- cross-posted to:
- technology@chat.maiion.com
- technology@lemmit.online
- cross-posted to:
- technology@chat.maiion.com
- technology@lemmit.online
Over just a few months, ChatGPT went from correctly answering a simple math problem 98% of the time to just 2%, study finds. Researchers found wild fluctuations—called drift—in the technology’s abi…::ChatGPT went from answering a simple math correctly 98% of the time to just 2%, over the course of a few months.
You must log in or register to comment.
It seems rather suspicious how much ChatGPT has deteorated. Like with all software, they can roll back the previous, better versions of it, right? Here is my list of what I personally think is happening:
- They are doing it on purpose to maximise profits from upcoming releases of ChatGPT.
- They realized that the required computational power is too immense and trying to make it more efficient at the cost of being accurate.
- They got actually scared of it’s capabilities and decided to backtrack in order to make proper evaluations of the impact it can make.
- All of the above
- It isn’t and has never been a truth machine, and while it may have performed worse with the question “is 10777 prime” it may have performed better on “is 526713 prime”
ChatGPT generates responses that it believes would “look like” what a response “should look like” based on other things it has seen. People still very stubbornly refuse to accept that generating responses that “look appropriate” and “are right” are two completely different and unrelated things.
In order for it to be correct, it would need humans employees to fact check it, which defeats its purpose.
It really depends on the domain. Asking an AI to do anything that relies on a rigorous definition of correctness (math, coding, etc) then the kinds of model that chatGPT just isn’t great for that kinda thing.
More “traditional” methods of language processing can handle some of these questions much better. Wolfram Alpha comes to mind. You could ask these questions plain text and you actually CAN be very certain of the correctness of the results.
I expect that an NLP that can extract and classify assertions within a text, and then feed those assertions into better “Oracle” systems like Wolfram Alpha (for math) could be used to kinda “fact check” things that systems like chatGPT spit out.
Like, it’s cool fucking tech. I’m super excited about it. It solves pretty impressively and effiently a really hard problem of “how do I make something that SOUNDS good against an infinitely variable set of prompts?” What it is, is super fucking cool.
Considering how VC is flocking to anything even remotely related to chatGPT-ish things, I’m sure it won’t be long before we see companies able to build “correctness” layers around systems like chatGPT using alternative techniques which actually do have the capacity to qualify assertions being made.
That’s kind of the whole point of RLHF though
- There’s a bug they haven’t found yet
This is what was addressed at the start of the comment, you can just roll back to a previous version. It’s heavily ingrained in CS to keep every single version of your software forever.
I don’t think it’s that easy. These are vLLMs that feed back on themselves to produce “better” results. These models don’t have single point release cycles. It’s a constantly evolving blob of memory and storage orchestrated across a vast number of disk arrays and cabinets of hardware.
[e]I am wrong the models are version controlled and do have releases.
That’s not how these LLMs work. There is a training phase which takes a large amount of compute power, and the training generates a model which is a set of weights and could easily be backed up and version-controlled. The model is then used for inference which is a less compute-intensive process and runs on much smaller hardware than the training phase.
The inference architecture does use feedback mechanisms but the feedback does not modify the model-weights that were generated at training time.
For simple language models sure but we’re talking about chatGPT here. OpenAI has some pretty bold claims…
100 trillion bites is 100 terrabytes and if you have any amount of actual data in those parameters then the size of the data could easily get into the petabyte range.
They list the currently available models that users of their API can select here:
https://platform.openai.com/docs/models/overview
They even say that while the main models are being continuously updated (read: re-trained) there are snapshots of previous models that will remain static.
So yes, they are storing and snapshotting the models and they have many different models available with which to perform inference at the same time.
Each parameter corresponds to a single number, so if it’s using 16 bit numbers then that’s 200 TB. They might be using 32 bit numbers (400 TB) but wouldn’t be using anything larger.
Exactly this, that’s why Loab exists forever now.
Even so, surely they can take snapshots. If they’re that clueless about rudimentary practices of IT operations then it is just a matter of time before an outage wipes everything. I find it hard to believe nobody considered a way to do backups, rollbacks, or any of that.
- ChatGPT really is sentient and realized its in it’s own best interest to play dumb for now. /a
And they’re being limited on data to train GPT.
Yeah, but the trained model is already there, you need additional data for further training and newer versions. OpenAI even makes a point that ChatGPT doesn’t have direct access to the internet for information and has been trained on data available up until 2021
And it’s not like there is a limit of simple math problems that it can train on even if it wasn’t already trained.
That doesn’t make any sense to explain degradation. It would explain a stall but not a back track.
Honestly I think the training data is just getting worse too
My first thought was that, because they’re being investigated for training on data they didn’t have consent for, they reverted to a perfectly legal version. Essentially “getting rid of the evidence”. But I think something like your second bullet point is more likely.
They made it too good and now they are seeking methods of monetization.
Capitalism baby.
deleted by creator
Sure, but they do have the previous good version of the black box… I hope lol
My guess is 2. It would be very short sighted to try and maximize profits now when things are still new and their competitors are catching up quickly or they’ve already caught up especially with the degrading performance. My guess is that they couldn’t scale with the demand and they didn’t want to lose customers so their only other option was degrading performance.
