inacionline's Avatar

inacionline

@inacionline.bsky.social

informatics engineer; explainability and meta-learning research;

81 Followers  |  166 Following  |  90 Posts  |  Joined: 07.07.2023  |  2.2432

Latest posts by inacionline.bsky.social on Bluesky

the damage llms made to the socially acceptable syntactic landscape have been disastrous to the online writing community

30.05.2025 21:26 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

landlord is the kind of job where the line between a vulture and a stork is too far to discern

30.05.2025 21:21 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

one thing i’ve recurrently noticed with software is that it makes life so easy, that it brings sort of a entitlement feeling in users, whereas something that breaks society today, will be deemed useless if it stops being society-breaking. the weird part is how quickly it happened for llms

30.05.2025 20:28 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

people were paying 20 a month for gpt 4, which seems on par with people paying for dsl 20 years ago, even if it was just 2 years ago! who knows where we’ll go

30.05.2025 20:14 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

add a friend-tagging system to the reminders and then we are talking

24.05.2025 17:36 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

isn't that just the reminders app though?

24.05.2025 17:35 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

more papers should be like this

24.05.2025 17:16 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

when the the abridged ends are finally justified by the substantially broader means?

24.05.2025 07:58 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

funny how interaction networks seem to re-grow node by node even in different digital plains

24.05.2025 07:55 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

i would attribute this to the β€œforum format -> endless algorithmic scroll” and the imposition of character limits. Since you only have [The one board], the prospect of getting lost on it, or worse, to β€œlose” to someone on it will squeeze the best predicates out of you

24.05.2025 07:52 β€” πŸ‘ 8    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

to watch the algorithmic threads unfold while you are weaving them

23.05.2025 22:33 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

instagram-filter-based chronological identifier

04.03.2025 22:50 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Figure 1: How the path kernel measures similarity between examples. In this two-dimensional illustration, as the weights follow a path on the plane during training, the model’s gradients (vectors on the weight plane) for x, x1 and x2 vary along it. The kernel K(x, x1) is the integral of the dot product of the gradients βˆ‡wy(x) and βˆ‡wy(x1) over the path, and similarly for K(x, x2). Because on average over the weight path βˆ‡wy(x) Β· βˆ‡wy(x1) is greater than βˆ‡wy(x) Β· βˆ‡wy(x2), y1 has more influence than y2 in predicting y, all else being equal.

Figure 1: How the path kernel measures similarity between examples. In this two-dimensional illustration, as the weights follow a path on the plane during training, the model’s gradients (vectors on the weight plane) for x, x1 and x2 vary along it. The kernel K(x, x1) is the integral of the dot product of the gradients βˆ‡wy(x) and βˆ‡wy(x1) over the path, and similarly for K(x, x2). Because on average over the weight path βˆ‡wy(x) Β· βˆ‡wy(x1) is greater than βˆ‡wy(x) Β· βˆ‡wy(x2), y1 has more influence than y2 in predicting y, all else being equal.

i mean, look

04.03.2025 21:50 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

if it's based, it's no longer in the cloud

04.03.2025 21:32 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

as big as it needed to be (maybe too big at times)

04.03.2025 21:31 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

from the folder: "papers that make something click"

arxiv.org/pdf/2012.00152

04.03.2025 21:29 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

there is a specific tongue used exclusively by reply/spambots that represents something profound about the way we connect and present to others that still needs to be looked into

03.03.2025 22:00 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

agreed, past-induced cringe is evolutionary, we should embrace it more

02.03.2025 22:39 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

nice try

01.03.2025 23:07 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I always found it funny that the way we train AI models, by trying to render them as errors-free as possible, to allow them to more reliably replace humans at some tasks, is so contrastive to our clumsy nature, that it won't necessarily make them "perfect replacements"

01.03.2025 23:05 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

this bag would be a good thread tbh

01.03.2025 01:18 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

yeah sorry i just chimed in, but yes, it ringed a bell to this post, basically, notch picked the parts people liked about infiniminer, and made what-would-be minecraft. my first pick from β€œthe-inspired-prevails” bag

01.03.2025 01:17 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0
Preview
Infiniminer – Minecraft Wiki Infiniminer is an open source multiplayer block-based sandbox building and digging game, inspired by the games Infinifrag, Team Fortress 2, and Motherload,[1] in which the player plays as a mi...

minecraft.wiki/w/Infiniminer

28.02.2025 23:43 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

another one to the folder of β€œthe-inspired-prevails” files

28.02.2025 23:40 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Feel-your-way-through-UIs

01.02.2025 20:59 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Haptics should not only be more ubiquitous, but should be shaped by the magnitude of the action. Like a post? small vibration. Delete a database? Now we’re talking

01.02.2025 20:57 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

β€œstruggle-based optimization” also has a nice ring

22.01.2025 13:23 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

*all* art is basically adding and subtracting latent vectors to reach a desired location of semantic meaning in the end

28.12.2024 16:23 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

β€œvisible mending” but in the software sense β€” the patches are the solutions anyways

25.12.2024 21:23 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

slightly related, but if I am following it right, I think you are describing scenarios where people defend their thesis, and at the same time, demonstrate how even if adversarial ones were viable, they'd be obviously less favourable. In a sense that, if one cross the line, they're "far-too-gone"

21.12.2024 23:52 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

@inacionline is following 19 prominent accounts