Lossfunk

A lab for time travellers.

We host and partner with deeply curious people shaping the future of their fields. Based in Bangalore, 🇮🇳 India.

Featured Questions

• What does it take to make a scientific model of consciousness or qualia?
• Why do the physical constants and laws in our universe appear fine-tuned for life?
• How do we confront metaphysical conundrums such as if AIs are conscious?
• What does the future of science and innovation look like?
• How do we create abundance for everyone?
• How do we retain human agency in a world where all economic work is automated?
• What new institutions should be created in a world swarming with agents?
• What new forms of human expression are possible now that weren't possible before?
• What would it take to accelerate India?

You can see our request for curiosity (frequently refreshed) for more concrete examples.

Join Lossfunk

• Sabbatical (for exceptional, experienced people): a 3-6 month fully funded sabbatical to work uninterruptedly on a risky yet important project
  • We host people and their projects that wouldn’t typically be supported by traditional institutions (academia, industry, VCs)
  • Typically meant for people who've had an impressive career but now seek a break to explore a burning question
  • Bangalore based + financial support included
• Fellowship (for early-career strongly-autonomous people): a 6 month paid fellowship based out of Bangalore to research, create and explore
  • We host young people who want to attack an ambitious, open problem and can work independently
  • Typically meant for early-career researchers and builders who already possess some work experience
  • Bangalore based + financial support included
• Residency (for builders, creators and founders): a 6-week cohort in Bangalore with 10-15 peers building the future
  • For people working in AI, science, deep tech and related fields
  • Bangalore based, part-time OK; program runs alongside main job / startup
  • We take no equity or fees

Latest Releases

• Representations • Denoising Models Develop Human-Like Perceptual Illusion Representations Across Architectures [Preprint]
• Representations • Geometry of Human Perceptual Domains Emerges Transiently in LLM Representations [⭐ ICML Mechanistic Interpretability Workshop, 2026]
• LLMs • Frontier Coding Agents Use Metaprogramming to Adapt to Unfamiliar Programming Languages [Preprint]
• World Models • Hybrid Neural World Models [Preprint]
• Philosophy • Metaphysical Concepts in AI Should Be Judged by Consequences [⭐ ICML 2026 - main]
• RL • Discovering Reinforcement Learning Interfaces with Large Language Models [⭐ Reinforcement Learning Conference (RLC) 2026 - main]
• Consciousness • AI Consciousness Requires Validated Models of Human Consciousness [AAAI 2026 Symposium on Machine Consciousness]
• LLMs • EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages [Logical Reasoning and ICBINB workshops at ICLR, 2026]
• AI for science • Why LLMs Aren't Scientists Yet: Lessons from Four Autonomous Research Attempts [Report]
• LLMs • Think Just Enough: Sequence-Level Entropy as a Confidence Signal for LLM Reasoning [Foundations of Reasoning Models Workshop at NeurIPS, 2025]
• LLMs • IPO: Your Language Model is Secretly a Preference Classifier [⭐ ACL 2025 - main]

Check out all our releases (papers, reports, code).

Attend Lossfunk Events

• Deeply technical talks by researchers, thinkers and experts (see past talks on youtube)
• Salons & Mixers to curate and bring together people who should know each other (see previous salons)
• Curated events with other orgs and people shaping the future of a field (e.g. conference for ai scientists)

Have an idea for a talk or an event? Pitch us: info@lossfunk.com

Want to attend an upcoming event (physical or online)? Join our WhatsApp announcement group (below), or follow us on luma.com/lossfunk.

🔔 To get notified about our upcoming events or releases, join our Whatsapp community (we post ~1x/week)

Sponsors

We’re a lab focused on benefitting the public at large by curating curious people and supporting their projects. To help with our mission, in the past or at present, we’re supported by following organizations.

AWS, OpenAI, Anthropic, Google, Modal, E2E / Jarvis and Sarvam <- thanks for helping Lossfunk!

We mention our sponsors in our releases, social media and events and they get recognized by deeply curious people.

In case you’d like to sponsor or partner, shoot us an email: info@lossfunk.com

Links

• Contact: info@lossfunk.com
• Blog: letters.lossfunk.com
• Github: github.com/lossfunk
• X: @lossfunk

About

Hi! I'm Paras, founder of Lossfunk where I'm constantly chasing the gradient of my curiosity. I read, write, research and build.

Previously, I founded Wingify, scaled it and then exited to a private equity group.

+[----->+++<]>++++.+++.+.++++++++.+[---->+<]>++.--[->++++<]>-.+[->+++<]>.-------.--[--->+<]>-.+[--->+<]>.-[->+++<]>+.+[---->+<]>+++.---[->++++<]>.--.+++.++[->+++<]>.--[--->+<]>-.[->+++<]>+.+++++++++++++.----------.-[--->+<]>-.+[->+++<]>+.+++++++++++.[--->+<]>-----.--[->++++<]>-.+[->+++<]>.-------.--[--->+<]>-.+[--->+<]>.-[->+++<]>+.+[---->+<]>+++.---[----->++<]>.---------.--.+.++++++++++++.

Stay curious :)

.

.

.

.

.

.

Still there? Here's a wall of testimonials