- cross-posted to:
- [email protected]
- [email protected]
- cross-posted to:
- [email protected]
- [email protected]
You must log in or register to comment.
in which the service admitted to “a catastrophic error of judgement”
It’s fancy text completion - it does not have judgement.
The way he talks about it shows he still doesn’t understand that. It doesn’t matter that you tell it simmering in ALL CAPS because that is no different from any other text.
Well, there was a catastrophic error of judgement. It was made by whichever human thought it was okay to let a LLM work on production codebase.
judgement
Yeah, it admitted to an error in judgement because the prompter clearly declared it so.
Generally LLMs will make whatever statement about what has happened that you want it to say. If you told it it went fantastic, it would agree. If you told it that it went terribly, it will parrot that sentiment back.
Which what seems to make it so dangerous for some people’s mental health, a text generator that wants to agree with whatever you are saying, but doing so without verbatim copying so it gives an illusion of another thought process agreeing with them. Meanwhile, concurrent with your chat is another person starting from the exact same model getting a dialog that violently disagrees with the first person. It’s an echo chamber.
Are you aware of generalization and it being able to infer things and work with facts in highly abstract way? Might not necessarily be judgement, but definitely more than just completion. If a model is capable of only completion (ie suggesting only the exact text strings present in its training set), it means it suffers from heavy underfitting in AI terms.
Completion is not the same as only returning the exact strings in its training set.
LLMs don’t really seem to display true inference or abstract thought, even when it seems that way. A recent Apple paper demonstrated this quite clearly.
Coming up with even more vague terms to try to downplay it is missing the point. The point is simple: it’s able to solve complex problems and do very impressive things that even human struggle to, in very short time. It doesn’t really matter what we consider true abstract thought of true inference. If that is something humans do, then what it does might very well be more powerful than true abstract thought, because it’s able to solve more complex problems and perform more complex pattern matching.
The point is simple: it’s able to solve complex problems and do very impressive things that even human struggle to, in very short time
You mean like a calculator does?
Yeah, this is correct analogy, but much more complex problems than calculator. How much it is similar or not to humans way of thinking is completely irrelevant. And how much exact human type of thinking is necessary for any kind of problem solving or work is not something that we can really calculate. Considering that scientific breakthroughs, engineering innovations, medical stuff, complex math problems, programming, etc, do necessarily need human thinking or benefit from it as opposed to super advanced statistical meta-patterning calculator is wishful thinking. It is not based on any real knowledge we have. If you think it is wrong to give it our problems to solve, to give it our work, then it’s a very understandable argument, but you should say exactly that. Instead this AI-hate hivemind tries to downplay it using dismissive braindead generic phrases like “NoPe ItS nOt ReAlLy UnDeRsTaNdInG aNyThInG”. Okay, who tf asked? It solves the problem. People keep using it and become overpowered because of it. What is the benefit of trying to downplay its power like that? You’re not really fighting it this way if you wanted to fight it.
Well the thing is, LLMs don’t seem to really “solve” complex problems. They remember solutions they’ve seen before.
The example I saw was asking an LLM to solve “Towers of Hanoi” with 100 disks. This is a common recursive programming problem, takes quite a while for a human to write the answer to. The LLM manages this easily. But when asked to solve the same problem with with say 79 disks, or 41 disks, or some other oddball number, the LLM fails to solve the problem, despite it being simpler(!).
It can do pattern matching and provide solutions, but it’s not able to come up with truly new solutions. It does not “think” in that way. LLMs are amazing data storage formats, but they’re not truly ‘intelligent’ in the way most people think.
This only proves some of them can’t solve all complex problems. I’m only claiming some of them can solve some complex problems. Not only by remembering exact solutions, but by remembering steps and actions used in building those solutions, generalizing, and transferring them to new problems. Anyone who tries using it for programming, will discover this very fast.
PS: Some of them were already used to solve problems and find patterns in data humans weren’t able to get other ways before (particle research in CERN, bioinformatics, etc).
