849

Karma: 34

849 19 Sep 2026 10:13 UTC
2 points
0
in reply to: Thane Ruthenis’s comment on: ricraz’s Shortform
My personal experience is that people are ok with using free ai but it is seen as shameful to pay for it as that would directly finance the evil billionaires

849 19 Sep 2026 9:44 UTC
1 point
0
in reply to: Seth Herd’s comment on: So where’s this AI thing going?
I’m french so this advice may not apply to the us but:
As a tip if you ever want to commission art for the first time, you could reach out for an art/animation/concept art school and ask them to transfer the job offer to their students. When I was in art school in France we would get similar offers for 50-100€, which is much lower than what you would pay a professional. Of course the quality may be lower, but you can select the candidate that you like the most before asking them to do the work.

849 11 Sep 2026 7:38 UTC
5 points
2
on: So where’s this AI thing going?
Minor quibble, I am very biased as people around me are artists who can recognize ai images pretty easily: I would not have used an ai image to talk about this. Most people hate ai, and just by using this image the first impression you give is “someone pro ai”. A human drawing, (even very crude or in the xkcd style) makes you much more relatable in my opinion.

849 9 Sep 2026 9:39 UTC
1 point
0
on: 849′s Shortform
Naive question:
Lying as a human is hard, it seems we could make it much harder for AIs too. When I lie I have to think about my posture, my smile, my eyes, my voice, my tone...
I understand this would not work in the long term but if CoT monitoring is getting harder, why can’t we just give way more channels of output and thinking to AIs? Give them access to my webcam when they talk to me, then monitor what they look at. Allow them to choose a face, a voice, a prosody, then monitor that. Give them the ability to see when I’m typing then check when they look. All of these would be genuinely useful so the ai would use them, but then when they lie it means they have to control more channels at once.

849 2 Sep 2026 18:48 UTC
11 points
4
on: Don’t be the vitamin B guy
On the LLM part, I actually recommend you start by asking an A.I. to push back on your brilliant idea before you recruit 20 friends to test it

849 30 Jul 2026 15:39 UTC
2 points
0
on: Can an LLM make a feature-length movie on its own?
- When making a (normal) movie, usually having a short storyboard is a blessing because everything becomes longer in production. So maybe the SB from Fable wasn’t so bad.
- I really dislike the look of current video models, everything seems way too clean. This works for ads which is how they gain money, but not movies. The fact that incentives are not aligned to make realistic image/video generation for movies makes me think it will take longer than necessary to be achieved
- I am very curious about what it would look like if you asked the models to make animated vector art instead of video generation. If the vector art is simple enough, the models could actually understand what is happening in every scene they make, and presumably the result would be much better ( I’m thinking a movie in the style of Thomas Was Alone would be possible today, and enjoyable)

849 17 Jun 2026 21:33 UTC
1 point
0
in reply to: Bart’s comment on: Most people can’t juggle one ball
Imagine the trajectory of ball1 as an arc. If you throw ball2 under the arc, it’s inside. Over the arc is outside

849 17 Jun 2026 21:31 UTC
3 points
0
in reply to: Adele Lopez’s comment on: Most people can’t juggle one ball
I think it’s overcorrecting for the instinct of following the first ball you see with your eyes until you catch it (which would make you miss all the following balls)

849 13 Jun 2026 16:58 UTC
4 points
−6
in reply to: Daniel Kokotajlo’s comment on: Daniel Kokotajlo’s Shortform
I know this sounds like science fiction but are we that far away from some kind of humanoid robot attack from the us if the war lasts more than a year?

849 10 Jun 2026 14:29 UTC
1 point
0
in reply to: Abhishaike Mahajan’s comment on: How to build a cancer vaccine, and whether they will work this time
Thank you! That makes things more clear to me.
Thank you also in a more general way for putting out accessible articles on these topics, I really enjoy learning from them

849 9 Jun 2026 19:00 UTC
2 points
0
in reply to: Brinedew’s comment on: Mnemonic portraits for 19,023 human genes
Oh ok that makes sense! I should mention I know absolutely nothing about proteins, but since I use your extension it has become really fun to read about them. The timing with that post ( https://www.lesswrong.com/posts/nWwDciCqfodBifvEP/how-to-build-a-cancer-vaccine-and-whether-they-will-work ) is perfect

