Prompt: "Keeping the glass and the hand behind the glass the same, please change only the three brown candies in the glass into green, yellow, red, and orange candies. Make no other changes. Change the reflection to remove the brown candy too." Seed was 1070229954903864, but your setup is probably too different for that to help.
It seems like Gemini 2.5 Flash was the only model that successfully removed the reflections...it should get some points for that!
I understood that much, at least from the description you added on the Kontext result. I agree that you should provide more information here, though, especially around "we adapt the exact prompts depending on the model", since your strategy here could also reflect model strengths and weaknesses.
I have a 4080 RTX and Kontext runs great at fp8. I run several other models besides. If you want to get at all good at this, you need tons of throwaway generations and fast iteration and an API quickly becomes pricier than a GPU.
Good effort, somewhat marred by poor prompting. Passing in “the tower in the image is leaning to the right,” for example, is a big mistake. That context is already in the image, and passing that as a prompt will only make the model apt to lean the tower in the result.
I quit Google last year because I was just done with the incessant push for "AI" in everything (AI exclusively means LLMs of course). I still believe in the company as a whole, the work culture just took a hard right towards kafkaville. Nowadays when my relatives say "AI will replace X" or whatever I just nod along. People are incredibly naive and unbelievably ignorant, but that's about as new as eating wheat.
Did you read the whole thread and all of your own comment each time you had to type another half-word? If not, I’m afraid your first statement doesn’t hold.
All models are wrong, but some are useful. However when it comes to cognition and intelligence we seem to be in the “wrong and useless” era or maybe even “wrong and harmful” (history seems to suggest this as a necessary milestone…anyone remember “humorism”?)
I think there is a kernel of truth in what you said but your language is a bit of an overreach. No athlete who trains only when told to train is making it to the Olympics.
The police does not like being screwed with either. These aren’t good things. People with significant authority perform a duty and ought to act independently of their personal feeling.
I wanted to look up Japanese vocab easily with my voice while running. Wouldn’t let me do it (it could show me dictionary pages but wouldn’t speak the translation into my AirPods). However, I could look up English words just fine.
So I had to set my Siri language to Japanese, and now I can look up English translations of Japanese words…though I do have to speak Japanese.
If MacOS, an OS with posix style permissions, app level permissions, and folder access limits per app does not have a “granular permissions model”, which OS does? What are you trying to say?
Given that Wanamaker died in 1922 it’s safe to say his quote was in the context of a different problem entirely (which still exists, on top of the bot issue). Maybe it’s time to update to “three fourths”?
I had the same feeling, but also the feeling that it was written for AI, as in marketing. That’s probably not the case, but it looks suspicious because this person only found this issue using AI and would’ve otherwise missed it, and then made a blog post saying so (which arguably makes one look incompetent, whether that’s justifiable or not, and makes AI look like the hero).
Neither, you’re reading it wrong. Think of it as codebases getting more reliable over time as they accumulate fixes and tests. (As opposed to, say, writing code in NodeJS versus C++)
