真心请教：一些"最前沿"的AI代理到底有啥区别
Genuine question: what is the difference between some "state of the art" AI agent/subscription and a SEA remote worker who can do the same task but better (at least not quitting and going on weird moody tangents) for $1 an hour?
真心求问：某些“最前沿”的AI智能体/订阅服务和一个月1美元的东南亚远程工人有什么区别？后者能干同样的活还干得更好（至少不会撂挑子或者突然闹情绪）。
>>109470303
Would you let a SEAmonkey access your emails?
你会让东南亚猴子碰你的邮箱吗？
>>109470343
definitely, with a signed contract and all, all within a legal framework.
确实，签了合同什么都齐了，完全在合法框架内。
and again, they would not quit or hallucinate half the time. there would be no such situations like
而且再说一遍，它们不会干一半就撂挑子或者胡言乱语。不会有这种局面：
>hey you done any work today?
>嘿，今天干活了吗？
>you're right to push back, i didn't
>你怼得对，我没干
because there would be consequences like losing their fucking job.
因为会有后果，比如丢了饭碗。
what is the consequence of claude rm-ing your root partition or nuking your emails? zero, nothing.
Claude删掉你的根分区或者清空你邮件的后果是什么？零，啥也没有。
>>109470303
just get another agent to supervise
再找个agent监督不就得了
>>109470386
you are trying too hard
你用力过猛了
>>109470303
>coding model
>编程模型
>using for emails
>拿来处理邮件
how do people like this even exist?
这种人是咋活到现在的？
>>109470556
i'm gonna ask again, try to reply without seething this time
我再问一次，这次别气急败坏地回
what are the consequences of claude doing fuckall while i am paying for it? who is responsible for that?
Claude在我花钱的时候啥都不干，后果是什么？谁负责？
>>109470749
None of the models are “coding models”, they are all unspecialised general model that the developers benchmaxxed on coding because that’s the best use case for them for now. All and none of the models are coding models.
根本没有“编程模型”这回事，它们全是不分专长的通用模型，开发者只是把跑分刷在编程上而已，因为那是目前最好的用例。所有模型是编程模型，也都不是。
>>109471490
They are.
确实垃圾。
Try ERP and you will understand.
试试角色扮演模型你就懂了。 ⟦ 说实话我并不觉得远程员工能做得更好。与其跟什么三流外包聊，我宁愿跟AI客服说话。
>>109470303
I don't think a remote worker can do the job better to be honest. I'd rather talk to an AI agent on a customer help line than some thirdie at this point.
说实话，我不觉得远程工作者能把这活儿干得更好。到这份上了，我宁愿跟客服热线上的AI机器人聊，也不愿搭理那个外包的。
Although I'd still vastly prefer a local worker who's paid well, but CEOs need their millions so that can't happen.
不过我还是更想跟拿高薪的本土员工打交道，但CEO得赚他们的百万年薪，所以没戏。
Can a SEA remote worker disprove the Jacobian conjecture for $1/hour?
一个东南亚远程员工能拿每小时一美元去推翻雅可比猜想吗？
>>109471536
>don't think a remote worker can do the job better to be honest.
>说实话我并不觉得远程员工能做得更好。
doubt
because if a human does pic related = fired
因为如果人类干出图里这事=被开除
when AI does it = haha lol woopsie, what are you gonna do, sue the computers lmoa
AI干出来=哈哈笑死，你能咋办，去告电脑？笑死我了
>>109470303
sounds like a harness problem to me
这听起来是个控制框架的问题
>>109470303
AI took away our affordable hardware.
AI把我们的平价硬件弄没了。
Fuck AI. Stop giving money to AI companies.
去他妈的AI，别再给AI公司送钱了。
They've made it really "umm ackshually" for no reason.
他们搞得特别“嗯，其实吧”味儿，莫名其妙。
Like if you show it demo code, it picks at obvious shit that you are still implementing and it fails to actually answer your questions
比如你给它看演示代码，它尽挑那些你本来还在实现的显而易见的毛病，结果根本没回答你的问题
>>109471918
That's why I run Claude in Docker and only mount the folders that are needed.
所以我把Claude跑在Docker里，只挂载需要的文件夹。
>>109470303
Oh no no no no no no, paypigxisters?
哦不不不不不不，给AI掏钱的韭菜们？
>>109471918
And what does a thirdie getting fired do for me anon? Doesn't fix my shit that's for sure.
那外包伙计被开除对我有啥用？肯定救不了我的烂摊子。
>>109471918
>no backup
>没备份
>Windows pathing
>Windows路径
fault = you
错在你自己
>>109470303
As we know, AI = Actually Indians
众所周知，AI=其实是印度人（Actually Indians）
>>109472161
>accountability? bruh whats that
>问责？老哥那是什么
Calm down SEA
冷静点，东南亚人
>>109470303
i can't tell if this is bait, or /g/ is just this retarded now.
