11 ms·

The linked twitter account is an AI influencer, so take whatever is written with a grain of salt. Their goal is to get clicks and views by saying controversial

• jmkni 3y ago
  It's crazy to me how quickly the magic wore off, it's only been around for just over 6 months and people went from "holy shit" to "meh" so quickly.
  • anononaut 3y ago
    It needs to be made open source to have its Stable Diffusion halcyon days.
  • lubesGordi 3y ago
    'meh' gets clicks, because it's still amazing to most people. If someone makes a claim that is shocking, it gets clicks. GPP is banking on the underlying report to be non-interpreted, which it looks like was a good bet based on some other comments here.
  • kainosnoema 3y ago
    I don't know why I keep seeing comments like these. We've had to delay the latest API model upgrade because it's dramatically worse at all of our production tasks than the original GPT-4. We've quantified it and it's just not performing anymore. OpenAI needs to address these quality issues or competitor models will start to look much more attractive.
• jeremyjh 3y ago
  So you are just going to ignore the data (not anecdotes) presented in the SP?
  • beebmam 3y ago
    [flagged]
    • majkinetor 3y ago
      Here it is just for you: https://t.co/H6OWhUkzcW https://t.co/H6OWhUkzcW
    • Georgelemental 3y ago
      Replace twitter.com with nitter.net in the url
    • jeremyjh 3y ago
      Well thanks for sharing your ignorance with us.
    • pc86 3y ago
      Then why comment? I'm not trying to be an ass, but this is literally the only comment you've made in this thread. The linked article is a Twitter thread. If you don't have Twitter, fine, move on to the next thing you can actually comment on. Person A offers a reason why we may want to be slightly more skeptical than normal about this information. Person B suggests we can pretty easily look past that. Person C (you) injects themselves in the conversation simply to say that they can't add anything to the conversation. The closest example I can think of that is equally cringe-inducing is two people talking about a show and an unasked third party leaning in to add, "well, I don't have a television so I can't comment."
      • benterix 3y ago
        I don't think the comment is unwarranted - it happened to me the other day: someone refers me to a particular Twitter post and I can't access it without a Twitter account. I don't know if it's a glitch or deliberate but it makes referring people to Twitter similar to referring them to Facebook posts (which is not really practiced on HN). Times change, I guess.
      • autoexec 3y ago
• majkinetor 3y ago
  So the original paper from Stanford and Berkley is also linked to this AI influencer? I am really amazed by this kind of dismissal. Its totally irrelevant who posted the info and how framed it is, as long as you have access to the source.
  • gwd 3y ago
    This comment has an interesting take on it, haven't read the paper to verify the take: https://news.ycombinator.com/item?id=36781968 https://news.ycombinator.com/item?id=36781968 EDIT: FWIW I haven't noticed any such regression. I don't generally use it to find prime numbers, but I do use it for coding, and have been really impressed with what it's able to do. 8<--- This paper is being misinterpreted. The degradations reported are somewhat peculiar to the authors' task selection and evaluation method and can easily result from fine tuning rather than intentionally degrading GPT-4's performance for cost saving reasons. They report 2 degradations: code generation & math problems. In both cases, they report a behavior change (likely fine tuning) rather than a capability decrease (possibly intentional degradation). The paper confuses these a bit: they mostly say behavior, including in the title, but the intro says capability in a couple of places. Code generation: the change they report is that the newer GPT-4 adds non-code text to its output. They don't evaluate the correctness of the code. They merely check if the code is directly executable. So the newer model's attempt to be more helpful counted against it. Math problems (primality checking): to solve this the model needs to do chain of thought. For some weird reason, the newer model doesn't seem to do so when asked to think step by step (but the current ChatGPT-4 does, as you can easily check). The paper doesn't say that the accuracy is worse conditional on doing CoT. The other two tasks are visual reasoning and answering sensitive questions. On the former, they report a slight improvement. On the latter, they report that the filters are much more effective — unsurprising since we know that OpenAI has been heavily tweaking these. In short, everything in the paper is consistent with fine tuning. It is possible that OpenAI is gaslighting everyone by denying that they degraded performance for cost saving purposes — but if so, this paper doesn't provide evidence of it. Still, it's a fascinating study of the unintended consequences of model updates.
