throwaway1851

9 ms·

31.
by throwaway1851 4y ago
Yes. I have been using ChatGPT quite a bit with programming tasks. One of the things I've been trying to do is using chain-of-thought prompting to ask the model to review its own code, line by line, evaluating it for the presence of a
32.
by throwaway1851 4y ago
I’m interested in this as well. Comparatively little attention has been paid to those 7B model results, but they look quite good against 175B GPT-3. As for ChatGPT, that is GPT-3.5 (same 175B model, but with instruction fine-tuning), plus t
33.
by throwaway1851 4y ago
I actually think this suggests it would be beneficial to just eliminate time limits on standardized tests. There is no inherent need to make time pressure the hard part of the exam. I’ve always thought it reflects a kind of laziness in the
34.
by throwaway1851 4y ago
I have often wished that there were a search engine that would let me filter out pages with monetization (affiliate links, calls to action, ads, etc.). However, every time I consider this idea, I wonder how much of the web would be left. Be
35.
by throwaway1851 4y ago
No, the 13B model outperforms GPT-3. Judging from the metrics published in the paper, it does look like the 7B model is not far off from GPT-3 however.
36.
by throwaway1851 4y ago
The disconnect here is that the model code isn’t the important part. If you read the paper, LLaMA is all about the training data and the training process (which required a ”massive quantity” of compute, to quote the paper). Without relea
37.
by throwaway1851 4y ago
Take y=3x+7. 3 and 7 are parameters, which would be learned by using gradient descent to find the values that make the x’s in the training data produce the corresponding y’s. LLMs are essentially functions like this, only they have billions
38.
by throwaway1851 4y ago
For embeddings, it may be overkill. Smaller BERT-type models can provide good embeddings when fine tuned with a contrastive learning objective. Eg: https://sbert.net .
39.
by throwaway1851 4y ago
This is great tech, and I’m really impressed with the quality of the results. I particularly like how the model declines to answer when the context doesn’t answer the question. One note, though: I’m finding the (mobile) UI very chaotic. Th
40.
by throwaway1851 4y ago
> “And it's a human communication - I am helping someone express their feelings, desires, and fantasies. I am giving them permission to indulge in crazy fantasies and to feel beautiful, powerful, and desirable. Thank fuck that commu
41.
by throwaway1851 4y ago
> You can have a wonky nose and a crooked mouth and a double chin and stick-out teeth >You can have a wonky nose and a crooked mouth and stick-out teeth I’m somewhat amused that wonky noses, crooked mouths, and stick-out teeth are fai
42.
by throwaway1851 4y ago
Alcoholism is also highly prevalent (at about 6% of US adults). Yet, we don’t have public policy that is phobic of alcohol. I think a small percentage of people are very susceptible to addiction. It would be a far better idea to improve met
43.
by throwaway1851 4y ago
The information retrieval step doesn’t use the language model, though. And the query fed to the model doesn’t need to contain any user-identifying information.
44.
by throwaway1851 4y ago
Chatgpt has such a recognizable tone. I recognized it partway through the the first sentence, but I can’t quite put my finger on what tipped me off.
45.
by throwaway1851 4y ago
One of the most critical parts of nontrivial search tasks is evaluating the quality and credibility (and motives) of the sites asserting various facts or opinions. I wouldn’t outsource this part to a statistical model. Even if it cites its
46.
by throwaway1851 4y ago
> a good and trustworthy government Know of any?
47.
by throwaway1851 4y ago
That makes sense in a “but for” type of causality. However, I don’t think that’s what this website is aiming for (however flawed or misleading it might be so far as its methodology is concerned). I think the idea of attribution here is more
48.
by throwaway1851 4y ago
Thanks for the reply! It’s interesting (counterintuitive?) that the problem of models drawing too much from their stored knowledge is mitigated by using models with much greater stored knowledge.
49.
by throwaway1851 4y ago
Hi. I remember reading a question answering paper from Google in 2021 that concluded there was little evidence that models were actually grounding the generated answers in the contexts. Do you know if this problem has been solved?
50.
by throwaway1851 4y ago
Causal attribution would indeed be cost prohibitive. But what about comparing neural activations for pairs or sets of input/output images? It might shed some light on how much “extra” knowledge SD pulls in beyond the memorized features
51.
by throwaway1851 4y ago
That in itself isn’t an indictment of the website. The tool is built on the assumption of being provided an image generated from Stable Diffusion. If you violate that assumption, it isn’t surprising that the tool fails your “test.” By comp
52.
by throwaway1851 4y ago
> Clearly seeing at least one photo of me (and AFAIK it was only trained on thousands of copies of that single photo) was absolutely crucial to the construction of this image, and yet this website isn't finding any I think there are
53.
by throwaway1851 4y ago
I love that your completely reasonable statement of opinion has been grayed out by anonymous, petty censors in this forum.
54.
by throwaway1851 4y ago
I’ve found ChatGPT moderately useful to locate solutions to coding questions that are hard to find on StackOverflow through Google search. (Google seems to have gone all in on embedding-based retrieval these days - all I can get are results
55.
by throwaway1851 4y ago
Those results aren’t real. If you actually try to retrieve them, you will find out that Google stops delivering unique results after a few pages. (Other search engines are the same in this respect.)
56.
by throwaway1851 4y ago
The world doesn’t need to be unfucked for you to be happy, I believe is the point. And that’s a good thing, because the world always has had and always will have its problems.
57.
by throwaway1851 4y ago
Public service announcement: receiving a salary != being exempt from overtime, depending on where you live. In, for example, California, exempt status depends on the actual tasks you perform and the percentage of your hours each week you sp
58.
by throwaway1851 4y ago
Sadly, no, it was the product of 18 months of collaboration between a committee and “stakeholder groups”. Had it been made by a single overzealous HR person, that would’ve been significantly less embarrassing for Stanford.
59.
by throwaway1851 4y ago
That’s addressed in the source.
60.
by throwaway1851 4y ago
I think what’s strange about this ad is the fact that she’s doing a handstand. That isn’t a normal part of getting fit for most people, and it isn’t even safe to attempt by most people in the model’s weight range. Had they shown her using a

More ›