Anmelden

Frühere Suchanfragen

Keine früheren Suchanfragen

Suchoptionen

Only available when logged in.
ard.social ist Teil eines dezentralisierten sozialen Netzwerks, angetrieben von Mastodon.
Die offizielle Mastodon-Instanz der Arbeitsgemeinschaft der öffentlich-rechtlichen Rundfunkanstalten der Bundesrepublik Deutschland (ARD). Bestimmt für die redaktionellen Angebote der ARD.

Verwaltet von:

Serverstatistik:

15
aktive Profile
Mehr erfahren

ard.social: ÜberProfilverzeichnisDatenschutzerklärung

Mastodon: ÜberApp herunterladenTastenkombinationenQuellcode anzeigenv4.2.18

#chatgpt

4 Beiträge4 Beteiligte0 Beiträge heute
Kerfuffle

I'm not someone to point fingers and blame developers for mistakes they made, but if I run into incomprehensible code and the author says "I dunno what the code does; I just asked #chatGPT to generats it", I'm gonna be pointing lots of fingers. Six on each hand at least. #ai #dev

Raphael Wimmer

Did you know that #ChatGPT 's new image generator fakes rendering an image line by line? In reality, the browser downloads partly finished images and then simulates the line-by-line rendering via JS.
Makes sense from a UX perspective but I don't like it.

Practical use case: even if ChatGPT tells you that it cannot render an image due to content policies, sometimes a part of the image has already been sent to the browser. This might make reverse-engineering the content policies a bit easier.

Screenshot of a browser window showing a ChatGPT session with the prompt "generate an image of a female professor at the university of regensburg". Below the prompt, a completely blurred image is shown (which later slowly gets sharper line by line). However, the developer console on the right side of the screenshot shows that the browser already downloaded a version of the image where the top third of the image is already sharp and only the lower two thirds are blurred.
ALT
Screenshot of ChatGPT session. On the left, the prompt says: "generate an image of an adventurous archeologist with a hat and a whip" (intentional typo). After a while, ChatGPT responds with "I wasn’t able to generate that image because the request violates our content policies. If you’d like to try something different—maybe a variation on the character or a new setting—I’d be happy to help. Just let me know your new prompt!". However, the developer console on the right shows that the browser already received a partially rendered image showing someone slightly resembling Indiana Jones. The facial features are different, however.
ALT
c't Magazin

c't-Webinar: KI-Sprachmodelle effizient nutzen

Das Webinar gibt praxisnahe Tipps und Hinweise zu Sprach-KI im Berufsalltag. Es beleuchtet technische Hintergründe, Kostenstrukturen und rechtliche Neuerungen.

heise.de/news/c-t-Webinar-KI-S

heise online · c't-Webinar: KI-Sprachmodelle effizient nutzenVon Anke Brandt
#ChatGPT#IT#KünstlicheIntelligenz
EntdeckenLive-Feeds
Über