我分不清这是钓鱼帖，还是/g/版现在真就这么蠢了。
If you can't figure out how to use subagents, build actual software with AI oversight, cron jobs, github, you aren't going to make it.
你要是连子agent、带AI监督的真实软件、定时任务、GitHub都搞不明白，那你真混不下去。
>>109471529
yeah I've tried, it's still a shitty language model. not a coding model, not a writing model, not an erp model, not a math model. a shitty language model
对，我试过了，还是个垃圾语言模型。不是编程模型，不是写作模型，不是角色扮演模型，不是数学模型。就是个垃圾语言模型。
the fact that it does EVEN WORSE at some other task does not make it good at the one it's benchmaxxed for to entice investors
它在别的工作上干得更差，不代表它在那项为骗投资人而刷分的任务上就干得好。
>>109472876
please regale us with the litany of amazing cutting edge "actual software" you've developed with these amazing tools. you said github so you can just link your repos instead of larping
请给我们讲讲你用这些神奇工具搞出了什么牛逼“真实软件”吧。你提到GitHub了，直接甩repo链接呗，别在这儿装。
>>109470303
>>109471918
this "prompt engineer" deserves to get fired
这个"提示工程师"活该被炒
>>109470303
Agents do sometimes fuck things up in ways that humans wouldn't, and it's very frustrating when it happens. Try not to anthropomorphise them too much ("lazy", "caught", "moody"), it's unhealthy and it won't get you better results.
代理有时候会以人类不会的方式搞砸事情，遇到时特别窝火。别过度拟人化它们（说它们"懒"、"被抓包"、"闹脾气"），这不健康，也不会让你得到更好的结果。
Maybe you have a workload that would be better served by a remote worker in a VLCOL area. If you do then then there's nothing stopping you.
也许你的工作负载更适合雇个低生活费地区的远程员工。如果真是这样，也没人拦着你。
To be honest though, most of the time the agent is still going to be faster and easier and cheaper even accounting for the times you have to reword and retry when it fucks up.
说实话，大多数时候agent还是更快更省事更便宜，就算算上它搞砸了要你重新措辞重试的次数也一样。
>>109470303
The SEAmonkey will just put it into a shatbot for you, and the joke is on you if you think otherwise.
SEAmonkey只会帮你把东西扔进一个屎bot里，你要是觉得不是这样，那笑话就在你身上。
>>109470303
I used claude to build a crypto trading agent and hosted it myself on my own server using my own LLM. so far I am up 60%
我用claude建了一个加密货币交易agent，自己托管在自己服务器上，用自己的LLM。目前涨了60%。
it made trades for me that i would have missed
它替我做了一些我本来会错过的交易。
I love AI, I use it for a lot things and agents have made it even better.
我爱AI，我拿它干很多事，agents让体验更上一层楼。
>>109471032
then go to bed next time
那就下次去睡觉吧。
metadoom is epic, please understand.
metadoom是真牛，请理解这一点。
>>109470303
This happened to me. I broke a very large pdf into parts and gave it to claude, to ask it questions. It denied parts I knew, and when I corrected it, it literally said "oh yea your right" before printing out a 1000 page report of what I asked it. It was super embarassing, because I was demoing it to a boomer using a pdf book he read and claimed to know all about. Claude was dead wrong, and the boomer was right. The boomer was right. I feel sick.
这事我遇到过。我把一个非常大的PDF拆成好几部分给claude，问它问题。它否认我明知道的事实，等我纠正它的时候，它直接说“哦你说得对”，然后打出一份我要求的1000页报告。超级尴尬，因为我在给一个老古董演示，用的是他读过并声称了如指掌的一本PDF书。Claude错得离谱，老古董是对的。老古董是对的。我觉得恶心。
>>109470386
>SEAmonkey giving a single fuck about a contract, with someone out of country at that
>SEAmonkey会对合同上心？还是跟境外的人签的那种？
lol
Lmao, even
Lmao，还真敢想。
>>109474144
Lolno. The proof is in the pudding: 0 software or updates that people use or like. Constant outages everywhere. Companies needing to do company-wide addresses telling people to stop the slop right after they told them to tokenmaxx. There are only 3 ways you can be productive with AI: 1) be called pajeet 2) be profoundly clinically retarded (same thing as 1) or 3) lie
Lolno。事实胜于雄辩：0个有人用或有人喜欢的软件或更新。到处频繁宕机。公司需要全公司发通知让人别继续灌水，就在刚让他们tokenmaxx之后。想用AI产出东西只有三条路：1）叫pajeet 2）严重临床智障（跟1一样）或 3）撒谎。
>>109474420
I've wanted to try this and bug bounty shit in the AI age. Do you have any pointers, or did you just freeball it?
我一直想试这个还有AI时代的bug bounty那类活儿。你有什么指点吗，还是纯粹瞎搞过来的？
>>109470343
No, but I have a small mexican man named Pablo in charge of my facebook invites and sexting.
