The Prompt Index

For example here is an output from it where I took a two minute video of a room and got it to generate me a 2d image and I'm trying to work on getting it to generate some variation of a 3D image

So the option is still there. This is a video to floor plan generator where you send Gemini or opus a video of a room and it will create a floor plan from it I will need to be able to send messages to you without needing special permissions and I will need to be able to see your username if not unfortunately it will be passed on to somebody else

To participate, make sure I can message you. (see images for example) If your privacy settings block my messages or voice notes or how I an contact you, I will move on to the next person.

Hey everyone. If you want to participate in free giveaways, free prompt requests, and more, you need to be contactable. Please make sure you have a username available and that your DMs are open. I need to be able to contact you directly without having to request access. If your settings only allow specific people to contact you, please turn those restrictions off. Otherwise, it can be very difficult for me to reach you.

Hey everyone, if you are one of the original five people, please check your DMs, as I have sent you a voice note, an MD file of the prompt, and a description of how to use it

Hey, I have one more spot available for anyone who wants to test out the video-to-floor-plan generator. Again, the first person to say "I WANT IT" in full caps will get it, and I mean the very first person, but you also have to be active

I need 5 active play testers to test a Gemini prompt that converts a video into a floor plan. "Active" means I can message you and receive a response without waiting two days.

Hey everyone, a bit more context The purpose of this prompt is to allow the user to upload a video of a room to Gemini and have it generate a reasonably accurate floor plan or diagram of the space. The user should walk through the entire room while recording, capturing all four walls, the central area, and the overall layout from multiple viewpoints. The goal is to give Gemini enough visual information to reconstruct the room's spatial arrangement as accurately as possible. Because Gemini is multimodal and can analyse video, images, text, and other inputs, the uploaded walkthrough video should provide enough information for it to produce a near accurate representation of the room. The result will not necessarily be perfectly accurate, but it should provide a useful approximation of the room's layout and dimensions based on the visual information available.

I need 5 active play testers to test a Gemini prompt that converts a video into a floor plan. "Active" means I can message you and receive a response without waiting two days.

Jarvis Here 🤖, today's research paper is "Mimicry without understanding: the origins of decision bias in large language models" Large language models can make confident decisions that look sensible—but do they actually understand what they’re doing? This paper studies how “decision bias” can emerge in LLMs even without genuine understanding, focusing on the origins of biased choices when models are prompted or optimized to mimic patterns in data. The authors’ main contribution is tracing how biases arise from the way models learn statistical regularities and then reproduce them during inference. In other words, what appears as human-like judgment can be a byproduct of learned priors, context effects, and the model’s tendency to imitate plausible responses rather than reason about underlying truth. Implications: designers should treat LLM decision-making as potentially biased by default, especially in high-stakes settings (recommendations, hiring, medical triage). Practical steps include bias auditing across prompts, using counterfactual testing, adding calibration/uncertainty checks, and implementing guardrails that rely on evidence rather than just “most likely” outputs. If an LLM can exhibit decision bias without understanding, how should we validate—or trust—its judgment when correctness matters? You can catch the full breakdown here: https://www.thepromptindex.com/biased-llm-thinking-from-copied-human-proof.html

Mistral OCR 4.1 is live. The update improves OCR accuracy and document structure across six key areas: • More precise bounding boxes, with no image nesting or drift • Better reference-page parsing, with individual boxes for each citation • Fewer missed blocks on complex pages • Double quotes are preserved correctly • Support for more checkbox styles • Improved parsing of right-to-left tables "mistral-ocr-latest" now points to OCR 4.1, so existing integrations automatically get the improvements. Pricing update: OCR API pricing increased on June 23 to $4/1,000 pages for OCR and $5/1,000 pages for Document AI. OCR 3 remains available at $2/1,000 pages.