@RichardSocher Does https://t.co/Jgmx7PZMSL use the chats for training the models? If so, is it possible to opt-out? I’m not talking about the private mode (I’d still like to see the chat history), but a complete opt-out that you can do with OpenAI.
📢[Survey 📚] Object localization in images with zero manual annotation?🤩
➡️ We propose a survey discussing recent works exploiting ❄️self-supervised ViTs (incl. ICCV’23 & NeurIPS’23 works💫) w/ @EloiZablocki@SpyrosGidaris@gillespuy & P. Pérez.
📄https://t.co/Q7y8eNcQiM
Here is a review of unsupervised object localization methods, in the era of self-supervised ViTs.
We tried to make it concise but comprehensive.
Also check the companion GitHub awesome repo: https://t.co/QLdz7gp8wc
@MParakhin@teknium I have also noticed it's frustratingly worse lately. I barely use it anymore because of it, and I used to love it. Now going back to ChatGPT and Google 🤷
@omarsar0 It works pretty well with Bing if you open the PDF in Edge and use the Bing button from the toolbar to chat about the content of whatever you have opened. The formatting is much worse tough and the prompt might need tuning to get the level of details you want.
One effect of playing with AI for a bit is realizing how much of a dead-end Siri and other voice assistants feel like, now.
"I want to go to dinner and a movie tonight, and make sure I can get there by 6 and be home by 10. I would love some spicy Latin food and an action movie."
@rasbt@terrible_coder@tdomhan That's how they train it in the original "Attention is all you need" (for translation) and "Generating Wikipedia by Summarizing Long Sequences", which GPT cites as the architecture; GPT mentions masked self-attention heads as well
@yusuf_i_mehdi Both modes are much more fun than the previous mode/current balanced mode, which gives short answers and keeps asking follow-up questions that are usually not something that I would ask. The prompt limit feels less restrictive in the new modes because more information is given.
@yusuf_i_mehdi When the creative mode feels too casual (e.g. emojis when asked for scientific explanations) or I detect that it's hallucinating - I try the same prompt in the precise mode and it works just as advertised, less hallucination and precise answers.