[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1748502769022215.png (205 KB, 960x461)
205 KB PNG
That's so freaky. How does it know who hiro is?
https://archive.4plebs.org/_/view_similar/vbYTTOJ6l5ehzFbebgt7kg/
>>
File: 1752091838233218.jpg (52 KB, 600x543)
52 KB JPG
>>108800651
>>
>>108800651
The same way the cameras on the streets let your government know who you are and where you are going? What are you confused about?
>>
>>108800651
4plebs has added AI features to the archive. If you hover over certain image thumbnails, it will give you a detailed LLM-generated description in the tooltip.
>>
>>108800797
clearly WRONG.
>>
File: file.png (2.04 MB, 1600x883)
2.04 MB PNG
>>108800651
he is extremely popular in japan. I was watching japanese youtube vids and I always see him in advertisements on the street and even game shows.
>>
File: 1778523456416.png (685 KB, 1080x2057)
685 KB PNG
>>108800651
>>108801023
>How does it know who hiro is?
It doesn't.

Like with most image search implemented by non-glowie projects, 4plebs is using embeddings to search images. This is not a new concept at all. OpenAI related their CLIP idea a few years ago. I experimented with it and it doesn't yield good results imo. A small dataset yields unrelated results. A large dataset yields... unrelated results. See attached pic.

Computing the embeddings from your image yielded better results from computing embeddings from the name "Hiroyuki Nishimura". This is not a surprise.

If you really want a surprise, just wait 2-3 more leaps equivalent to going from to Gemma 3 to Gemma-4, and your ass will be wholly owned by OSINT fags because it will take 0.05s to generate accurate image/text attributes per post, rather than 1s to generate semi-accurate attributes.

By properties, I'm talking,

>person in image, full name
>person in image, height
>race, skin tone, age, hair color

Same highly specific and discerning attributes for location.

All this for very little cost.

t. knower who has their own special image search that doesn't use embeddings.
>>
>>108801230
ah so its just going 'this dude looks kinda jap-y let's return the other jap-y images' and it happens to be hiro
>>
>>108801494
Yes

Search for Mitchell, and you'll probably get sports players, politicians, and celebrities that are white or black men

The future will be like,

Feed the image into an AI, ask it "what is this person's name (if no name, repond null)?". Store and query those results.

But now the user has to provide the correct spelling and even know what his full name is.
>>
>>108800651
i like 4plebs
>>
>>108801230
blah blah blah
I've been using the large CLIP to sort my 12GB furry porn database for a couple years, works wonders.
>>
>>108800651
4plebs has been shit ever since they removed the search function

fucking faggots
>>
>>108802280
thats good you have a solution that meets your standards anon-kun
>>
>>108802538
explain



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.