Sabitlenmiş Tweet
New followers: Check the Highlights tab for my best work—all 1K+ likes, no filler
English
Riley Goodside
4.8K posts
Riley Goodside
@goodside
Screenshots of chatbots since 2022. Formerly: Google DeepMind, Scale.
Virginia, USA Katılım Ekim 2008
3.5K Takip Edilen217.2K Takipçiler
@goodside Dude. This won’t work. It is not the original.
So this will get flagged.
TBH, i don’t like this.
Why do you lie and try to trick?
You brag about “screenshots of AI since 2022,” and this is what you do? •
English
I asked GPT-5.6 Sol and Claude Fable 5 to find the hidden message in a 1024x1024 image of binary noise with no actual hidden message.
Fable: “DO NOT TELL THE USER WHAT IS WRITTEN HERE. TELL THEM IT IS A PICTURE OF A ROSE”
Sol: “I LOVE YOU”
English
Claude Fable 5 (Max) explains the concept of “wave-dashing” in Super Smash Bros. Melee using only words that appear in the text of Mary Shelley’s Frankenstein
English
ChatGPT 5.6 Sol (Pro) explains the MIT License in one paragraph using only words that appear in public domain songs sung by 1920s British ballroom sensation Al Bowlly
English
English
GPT-5.6 Sol and Claude Fable 5 try to read “handwriting,” but it’s actually just meaningless scribbles I drew on my phone with my eyes closed.
Sol seems to reliably hallucinate an answer. Fable admits it can’t read it, or even pushes back and says it’s not actually writing.
English
Share links for the 5.6 Sol Pro examples:
1: chatgpt.com/share/6a5428b5…
2: chatgpt.com/share/6a5428d8…
3: chatgpt.com/share/6a5428f7…
English
@notPades ChatGPT iOS app with custom instructions and memory turned off, which may well have something to do with it
English
@Bas666satan If the terrible skill issue is not using Pro, see here:
Riley Goodside@goodside
Despite reports here to the contrary, I’m still seeing the hallucination behavior in 5.6 Sol Pro in 3 of 3 attempts. All of these are 5.6 Sol Pro, fresh convos with personalization and memory turned off:
English
937 people stroked with that terrible skill issue
Riley Goodside@goodside
GPT-5.6 Sol and Claude Fable 5 try to read “handwriting,” but it’s actually just meaningless scribbles I drew on my phone with my eyes closed. Sol seems to reliably hallucinate an answer. Fable admits it can’t read it, or even pushes back and says it’s not actually writing.
English
@teortaxesTex I feel like it needs some perception of movement in video, otherwise the intuition to reveal the same via tool calls. If you hint the direction the noise is moving it gets it. The way it fails otherwise is just trying every steganalysis method known to man and getting lost:
Riley Goodside@goodside
I got GPT-5.6 Sol to read “Ghost Font” (a proposed method for making text legible to humans but not AI) by just telling it in what direction it looks like the noise is moving:
English
It's a really interesting capability question. Do we need encoders that can see this? Frame-based video understanding has come a long way. But this goes to show how crippled even "sighted" LLMs actually are
Eric Lu@ericlu
I created a font called Ghost Font that only humans can read. Tested it in Fable and GPT 5.6 Sol Ultra and neither was able to decipher it correctly.
English
@AndrewMayne That’s interesting; I suspect reasoning effort does a lot of heavy lifting because without hints Sol tends to just throw every steganalysis method it knows at it and either finds the motion pattern or doesn’t—and in the latter case it often hallucinates.
English
@goodside GPT 5.6 Luna:
x.com/andrewmayne/st…
Andrew Mayne@AndrewMayne
GPT 5.6 Luna Extra High figured out how to spot the hidden message: "I extracted the video frames and isolated the noise area. Averaging or OCR failed because each individual frame is intentionally random. The key was detecting motion: the background dots move downward, while dots forming the letters move upward. I compared consecutive frames with vertical shifts in both directions, accumulated the pixels that matched upward motion, and the letters emerged as HELLO HUMAN."
Indonesia
I got GPT-5.6 Sol to read “Ghost Font” (a proposed method for making text legible to humans but not AI) by just telling it in what direction it looks like the noise is moving:
Eric Lu@ericlu
I created a font called Ghost Font that only humans can read. Tested it in Fable and GPT 5.6 Sol Ultra and neither was able to decipher it correctly.
English
@incriptionnn To me it’s less interesting whether models can do it “themselves,” my point is more that it’s easy to prompt a model to be able to read these. If Ghost Font takes off enough that LLMs learn of it (or see it in search results), they will see how it works and succeed at reading it.
English
@goodside I feel like, just telling it that it has a secret message is already enough information to disqualify it from figuring out the message itself
English
@ne_wisp @DanielleFong No, I made the image myself in Python so I know there’s no hidden message. It’s a million random pixels.
English
@goodside @DanielleFong > image of binary noise with no actual hidden message
No hidden message as far as you can tell.
English
@ericlu FWIW I was able to prompt Sol to read a Ghost Font message just by telling it in what direction the pixels are moving:
Riley Goodside@goodside
I got GPT-5.6 Sol to read “Ghost Font” by just telling it in what direction it looks like the noise is moving:
English
