can better judge your situation. My guess: Your domain is weird/rare, so LLMs are terrible because their training data is very limited.

> My guess: Your domain is weird/rare, so LLMs are terrible because their training data is very limited.

The training data is always limited, because if there was an existing software that already did what I'm doing, I'd be using it instead of writing it :)

So by definition, unless I'm learning how to program and doing exercises that thousands have done before me, I'm doing something for which there is no training data.

Post reply on HN