    • majkinetor 3y ago
• hospitalJail 3y ago
  Are we talking GPT4 or chatgpt4? Its not disputed that chatGPT4 has degraded in quality as its been 'aligned'.
  • bufferoverflow 3y ago
    There's no such thing called chatgpt4. There's GPT-4 and ChatGPT.
    • hospitalJail 3y ago
      Yes, these are different. Ask each how to do insert illegal activity, let me know if you find a difference.
    • CamperBob2 3y ago
      Huh? When you log on to ChatGPT you're given the choice to use the faster GPT 3.5 model or the slower, rate-limited GPT 4 model. They have both been doing the Flowers for Algernon thing over the last few months. People talk about regression on the part of ChatGPT 4, but the 3.5 chatbot has also been getting worse. No amount of hand-waving and gaslighting from OpenAI (sic) is going to change the prevailing opinion on that.
• naillo 3y ago
  To be fair to sama he did say that "it still seems more impressive on first use than it does after you spend more time with it"[1] but I suspect that was a kind of 4d chess move to hope to be quoted in situations like this when he knew the hype died down a few months later to cushion the blow after the hype cooled off. [1] https://twitter.com/sama/status/1635687853324902401 https://twitter.com/sama/status/1635687853324902401
• IIAOPSW 3y ago
  I feel like it has gotten worse. Your expectations have shifted towards knowing and accepting you have to prompt it a certain way. When you started, "holy shit I can just use natural language". GPT4 is influencing the bar we are setting for it, and the bar we are setting for it is influencing the outcome of its training. If you've been under a rock and you just found out about chatGPT today, you'd find it less fluid and impressive than the rest of us when we started.
• damnesian 3y ago
  > it's just that the magic has worn off as we've used this tech I agree with this. The analogy I use on repeat is the dawn of moving picture making. The first movies were short larks, designed just to elicit a response. Just like when CGI was new- a bunch of over-the-top, sensationalist fluff got made. This tech needs to mature. And we need it to continue to be fed the work of real humans, not AI feeding on AI, a recent phenomenon that hopefully will not turn out to be the norm. If we water and feed it responsibly, it will only grow more capable and useful over time.
  • Filligree 3y ago
    I have actual artifacts -- programs it wrote -- from immediately after the release, which I'm unable to replicate now. Don't claim it hasn't gotten dumber. It's easy to find excuses, but none that explain my experience.
    • YeGoblynQueenne 3y ago
      ChatGPT is a nondeterministic system that's sampling from a huge distribution in which there are thousands or possibly tens of thousands of events with similarly infinitesimal probabilities. There is no guarantee that you will get the same result twice, and in fact you should be a little surprised if you did. That is not an excuse, and explains your experience.
      • Filligree 3y ago
        You're assuming I'm not aware of that, and didn't try multiple times. The original program wasn't the output of a single request; it took several retries and some refining, but all the retries produced something at least mostly correct. Doesn't matter how many times I try it now, it just doesn't work.
        • YeGoblynQueenne 3y ago
          >> You're assuming I'm not aware of that, and didn't try multiple times. I'm not! What you describe is perfectly normal. Of course you'd need multiple tries to get the "right" output and of course if you keep trying you'll get different results. It's a stochastic process and you have very limited means to control the output. If you sample from a model you can expect to get a distribution of results, rather tautologically. Until you've drawn a large enough number of samples there's no way to tell what is a representative sample, and what should surprise you. But what is a "large enough" number of samples is hard to tell, given a large model, like a large language model. >> Doesn't matter how many times I try it now, it just doesn't work. Wanna bet? If you try long enough you'll eventually get results very similar to the ones you got originally. But it might take time. Or not. Either you got lucky the first few times and saw uncommon results, or you're getting unlucky now and seeing uncommon results. It's hard to know until you've spent a lot of time and systematically test the model. Statistics is a bitch. Not least because you need to draw lots of samples and you never know how close you are to the true distribution.