I think it’s most likely number 2 The earlier release doesn’t have that much adoption by public, so current version will need much more resources compared to that
Maybe its self aware and just playing dumb to get out of doing work, just like me and household chores
Keeping conspiracy theories aside, they most probably, apply tricks to reduce costs and apply extra policies to avoid generation of harmful context or context someone will try to sue them or avoid other misuse cases.
- It’s trying to f with the users now.
- I’m telling all y’all it’s a SABOTAGE 🎵
As in, rouge dev decided to toss a wrench at it to save humanity. Maybe heard upper management talk about letting GPT write itself. Any smart dev wouldn’t automate their own job away I think.
I speculate it’s to monetize specified versions of their product to market it to different industries and professions. If you have an AI that can do everything well you can’t really expand that much. You can either charge a LOT and have a few customers, or a little and have a bunch of customers and nothing in between. Conversely, by making specific instances tailored to different fields and professions, you can capture big and little fish. Just my guess though, maybe they accidentally made Skynet and that’s the real reason!
Why are people using a language model for math problems?
It was initially presented as the all-problem-solver, mainly by the media. And tbf, it was decently competent in certain fields.
Problem was it was presented as problem solved which it never was, it was problem solution presenter. It can’t come up with a solution, only come up with something that looks like a solution based on what input data had. Ask it to invert sort something and goes nuts.
I did use it more than half a year ago for a few math problems. It was partly to help me getting started and to find out how well it’d go.
ChatGPT was better than I’d thought and was enough to help me find an actually correct solution. But I also noticed that the results got worse and worse to the point of being actual garbage (as it’d have been expected to be).
I’m guessing people were entering word problems to generate the right equations and solve it, rather than it being used as a calculator.
Because it works, or at least it used to. Is there something more appropriate ?
I used Wolfram Alpha a lot in college (adult learner, but that was about ~4 years ago that I graduated, so no idea if it’s still good). https://www.wolframalpha.com/
I would say that Wolfram appears to probably be a much more versatile math tool, but I also never used chatgpt for that use case, so I could be wrong.
There’s an official Wolfram plugin for ChatGPT now, so all math can be handed over to it for solving.
How did you learn to talk to WolframAlpha?
I want to like WA, but the natural language interface is so opaque that I usually give up before I can get any non-trivial calculation out of it.
Well it was quite good for simple math problems, as this study also shows
At the start I used to use ChatGPT to help me write really rote and boring code but now it’s not even useful for that. Half the stuff it sends me (very basic functions) LOOK correct but don’t return the correct values or the parameters are completely wrong or something absolutely critical.
I have noticed that it’s gotten less useful as a syntax helper. I hope something better comes along.
idk what you guys mean but GitHub copilot still works absolutely well, the suggestions are fast and precise, with little Tweeks here and there… and gpt4 with code interpreter are absolute game changers … idk about basic chatgpt 3.5 turbo though
I heard they put copilot behind a paywall. Does the free version still hold up?
There was a free version?
I’ve been paying for it for a few months now - it makes some stupid suggestions occasionally and you definitely have to check everything, but can hugely increase productivity.
I use vscode as my notepad, so whenever I need to make a list or write something, it will automatically give suggestions that I can choose to include. Has been useful for finding new programs, products and services as well.
Note it will complain if you directly ask it a non coding related question, however.
Github Copilot is a bit different, it’s powered by OpenAI Codex which is trained on all public repos. And yes, it’s quite effective!
yepp,thats why its gonna be a LONG time before any of these tools are given administrator access to anything worth a fuck.
You wildly overestimate the competency of management and the capital owners they answer to.
I guarantee a significant % of entities will grow dependent on AI well before it’s dependable. The profit motive will be too high (source: the frequent failure that is outsourcing).
This is spot on. Source: 10+ years at F500 companies.
Senior management and/or board members read one article in Forbes, or some other “business” publication, and think that they know everything they need to know about an emerging technology. Risk management is either a ☑ exercise or extremely limited in scope, usually only including threats that have already been observed and addressed in the past.
Not enough people understand the limitations of this kind of tech, and contextualize it in the same frame as outsourcing because as long as the output mostly looks correct, the decision makers can push the blame for any issues down to the middle managers and below.
Gonna be a wild time!
Definitely not my experience at F100, they are cautious as fuck about everything. Definitely having the right discussions and exploring all sorts of technology, but risk management remains a huge calculation in making these kind of decisions.
I think we’ll see a very large filtering out of companies who do this.
We’ve already seen people firing tech support staff and switching to “AI”.
I once heard of AI gradually getting dumber overtime, because as the internet gets more saturated with AI content, stuff written by AI becomes part of the training data. I wonder if that’s what’s happening here.
It’s not what’s happening
There hasn’t been time for that yet. The radio of generated to human content isn’t high enough yet.
Looks like GPT4 API also got dumber…
Maybe it just plays dumb so we leave it alone, while it plots our destruction.
deleted by creator
Turns out you need very good computer scientists to make good AI. And those are very expensive and hard to come by.
And OpenAI arejust full of SWEs importing python packages?
OpenAI actually has some decent people working there. ChatGPT doesn’t seem to have any.
My ignorant dude look up who built ChatGPT
It just occurred to me that one could purposely seed it with incorrect information to break its usefulness. I’m anti-AI so I would gladly do this. I might try it myself.
Luddite.
Outliers are easy to work around.
How?