849 9 Jun 2026 16:41 UTC
1 point
0
in reply to: 849’s comment on: How to build a cancer vaccine, and whether they will work this time
Also, since all the peptides that we are showing them are produced by cancer cells, why is it so bad that the T-cells “pick a handful” of winners ? Since the winners will point them to cancerous cells anyway

849 9 Jun 2026 16:36 UTC
1 point
0
on: How to build a cancer vaccine, and whether they will work this time
Bu it is worth asking a question: why bother with the candidate list at all? Can’t we just be maximalist about it and stuff thousands of candidates into the vaccine? It only takes one (or maybe a few) to hit. It’s not like there are any downsides to being aggressive here.
Sadly, there is a downside to being aggressive.
Namely, a concept called “immunodominance”, which is the observation that when you present the immune system with a mixture of antigens, the resulting T-cell response tends to concentrate on one or a small handful of “winners,” with the remaining antigens getting ignored or generating responses so weak they might as well not be there.
I am confused: aren’t the T cells already exposed to all the neoantigens anyway ? Or we are wrapping the neoantigens into an adjuvant that will tell them they are dangerous, and immunodominance happens when the adjuvant wraps many peptides at once ? I guess this is it but since all my oncology knowledge is basically this post, I’m not sure

849 9 Jun 2026 9:21 UTC
3 points
2
on: Mnemonic portraits for 19,023 human genes
Image models consistently create images with high vibrance, so I think the vibrance parameter should be a filter on top of the output (vibrance filters, unlike saturation filters, should still make the image look “normal” even with an important offset). You can also color correct on top of that for the right hue. I did a before and after example for cathespin B by hand here (adjusting for low vibrance and orange hue): https://imgur.com/a/0BCEzxi
but all this could be done automatically

849 9 Jun 2026 8:49 UTC
5 points
0
in reply to: Brinedew’s comment on: Mnemonic portraits for 19,023 human genes
Isn’t it simpler to limit the bounds of lightness from 20% to 80% so you don’t have pure white or black and can still see color? I might be missing something

849 21 May 2026 7:17 UTC
9 points
3
on: Numb mental state shifts
- Doing something you have never done before seems very hard until you do it.
- When you have an idea and you think you will remember it, you probably won’t
- When you are tired, you naturally think you need to rest for hours when 20min is generally enough.
- When you are picking up a habit, not doing the thing once seems like no big deal, but it decreases a lot the probability you will pick it up again the next day. Not doing the thing two days in a row is basically the same as giving up but doesn’t feel like it.
- Spaced repetition (like anki) to learn something you don’t care about seems useless every time I do it, except it works.
- When people say nice things to you, you like them more even when you know they are motivated by something or you think you don’t care, especially after repetition. Something something Pavlov.
I’m not sure all of these exactly apply to what you describe, but this is what I got from the top of my head.

849 8 May 2026 20:25 UTC
1 point
0
on: Chess as a prediction model of the artificial intelligence impact on culture
Hello, this post was originally written in french, I did not check the translation very closely. I will fix any typos/weird things you point out! Also I’m very unsure about the part where I mention porn… I hope it’s at least food for thought quality.
Also the premise is that we don’t all die but the post is intended for more general audiences.

849 29 Apr 2026 17:45 UTC
6 points
0
on: Takes from two months as an aspiring LLM naturalist
One disanalogy I do see: humans sleep, and probably would for psychological reasons even if we didn’t need to physically; today’s LLMs don’t. I expect there’s more; maybe you can help me out in the comments?)
I would bet (a very small amount) that infinite context will be solved by introducing something analogous to sleeping for LLMs

849 21 Apr 2026 12:11 UTC
2 points
0
on: 10 non-boring ways I’ve used AI in the last month
I use llms in combination with anki to learn vietnamese, it’s so much better than any available online resources (this is also because there are relatively few english/french to vietnamese ressources) ! You can have conversations, learn if a word has more than one meaning or can be used in an expression, ask what is the opposite of a word, ask for sentences using a specific word or structure… Really recommend it.

849 6 Mar 2026 22:36 UTC
1 point
0
in reply to: Ninety-Three’s comment on: Rational Chess
In very important tournaments, games can be intense enough that if you play a very long game (up to 6 hours of focus) you will be tired/play more poorly the next day