Akshit jindal

[🔁 Please repost!] Help us improve emotion tracking tools! Take a 15–20 min interactive survey testing different ways to log emotions. 📍 For those 18+ & living in India 💬 No experience needed! 👉 alchemy18.github.io/EmotionDairy-U… #UXResearch #MentalWellness #MentalHealthAdvocate

It's no secret. Quite visible actually.

It's worth remembering, US bombings are lower than they used to be. I doubt AI has affected this trend – and it's too early to tell what will happen. But we have now seen two actual cases this year (Palm Springs IVF + Las Vegas cyber-truck). This threat is no longer theoretical.

FBI says Palm Springs bombing suspects used AI chat program to help plan attack cnbc.com/2025/06/04/fbi…

Basically AI says NTA most of the time. That's why narcissists love it. Who doesn't like being surrounded by yes men? People who actually use their brains, that's who.

You can make LLMs more creative by training them on human "creativity signals" (novelty, diversity, surprise, quality). Result: Even small models score higher on all 4 creativity dimensions simultaneously. Looks like we can optimize AI for creativity just like any other metric

So a stupid person who wants to blow up his friend won't be able to, but an intelligent terrorist can easily get detailed instructions for the same. This just shows that safety is merely an afterthought for the big companies. All they care about is money.

Kudos to the @SarvamAI team for leading the charge in sovereign AI development and advancing Indic language technology. Leaving the naysayers aside, this is a great step forward in growing India's AI capabilities. We need more people to work in a low resource and inclusive way.

Today is a big day for AI Safety. We released Claude Opus 4 under the ASL-3 deployment standard Here's what that means:

Going through any system prompt, be it Claude, Gemini, ChatGPT, or any other, makes the system look like a rule-based system rather than an intelligent one. If so many instructions have to be hard coded basically, where is the intelligence?

That's the catch, it's not good at all if you know how to code.

I don't think the NeurIPS submission ID directly correlates to the number of submissions. But maybe that's just me being optimistic 🙃 @NeurIPSConf #NeurIPS

Reviewer calling themselves an expert in the field when they have so obviously misunderstood the concept is hilarious to me. I'm being asked to evaluate things that either do not exist, or are present in the supplementary material already. Unexpected from @ICCVConference #ICCV

I heard something along the lines of "You need to know what's wrong in order to do no wrong". Guess this quantifies it. Teach a model about toxicity instead of removing it completely from the training data.

Wonder what evals the @OpenAI team uses.🤔

Everyone's rushing to implement MCP but nobody's talking about the easiest way to do it. import gradio as gr demo = gr.Interface(fn=your_function, ...) demo.launch(mcp_server=True) # ← that's it 🤯 Your Gradio app is now an MCP server for any LLM. Learn how:👇