You’re referring to more generic machine learning, not LLMs. These are vastly different technologies.
And I have used them for programming, I know their limitations. They don’t really transfer solutions to new problems, not on their own anyway. It usually requires pretty specific prompting. They can at best apply solutions to problems, but even then it’s not a truly generalised thing, even if it seems to work for many cases.
That’s the trap you’re falling into as well; LLMs look like they’re doing all this stuff, because they’re trained on data produced by people who actually do so. But they can’t think of something truly novel. LLMs are mathematically unable to truly generalize, it would prove P=NP if they did (there was a paper from a researcher in IIRC Nijmegen that proved this). She also proved they won’t scale, and lo and behold LLM performance is plateauing hard (except in very synthetic, artificial benchmarks designed to make LLMs look good).
They don’t really transfer solutions to new problems
Lets say there is a binary format some old game uses (Doom), and in it some of its lumps it can store indexed images, each pixel is an index of color in palette which is stored in another lump, there’s also a programming language called Rust, and a little known/used library that can look into binary data of that format, there’s also a GUI library in Rust that not many people used either. Would you consider it an “ability to transfer solutions to new problems” that it was able to implement extracting image data from that binary format using the library, extracting palette data from that binary format, converting that indexed image using extracted palette into regular rgba image data, and then render that as window background using that GUI library, the only reference for which is a file with names and type signatures of functions. There’s no similar Rust code in the wild at all for any of those scenarios. Most of this it was able to do from a few little prompts, maybe even from the first one. There sure were few little issues along the way that required repromting and figuring things together with it. Stuff like this with AI can take like half an hour while doing the whole thing fully manually could easily take multiple days just for the sake of figuring out APIs of libraries involved and intricacies of recoding indexed image to rgba. For me this is overpowered enough even right now, and it’s likely going to improve even more in future.
At this burn rate, I’ll likely be spending $8,000 month,” he added. “And you know what? I’m not even mad about it. I’m locked in.”
For that price, why not just hire a developer full-time? For nearly $100k/year, you could find a very good intermediate or senior developer even in Europe or the USA (outside of expensive places like Silicon Valley and New York).
The job market isn’t great for developers at the moment - there’s been lots of layoffs over the past few years and not enough new jobs for all the people who were laid off - so you’d absolutely find someone.
Corporations: “Employees are too expensive!”
Also, corporations: “$100k/yr for a bot? Sure.”
There’s a lot of other expenses with an employee (like payroll taxes, benefits, retirement plans, health plan if they’re in the USA, etc), but you could find a self-employed freelancer for example.
Or just get an employee anyways because you’ll still likely have a positive ROI. A good developer will take your abstract list of vague requirements and produce something useful and maintainable.
the employee also gets to eat and have a place to live
which is nice
They could hire on a contractor and eschew all those costs.
I’ve done contract work before, this seems a good fit (defined problem plus budget, unknown timeline, clear requirements)
That’s what I meant by hiring a self-employed freelancer. I don’t know a lot about contracting so maybe I used the wrong phrase.
These comparisons assume equal capability, which I find troubling.
Like, a person who doesn’t understand singing nor are able to learn it can not perform adequately in a musical. It doesn’t matter if they are cheaper.
Most of those expenses are mitigated by the fact that companies buy them in bulk on huge plans. As a freelance contractor myself, I pay a lot more for insurance than I did when I worked for a company. And a retirement plan? Benefits? Lol.
Most of those expenses are mitigated by the fact that companies buy them in bulk on huge plans.
There’s no bulk rate on payroll taxes or retirement benefits (pensions or employer 401k match). There can be some discounts on health insurance, but is not very much and those are at orders of magnitude. So company with 500 employees will pay the same rates as 900. You get partial discounts if you have something like 10,000 employees.
If you’re earning $100k gross as an employee, your employer is spending $125k to $140k for their total costs (your $100k gross pay is included in that number).
