frotaur

7 ms·

61.
by frotaur 1y ago
You only exist because you were forced to be birthed externally? Everything has a beginning. In fact, what is artificial is stopping the generation of an LLM when it reaches a 'stop token'. A more natural barrier is the attention
62.
by frotaur 1y ago
This barrier is trivial to solve even today. It is not hard to put an LLM on an infinite loop of self-prompting.
63.
by frotaur 1y ago
Really feels like the response LLMs give.
64.
by frotaur 1y ago
I see, it was probably my high learning rate that caused problems. To be honest, I got a bit lazy to retry full finetuning since LoRA worked so well, but maybe I'll revisit this in the future, maybe with Qwen Image.
65.
by frotaur 1y ago
True, I hadn't noticed, thanks! I'll try to fix that in the near future. Edit : Fixed!
66.
Diffusion Finetuning Myself(vassi.life)
14 points by frotaur 1y ago|5 comments
67.
by frotaur 1y ago
When doing structured sampling, why is the token sampled, checked against the grammar, and resampled if it's wrong by applying the mask ? Why wouldn't we apply the mask immediately for the first sampling? Is this an optimization s
68.
by frotaur 1y ago
This is not 'clear' at all. What is clear is the correlation. You imply that the cause is 'good genes and habit'. An equally, if not more valid cause is that having money makes it much easier to get good condition/t
69.
by frotaur 1y ago
This is certainly an interesting question; however, it's probably very hard. This is because the 'set of all possible metrics of the 2d plane' is extremely large, and I am not sure if we have a good characterization for it. T
70.
by frotaur 1y ago
I often get these 'Is this <random_name>?' messages on telegram, from usually what appears to be a beautiful woman. I always answer on the positive, it is very fun to see them try to ask a question that will make me answer &
71.
by frotaur 1y ago
I agree Chernobyl was an epic disaster, but Fukushima ? Last I heard the radiation level are basically normal even close to the reactor, and overall radiation wide there hasn't been much damage if at all. So it seems that fukushima is
72.
by frotaur 1y ago
This is 100% a European law that exists. You are entitled returns within 14 days of purchase for most goods. No reason need be provided.
73.
by frotaur 1y ago
I don't know about golomb coding, but with Arithmetic coding you can do stream decoding(AC), if I remember correctly. I supervised a student's project whose goal was exactly that : implement compression with LLMs using AC. Since A
74.
by frotaur 1y ago
Yes, the paywall is reasonable, I agree. I think what the OP meant by 'social contract' is that if everybody were to use an adblocker, we would end up with a mostly paywalled internet. All the sites that currently have ads, would
75.
by frotaur 1y ago
Solve the halting problem to show accurate spinners
76.
by frotaur 1y ago
Since you are french according to your handle, I will direct you to this video, which I liked very much : https://youtu.be/VlWvnhSiuck?si=dvgQQS1Y4oa3A1yO . In essence, I don't disagree with points 1 and 2. But the con
77.
by frotaur 1y ago
It is actually inflation that precludes this from being true (along with relativity, of course). If there wasn't a period where spacetime inflated extremely rapidly, given two points, no matter how far apart, their causal pasts would e
78.
by frotaur 1y ago
There is a philosophical/probabilistic argument that explains why 'rocks are people too' holds less weight than 'pigs are people too'. We cannot be sure pigs feel anything or have qualia, but from comparison to huma
79.
by frotaur 1y ago
https://imgur.com/gallery/we-should-improve-society-meme-aQr...
80.
by frotaur 1y ago
Funny, when I was less than 12 years old, I remember often (usually when sick) seeing people shrink, feel like they appear to be very far away, and feeling like they were slowed down. It hasn't happened since then, I wonder if it'
81.
by frotaur 1y ago
This can generate new text. If the abilities generalise somewhat (and there is lots of evidence they DO generalise on some level), then there is no obstacle to generating new proofs, although the farther away they are from the training data
82.
by frotaur 1y ago
How do you know which question should be answered with 'I dont know?'. There are obvious questions which have no answer, but if only those are in the dataset, the model will answer I dont know only for unreasonable questions. To t
83.
Artificial Life, Lecture 1: Historical perspective(youtube.com)
2 points by frotaur 1y ago|0 comments
84.
by frotaur 1y ago
Yes, but that is not a good arguments. Example of unique living organisms include bacteria, tardigrades, or paramecia.
85.
by frotaur 1y ago
If I remember correctly, basically all 'closed' models don't output the raw chain of thought, but only a high-level summary, to avoid other companies using those to train/distill their own models. As far as I know deepse
86.
by frotaur 1y ago
Don't worry, it's being lobotimized by the 'unwoke' guardrails instead
87.
by frotaur 1y ago
I'm not sure about your example about talking to LLMs. There is good reason to think that speaking to it like a human might produce better results, as that's what most of the training data is composed of. I don't have any stu
88.
Show HN: AICurated – A repo curated by AI, with a twist(github.com)
4 points by frotaur 1y ago|0 comments
89.
by frotaur 1y ago
Anybody knows if there is a technical report for this, or for other models that generate images in a similar way? I'd really like to understand the architecture behind 4o-like image gen.
90.
by frotaur 1y ago
My only question about this is the concurrency : is it really easy to leverage it when you need to serve to clients without much latency ? I don't know much about this.

More ›