没有，但我有个叫Pablo的墨西哥小个子负责我的facebook邀请和调情短信。
>>109476382
It's fake.
这是假的。
>>109476382
a trading bot is a one shot problem for current models, there's a shitload of them on github and the exchange APIs are well documented, it's a fake post
交易bot对现在的模型来说是小菜一碟，github上有一堆现成的，交易所API文档也很齐全，这是个假帖子。
competitive strategies are another matter entirely
竞争性策略是另一回事。
>>109476382
Here is my flow, it works well.
这是我的流程，效果很好。
Make sure to paper trade for about a month or longer to tighten everything up and adjust the agent as needed before committing real money.
确保先模拟交易大概一个月或更久，把所有东西调顺，按需调整agent，再投入真钱。
It works and works well if you have the right rules and prompting in place.
只要规则和提示词设对了，它很管用，而且效果很好。
>>109476984
Even worse than I expected.
比我想的还糟。
Lmao. AItards really are braindead.
Lmao。AI白痴真是脑子进水。
>>109477041
>60% increase
>60%涨幅
>Worse
>The Europoor meme is real
>欧洲穷鬼梗是真的
Just stop being poor and a luddite
别穷了也别当老古董就行了。
>>109470303
SEAmonkeys are comparable to an 8b model from 2024
SEAmonkey跟2024年的8b模型差不多档次。
>>109471545
Probably some of them could, but I bet they wouldn’t.
也许有些人能做到，但我打赌他们不会。
>>109476984
lambo any day, huh
随时都能上lambo，是吧？
>>109470303
expecting ai to do good at a task like that is foolish imo. ai is great at iterating a piece of software to do exactly what you need, but if you trust it to go thru your inbox and output a table of information that isn't at least partially hallucinated, you will have a bad time. at best, you can funnel the hallucination to create something real. sadly this means its not much use for 95% of normies.
指望AI在这种任务上干得好，我觉得挺蠢的。AI擅长迭代软件实现你确切的需求，但你要是信它能翻完收件箱、输出一张不掺半点幻觉的信息表格，那等着你的就是糟心事。顶多你能把幻觉导向去创造点真实的东西。可惜，这对95%的普通人来说没多大用。
>>109477090
SEAmonkeys are comparable to a 1q model from 2240. That's how bad your gay eye is today.
SEA猴子搁那跟2240年的1Q模型比。你那个同性恋眼睛今天烂得就这水平。
>>109470303
You can not pay your SEA monkey remote worker and it'll still be loyal like a retarded dog.
你连SEA猴子远程工的钱都不用付，它照样像个傻狗一样对你忠心耿耿。
>>109474923
>lol
>Lmao, even
>哈哈哈，连这个都算上了
they take contracts seriously in SEA and that's a good thing
在SEA那边他们把合同当回事，这是好事
i've witnessed Americans offshore their tasks to some philipinos, who eventually tried screwing them one way or another and they're rotting in a philipino prison
我见过美国人把活外包给菲律宾人，结果那些人最后想方设法坑他们，现在烂在菲律宾监狱里了
>>109477356
>people are genuinely shilling for retarded, brown, third world programmers
>真有人正儿八经给那群弱智、棕皮、第三世界的程序员站台
you goyim have gone insane. thankfully you deserve every bad thing that happens to you.
你们这些goyim真是疯了。谢天谢地，你们活该遭一切坏事。
>>109479610
See, you didn't understand that post.
看吧，你压根没读懂那条帖子。
No, that is not at all what's happening. What's happening is that gay eye is impossibly terrible at even basic shit to the point even the shitskins are orders of magnitude more desirable despite being cheaper (and that's with subsidized AI prices).
不，根本不是那么回事。实际情况是那个同性恋眼睛烂到连基本活都干不好，烂到连那些shit skin都比它强好几个数量级，哪怕人家更便宜（还是在AI价格有补贴的情况下）。
>>109474808
>still no argument, just endless seething
>还是没论点，就搁那无限无能狂怒
calm down Rajeet
冷静点，Rajeet
>>109479653
it is loathsome to exist in the same fleshscape with hylics as detached from reality as you
跟你们这种跟现实脱节到这种程度的hyllic活在同一个肉身世界里，真特么恶心
>>109470303
based clanker shitting on lazy humans
基于事实的clanker在喷懒鬼人类
>>109479766
Ai psychosis status: terminal
AI精神病状态：晚期
>>109470343
Would you let SF billionaires access your emails?
你会让旧金山的亿万富翁看你邮件吗？
>>109479912
Oy vey, tune it down with the antisemitism goy
Oy vey，goy，反犹言论收一收
>>109470545
>supervisor goes to sleep
>主管去睡觉了
Whoops.
（如果你觉得这篇文章有启发，可以点击这里付费）