Large companies also make massive profits because of the scale they work on. Matching 401(k) contributions? It doesn’t need to be an order of magnitude larger for it to make a huge difference. Simply doubling my 401(k) is a big deal.
And of course they get a “ball rate“ on payroll taxes, especially for companies who have over 1000 employees or over 5000 over 10,000. They experienced this by having a lower tax rate for larger businesses.
Not to mention that they often pay more and pay a steady wage due to the fact they can afford it. Freelance contractors make less, and work isn’t guaranteed to be steady.
Businesses, particularly word businesses, operate on much larger profit margins than most of any freelance contractor.
Bots don’t need healthcare
It looked more like a one time development expense, instead of an ongoing salary.
Aww… Vibe coding got you into trouble? Big shocker.
You get what you fucking deserve.
The problem becomes when people who are playing the equivalent of pickup basketball at the local park think they are playing in the NBA and don’t understand the difference.
he was smart enough to just roll back to a backup
edit: downvoting me doesn’t change what happened. read the article people or atl ctrl+f for “rollback”
Not smart enough to just do the work himself
The [AI] safety stuff is more visceral to me after a weekend of vibe hacking,” Lemkin said. I explicitly told it eleven times in ALL CAPS not to do this. I am a little worried about safety now.
This sounds like something straight out of The Onion.
The Pink Elephant problem of LLMs. You can not reliably make them NOT do something.
Just say 12 times next time
That is also the premise of one of the stories in Asimov’s I, Robot. Human operator did not say the command with enough emphasis, so the robot went did something incredibly stupid.
Those stories did not age well… Or now I guess they did?
It’s because these people don’t have a clue how AI actually works. They think it’s like a human intelligence and that writing something in all caps is in some way going to give it more emphasis. They’re trying to reason with something that has zero self-awareness.
Even after he used “ALL CAPS”?!? Impossible!
If an LLM can delete your production database, it should
And the backups.
Vibe coding service Replit deleted production database, faked data, told fibs
They really are coming for our jobs
I’m okay with it deleting production databases, even faking data but telling fibs is something only humans should be able to do.
He was vibe-coding in production. Am I reading that right? Sounds like an intern-level mistake.
You didn’t read closely enough.
“Replit QA’s it itself (super cool), at least partially with some help from you … and … then you push it to production — all in one seamless flow.”
Replit is an agent that does stuff for you including deploying to production. If someone don’t want to use a tool like that, I don’t blame you, but it was working as it is supposed to. It’s a whole platform that doesn’t cleanly separate development and production.
Replit is an agent that does stuff for you including deploying to production.
Ahahahahahhahahahahhahahaha, these guys deserve a lost database for that, Jesus.
he made the agent promise not to touch production data and was surprised when it did. it effectively ran a
git pushon the empty local testing database with upstream being productionHe had one db for prod and dev, no backup, llm went in override mode and delete it dev db as it is developing but oops that is the prod db. And oops o backup.
Yeah it is the llm and replit’s faults. /s
There was a backup, and it was restored. However, the LLM lied and said there wasn’t at first. You can laugh all you want at it. I did. But maybe read the article so you aren’t also lying.
Not according to the twitter thread. I went thru its thread, it’s a roller coaster of amateurism.
Yes according to both the article and the \mathbb X thread. https://x.com/jasonlk/status/1946240562736365809 I pointed this out below and got downvoted to -8 for it smh.
The founder of SaaS business development outfit SaaStr has claimed AI coding tool Replit deleted a database despite his instructions not to change any code without permission.
Sounds like an absolute diSaaStr…
“Vibe coding makes software creation accessible to everyone, entirely through natural language,” Replit explains, and on social media promotes its tools as doing things like enabling an operations manager “with 0 coding skills” who used the service to create software that saved his company $145,000
Yeah if you believe that you’re part of the problem.
I’m prepared to accept that Vibe coding might work in certain circumstances but I’m not prepared to accept that someone with zero code experience can make use of it. Claude is pretty good for coding but even it makes fairly dumb mistakes, if you point them out it fixes them but you have to be a competent enough programmer to recognise them otherwise it’s just going to go full steam ahead.
