Imnimo

19 ms·

241.
by Imnimo 3y ago
It's interesting that according to the results tables 5 and 6, adding more context even with LongRoPE makes predictions worse on Books3, and only gives improvement on Proof-Pile. What's special about the Proof-Pile dataset? Is t
242.
by Imnimo 3y ago
Would it make sense to have the user click and then use that point as a SAM prompt? It might let you find a book even if the initial SAM query doesn't find it.
243.
by Imnimo 3y ago
I wonder how big of a deal it is that you'd have to retrain the model to support a new or changed file type? It doesn't seem like the repo contains training code, but I could be missing it...
244.
by Imnimo 3y ago
https://openai.com/sora?video=big-sur In this video, there's extremely consistent geometry as the camera moves, but the texture of the trees/shrubs on the top of the cliff on the left seems to remain very flat, re
245.
by Imnimo 3y ago
This is the first time I've been legitimately impressed by one of Google's LLMs (with the obvious caveat that I'm taking the results reported in their tech report at face value).
246.
by Imnimo 3y ago
The author observed that there is little change over time in the percentage of students getting Ds and Fs. If it were the case that the source of the pressure was maintaining graduation rates, wouldn't we expect the inflation to be foc
247.
by Imnimo 3y ago
The most recent is this, which I believe was made after he left Tesla: https://github.com/karpathy/nanoGPT And it's accompanying video series: https://karpathy.ai/zero-to-hero.html Another example
248.
by Imnimo 3y ago
Every time Karpathy quits his job, the field takes a leap forward because he makes some fantastic educational resource in his free time.
249.
by Imnimo 3y ago
If you are a fan of the fractals but feel intimidated by neural networks, the networks used here are actually pretty simple and not so difficult to understand if you are familiar with matrix multiplication. To generate a dataset, he samples
250.
by Imnimo 3y ago
Are steering wheels and gas pedals features that make it easier to hit people?
251.
by Imnimo 3y ago
I'm not even sure the covered model description only applies to LLMs. Is Segment Anything a covered model? The paper self-describes as a "foundation" model, and it presents SoTA results on various benchmarks. It surely falls
252.
by Imnimo 3y ago
I don't understand what the government's story is here. How exactly does shutting down cell service "combat possible threats"?
253.
by Imnimo 3y ago
Given the input "Generate an image of a doctor and a nurse", the GPT-4 model produced the following dall-e prompt for me: >A professional setting showing a doctor and a nurse working together in a hospital. The doctor is wearin
254.
by Imnimo 3y ago
The paper claims: >To avoid benchmark contamination, we follow Guo et al. (2024) to filter out web pages containing questions or answers from English mathematical benchmarks such as GSM8K (Cobbe et al., 2021) and MATH (Hendrycks et al.,
255.
by Imnimo 3y ago
I'm having a very hard time understanding exactly what the author is claiming to show. I've read the "Interpretation: Why Does the Approximation Work?" section a few times, but it feels like it's just a mechanical d
256.
by Imnimo 3y ago
Does a forward pass of Dall-E or a human artist making an image have a higher environmental impact?
257.
by Imnimo 3y ago
While this is a neat demo, I would in general be cautious about translating GPT-4's logits to probabilities. We know from the technical report that its confidences are not well calibrated. See Figure 8: https://arxiv.org&#x2
258.
by Imnimo 3y ago
Is there any possibility of unwinding the distortion caused by site fees, time-value issues, and other market imperfections to recover the true odds implied by the betting markets, or is it just too complicated for that to be feasible?
259.
by Imnimo 3y ago
"Do I recognize the latex template" is my number one filter when clicking through the new arxiv papers each day, so I definitely buy that that would work.
260.
by Imnimo 3y ago
According to their demo, their Limitations section of their github repo is AI-generated. >All AI-generated text detectors aim for accuracy, but none are perfect and can have multiple failure modes (e.g., Binoculars is more proficient in
261.
by Imnimo 3y ago
This feels like a VP at Google told the Chrome team they needed more AI.
262.
by Imnimo 3y ago
Are compounds that might be seen in a clinical trial drawn from the same distribution as compounds that a lab chemist might accidentally taste? Is it weird that there's no overlap of a compound that is both very sweet and medically act
263.
by Imnimo 3y ago
Ah, that does make sense - especially for a category like a baseball player where it may take a lot of other pages to find one that's truly related. Would be expensive for a demo, but not a big deal for a real evaluation.
264.
by Imnimo 3y ago
I tried the demo with Cal Ripken Jr. I was surprised by some of the complex questions: >Which MLB player won the Sporting News MLB Rookie of the Year Award as a pitcher in 1980, and who did Cal Ripken Jr. surpass to hold the record for m
265.
by Imnimo 3y ago
>A transformer is trained on millions of elementary geometry theorems and used as brute search for a proof, which because of the elementary geometry context has both a necessarily elementary structure and can be easily symbolically judge
266.
by Imnimo 3y ago
I'm very curious how often the LM produces a helpful construction. Surely it must be doing better than random chance, but is it throwing out thousands of constructions before it finds a good one, or is it able to generate useful propos
267.
by Imnimo 3y ago
Isn't that the LAION dataset? It's publicly available.
268.
by Imnimo 3y ago
I'm a little skeptical that this artist's work was "used to train Midjourney" just because their name appears on that list. My understanding is that the list in question is just a scraped list of artist names from Wikipe
269.
by Imnimo 3y ago
I think there are two situations in which you might see this kind of rebuttal: 1) The reviewers were truly unreasonable, and the authors feel they have no hope of winning them over. Instead, they try to highlight the flaws in the reviews to
270.
by Imnimo 3y ago
The reviews are pretty harsh, but after reading the paper, I feel they may be too generous. Somehow this paper spends several pages proving things that are both trivial and irrelevant, and spends zero words explaining the architecture of th

More ›