Imnimo

19 ms·

301.
by Imnimo 3y ago
In the video (at about 4:30), they identify an issue with novelty search - the agent is rewarded for finding areas of the map with irrelevant animations like waves in water, swaying flowers or NPCs walking randomly, as each new frame will h
302.
by Imnimo 3y ago
This is an extremely content-light paper. There's basically zero information on anything important. Just hand-waving about the architecture and the data. Instead it spends its space on things like the equation for MAE and a diagram dep
303.
by Imnimo 3y ago
Does Replit Ghostwriter reference live documentation of the API you're asking for, or is this just coasting on having memorized a recent version of the Slack API? If the latter, this seems like a completely unsustainable approach to de
304.
by Imnimo 3y ago
This feels like it's a mish-mash of concepts that are used in different settings (pre-trained word embeddings, unmasked attention, beam search, ...) but that aren't really used in things like ChatGPT. I guess if you're readin
305.
by Imnimo 3y ago
>In just five days, through thorough data collection and fine-tuning, we not only reduced the system message to under 200 tokens but also transformed the LLM into a robust reasoning tool using a custom JSON format. That sure sounds like
306.
by Imnimo 3y ago
Here's a chat message from a random character in the demo I clicked on: ALEX 8/15/2023, 1:53:43 AM Absolutely! Here's a glimpse of my latest masterpiece. [Attaches a photo of the painting] What do you think? === I feel l
307.
by Imnimo 3y ago
I am very skeptical that doing things like producing "Competitive Quadrant Charts" (see Figure 3) is a helpful form of prompt engineering. The extremely weak ablation (Section 4.4) rings a lot of alarm bells for me also.
308.
by Imnimo 3y ago
Yeah, if the TOS says one thing, and a blogpost pinky-promises another, only one of those two actually counts as far as I'm concerned.
309.
by Imnimo 3y ago
This is an embarrassingly defensive blogpost. Who are they even trying to convince?
310.
by Imnimo 3y ago
This is a very weird article. It should have been trivial to argue against Turner's tweet, because obviously a board game is not real economics. Instead, the author lands on "actually, Monopoly games take a long time" as a go
311.
by Imnimo 3y ago
>victims of domestic violence looking for resources without their abuser finding out or I don't really understand the threat model that would make Tor helpful here.
312.
by Imnimo 3y ago
>So $900k/yr per soldier in their godless AI army when that amount of earnings could qualify thirty-five actors and their families for SAG-AFTRA health insurance is just ghoulish, How many actors could Adam Sandler's $250M Netf
313.
by Imnimo 3y ago
Is it actually an indication that Deep Blue is more intelligent, or does it turn out that Deep Blue is succeeding at chess by means other than intelligence?
314.
by Imnimo 3y ago
The argument made in the article is that “Is Albert Einstein smarter than a toddler?” means things like "Einstein can do arithmetic better than the toddler". But this only makes sense because the toddler and Einstein are similar s
315.
by Imnimo 3y ago
>Einstein can do arithmetic better than the toddler Is Deep Blue more intelligent than Garry Kasparov?
316.
by Imnimo 3y ago
For the experiment in section 5.3, where they try to recover private information from embeddings of clinical notes, it's interesting that the model has to also spend capacity trying to reconstruct non-private information. I wonder if y
317.
by Imnimo 3y ago
Right, I get the goal of removing the outlier activations, but I just don't understand why outlier activations are a consequence of the model trying to "pass". The story from the linked paper earlier in the post ( https:&#x2F
318.
by Imnimo 3y ago
>The problem with using softmax is that it forces each attention head to make an annotation, even if it has no information to add to the output vector. Using softmax to choose among discrete alternatives is great; using it for optional a
319.
by Imnimo 3y ago
>The companies commit to developing robust technical mechanisms to ensure that users know when content is AI generated, such as a watermarking system. I've seen some papers on watermarking, but none of them are what I'd call &q
320.
by Imnimo 3y ago
I know this is really just get the model stop saying "since you've told me that you're an accountant from Peoria" in every reply, but "this feature is irrelevant 99% of the time" is not really selling me on the
321.
by Imnimo 3y ago
This paper rings a lot of alarm bells for me. It's rambling, written in a very odd style, and contains a lot of useless figures. Is this one of those fake paper pranks or something?
322.
by Imnimo 3y ago
I think whether or not this is legitimate would be made more clear if we cut out the AI middle man when explaining it. Rather than "cops have probable cause to search your car because an AI analyzed your driving patterns", say &qu
323.
by Imnimo 3y ago
You don't expect a human can figure out what to do if there is a cone on their hood?
324.
by Imnimo 3y ago
I tried this prompt: >Can you find me a picture of Winnie the Pooh, and describe the image? Bard gives me a link to the image on this article: https://www.psychologytoday.com/ie/blog/the-everyday-unconsc... It
325.
by Imnimo 3y ago
One of my test cases is to ask what the millionth prime is. Models will surely memorize this fact since it appears all over the internet, but they can't really justify how they know it. Claude does quite poorly. If asked point-blank, i
326.
by Imnimo 3y ago
"Inherently" might be too strong of a word, but the default implementations of a lot of key operations are nondeterministic on GPU. With the parallel nature of GPU compute, you can often do things faster if you're willing to
327.
by Imnimo 3y ago
This does at least correctly name the main character, but I think it's otherwise hallucinated. The plot of the story is that Jonathan Rugg is a Quaker who works as a clerk in Philadelphia. His boss sends him on a trip to Jamaica (credi
328.
by Imnimo 3y ago
>On information and belief, the reason ChatGPT can accurately summarize a certain copyrighted book is because that book was copied by OpenAI and ingested by the underlying OpenAI Language Model (either GPT-3.5 or GPT-4) as part of its tr
329.
by Imnimo 3y ago
This readme is very confusing. It says we're going to use the GPT-2 tokenizer, and use GPT-2 as an embedding model. But looking at the code, it seems to use the default LangChain OpenAIEmbeddings and OpenAI LLM. Aren't those text-
330.
by Imnimo 3y ago
I know everyone's on text-embedding-ada-002, so these particular embedding deprecations don't really matter, but I feel like if I were using embeddings at scale, the possibility that I would one day lose access to my embedding mod

More ›