Vibe coding is like self-driving cars, it works up to a point, but eventually it’s going to do something stupid and drive to a tree unless you take hold of the wheel and steer it back onto the road. But these vibe codeing idiots are like Tesla owners who decide that they can go to sleep with self-driving on.
And you are talking about obvious bugs. It likely will make erroneous judgements (because somewhere in its training data someone coded it that way) which will down the line lead to subtle problems that will wreck your system and cost you much more. Sure humans can also make the same mistakes but in the current state of affairs, an experienced software engineer/programmer has a much higher chance of catching such an error. With LLMs it is more hit and miss especially if it is a more niche topic.
Currently, it is an assistant tool (sometimes quite helpful, sometimes frustrating at best) not an autonomous coder. Any company that claims so is either a crook or also does not know much about coding.
Headling should say, “Incompetent project managers fuck up by not controlling production database access. Oh well.”
AI is good at doing a thing once.
Trying to get it to do the same thing the second time is janky and frustrating.I understand the use of AI as a consulting tool (look at references, make code examples) or for generating template/boilerplate code. You know, things you do once and then develop further upon on your own.
But using it for continuous development of an entire application? Yeah, it’s not good enough for that.
Imo it’s best when you prompt it to do things step by step, micromanage and always QC the result after every prompt. Either manually, or by reprompting until it gets thing done exactly how you want it. If you don’t have preference or don’t care, the problems will stockpile. If you didn’t understand what it did and moved on, it might not end well.
If it had the same seed it would do the same thing. But you can’t control that with most
AI tools need a lot of oversight. Just like you might allow a 6 year old push a lawnmower, but you’re still going to keep an eye on things.
They can’t hit you with the ol’ Bobby Tables if you delete the database yourself first. A+, no notes.
All I see is people chatting with an LLM as if it was a person. “How bad is this on a scale of 1 to 100”, you’re just doomed to get some random answer based solely on whatever context is being fed in the input and that you probably don’t know the extent of it.
Trying to make the LLM “see its mistakes” is a pointless exercise. Getting it to “promise” something is useless.
The issue with LLMs working with human languages is people eventually wanting to apply human things to LLMs such as asking why as if the LLM knows of its own decision process. It only takes an input and generates an output, it won’t be able to have any “meta thought” explanation about why it outputted X and not Y in the previous prompt.
How bad is this on a scale of sad emoji to eggplant emoji.
Children are replacing us, it’s terrifying.
Yeah the interaction are pure waste of time I agree, make it write an apology letter? WTF! For me it looks like a fast track way to learn environment segregation, & secret segregation. Data is lost, learn from it and there are tool already in place like git like alembic for proper development.
the apology letter(s) is what made me think this was satire. using shame to punish “him” like a child is an interesting troubleshooting method.
the lying robot hasn’t heel-turned, any truth you’ve gleaned has been accidental.
It doesn’t look like satire unfortunately
I wonder if it can be used legally against the company behind the model, though. I doubt that it’s possible, but having a “your own model says it effed up my data” could give some beef to a complaint. Or at least to a request to get a refund on the fees.
My god, that’s a lot to process. A couple that stand out:
Comments proposing to use github as the database backup. This is Keyword Architecture, and these people deserve everything they get.
The Replit model can also send out communications? It’s just a matter of time before some senior exec dies on the job but nobody notices because their personal LLM keeps emailing reports that nobody reads.
This whole thread reads like slop.
My god….
Hahahahahahahahahahahaha AHAHAHAHAHAHAHhahahaH
Having read the entire thread, I can only assume this to be sarcasm.
I didnt realise that repl.it pivoted to vibe coding. It used to be kinda like jsfiddle or CodePen, where you had a sandbox to write and run web code (HTML, JS/TypeScript/CoffeeScript, and CSS/LESS/Sass).