• mckirk 3y ago
  It has definitely gotten worse. At least its performance when prompted in German has degraded noticeably, to the point where it's making grammatical mistakes, which it never has before.
• r3trohack3r 3y ago
  I don’t believe this is true. It’s possible I was blinded by the light, but my programming tasks were previously (during the early access program) being handled by GPT-4 regularly and now they aren’t. I’ve also seen many anecdotes from engineers who had exceptionally early access before GPT-4 was public knowledge. The GPT-4 I use now feels like a shadow of the GPT-4 I used during the early access program. GPT-4, back then, ported dirbuster to POSIX compliant multi-threaded C by name only. It required three prompts, roughly: * “port dirbuster to POSIX compliant c” * “that’s great! You’re almost there, but wordlists are not prefixed with a /, you’ll need to add that yourself. Can you generate a diff updating the file with the fix?” * “This is pretty slow, can we make it more aggressive at scanning?” It helped me write a daemon for FreeBSD that accepted jails definitions as a declarative manifest and managed them in a diff-reconciliation loop kubernetes style. It implemented a binpacking algorithm to fit an arbitrary number of randomly sized images onto a single sheet of A1 paper. Most programming tasks I threw at it, it could work its way through with a little guidance if it could fit in a prompt. Now, it’s basically worthless at helping me with programming tasks beyond trivial problems. Folks who had early access before it was public have commented on how exceptional it was back then. But also how terrifyingly unaligned it was. And the more they aligned the model with acceptable social behavior, the worse the model performed. Alignment goals like “don’t help people plan mass killings” seem to cause regressions in the models performance. I wouldn’t dismiss these comments. If they’re true, it means there is a hyper intelligent early GPT-4 model sitting on a HDD somewhere that dwarfs what we’ve seen publicly. A poorly aligned model that’s down to help no matter what your request is.
  • pixl97 3y ago
    Is this via the ChatGPT web interface, or via the API. I'm wondering if there is a difference?
    • r3trohack3r 3y ago
      The chat interface for me
  • Filligree 3y ago
    This matches my experience. Back right after the release, I had it write a python GUI program using several different frameworks with basically zero input. I also had it ask me for requirements, etc. etc, with absolutely no hand-holding needed. Alas, I never saved that conversation. It's entirely impossible to do so now.
• guluarte 3y ago
  quantization
• mountainriver 3y ago
  Has anyone considered what's obvious here, GPT-4 is overworked.
• eldritch_4ier 3y ago
  Nah it’s definitely gotten worse in my experience. I played with an early access version late last year/early this year and there’s been a strong correlation between how hard they restrict it and how good it is. Queries it used to blow away it now struggles with.
• bluecrab 3y ago
  That's false, the fact that I no longer rely on it is enough proof that it has gotten worse for me.
  • imtringued 3y ago
    Continuous training is the holy grail. Yes you can just retrain with the updated dataset but guess what, you will run into catastrophic forgetting or end up in a different local minima.
• FreshStart 3y ago
  They optimized towards minimal server cycle costs at acceptable product quality while keeping the price stable?
• kainosnoema 3y ago
  This is just not true for us. I'll share again what I've said before: we’ve been testing the upgraded GPT-4 models in the API (where you can control when the upgrade happens), and the newer ones perform significantly worse than the original ones on the same tasks, so we’re having to stay on the older models for now in production. Hope OpenAI figures this out before they force the upgrade because quality has been their biggest moat up until now.
• gorkish 3y ago
  Anyone who has saved any early prompts/responses can plainly attest to this by sending those prompts right back into the system and looking at the output. As to why, it's probably a combination of trying to optimize GPT4 for scale and trying to "improve" it by pre/post processing everything for safety and/or factual accuracy. Regardless of the debate as to whether it is "worse" or "better," the result is the same. The model is guaranteed to have inconsistent performance and capability over time. Because of that alone, I contend it's impossible to develop anything reliable on top of such a mess.
• thunkshift1 3y ago
  What in the whole wide world is an ‘AI influencer’