Imnimo

34 ms·

451.
by Imnimo 4y ago
It will be interesting to see whether this sort of approach works better than something using GPT-4's vision capabilities. Obviously websites are built to be easy to use visually rather than easy to use via the DOM. On the other hand,
452.
by Imnimo 4y ago
I'm familiar with this ELIZA: https://en.wikipedia.org/wiki/ELIZA Is there some recreation of ELIZA that uses ChatGPT under the hood that the article is talking about?
453.
by Imnimo 4y ago
My main take away here is that Turkers are terrible at some of these tasks. The "stance" task is, "Classify the tweet as having a positive stance towards Section 230, a negative stance, or a neutral stance.", and the Tur
454.
by Imnimo 4y ago
I am very sympathetic to the idea that people doing labelling tasks in third world countries deserve more money. But I don't buy that I'm owed something for my tiny contribution to GPT-3's 300B tokens of training data, or Ll
455.
by Imnimo 4y ago
If you don't want the AI to use these words, why bother training it? Just mask those logits before sampling.
456.
by Imnimo 4y ago
In the example near the bottom, where it makes a restaurant reservation and a chickpea salad recipe, is it just generating that recipe from the model itself? It looks like they enable three plugins, WolframAlpha, OpenTable, and Instacart. I
457.
by Imnimo 4y ago
If I see an obviously AI generated face (and these are pretty obvious), I'm going to assume the person is non-existent and whatever communication they've had with me is a scam.
458.
by Imnimo 4y ago
It is absolutely true that the Glazed images will be scooped up by a fresh webscrape for training data for a new model. But there isn't any evidence that this would provide any actual defense. Their paper only studies the fine-tuning s
459.
by Imnimo 4y ago
Disagree with that. We're talking about defense against a situation where a user downloads a fine-tuning colab notebook, uploads a handful of the artist's pictures, and "steals" their style. The user here is interested i
460.
by Imnimo 4y ago
The FAQ says: >Once you add a cloak to an image, the same cloak can prevent different AI models (e.g., Midjourney, Stable Diffusion, etc.) from stealing the style of the cloaked image But my understanding of what they're actually do
461.
by Imnimo 4y ago
Yeah, I tried asking GPT-4 to write a simple Python program to find the millionth prime. It wrote the naive approach where you loop over numbers and for each number, check all of the possible divisors up to its square root, and then keep tr
462.
by Imnimo 4y ago
The absolute gall to say "you probably won't believe your eyes" on a simulated demonstration.
463.
by Imnimo 4y ago
> We will be inviting the top nominees to join a gathering of the brightest technologists, business leaders and ethicists working on trustworthy AI to help get your ideas off the ground. Participants will also have access to mentorship f
464.
by Imnimo 4y ago
It feels pretty obvious to me that it could not, but then again I've been wrong about LLM abilities before!
465.
by Imnimo 4y ago
A class of problem that GPT-4 appears to still really struggle with is variants of common puzzles. For example: >Suppose I have a cabbage, a goat and a lion, and I need to get them across a river. I have a boat that can only carry myse
466.
by Imnimo 4y ago
You can retrieve small amounts of specific information this way. But suppose the model was asked to write code in a new programming language that it doesn't have existing knowledge of. Could it do that given unguided access to the do
467.
by Imnimo 4y ago
Sure, but the value of logit matching is that it allows the smaller model to access richer representations that it has sufficient capacity to express but insufficient capacity to learn. In this case, we're not trying to distill the fu
468.
by Imnimo 4y ago
How expensive do you imagine collecting human feedback data is compared to the cost of doing the original language modelling training?
469.
by Imnimo 4y ago
Matching logits is merely a special case of knowledge distillation. See section 2.1 https://arxiv.org/abs/1503.02531
470.
by Imnimo 4y ago
Completely missed the point, as usual. Consider the amount of world knowledge present in a 175B model vs a 7B model. You can elicit instruction following behavior with a relatively small number of samples , but you're never going to ge
471.
by Imnimo 4y ago
>Alpaca cannot be used for [...] entertainment. Good luck with that.
472.
by Imnimo 4y ago
For a bubble that has no future, the author seems quite worried about its future impacts in the final paragraphs.
473.
by Imnimo 4y ago
Don't forget the cutesy nickname for its employees and users. "Dear Sveebles and Sveeblettes..."
474.
by Imnimo 4y ago
How should I interpret Figure 8? https://journals.plos.org/plosone/article/figure?id=10.1371/... If the difference between the 2012 and 2020 cohorts is roughly the same as between the 2012 and 2002 cohorts, w
475.
by Imnimo 4y ago
I spent so many hours hunting for cool bits of real world terrain and then massaging them into playable scenarios for myself (it's a lot of work to smooth out artifacts in the data and then add functioning water, vegetation, initial ro
476.
by Imnimo 4y ago
>Now, tbh it does rely on the assumption that we are actually seeing more undesired behaviour than before. If that was false then it would falsify the Waluigi hypothesis. This is exactly my point. There is no evidence given that we are s
477.
by Imnimo 4y ago
This is fun to read and think about, but it's also important to keep in mind that this is very light on evidence and is basically fanfic. The fact that the author uses entertaining Waluigi memes shouldn't convince you that it&#x
478.
by Imnimo 4y ago
I'd be curious to see more examination of the questions and answers at the token level, rather than by counting digits or calculating percentage error. For example, according to https://platform.openai.com/tokenizer ,
479.
by Imnimo 4y ago
Not this guy again... Haven't we already been through all of this?
480.
by Imnimo 4y ago
>Handle customer interactions – natural language Q&A, appointment setting, account management, and even tech support, available 24/7, pick up right where you left off, and switch from text to voice as needed. Customer service an

More ›