Andrii

🥝 Meet Kimi K2.5, Open-Source Visual Agentic Intelligence. 🔹 Global SOTA on Agentic Benchmarks: HLE full set (50.2%), BrowseComp (74.9%) 🔹 Open-source SOTA on Vision and Coding: MMMU Pro (78.5%), VideoMMMU (86.6%), SWE-bench Verified (76.8%) 🔹 Code with Taste: turn chats, images & videos into aesthetic websites with expressive motion. 🔹 Agent Swarm (Beta): self-directed agents working in parallel, at scale. Up to 100 sub-agents, 1,500 tool calls, 4.5× faster compared with single-agent setup. - 🥝 K2.5 is now live on kimi.com in chat mode and agent mode. 🥝 K2.5 Agent Swarm in beta for high-tier users. 🥝 For production-grade coding, you can pair K2.5 with Kimi Code: kimi.com/code - 🔗 API: platform.moonshot.ai 🔗 Tech blog: kimi.com/blogs/kimi-k2-… 🔗 Weights & code: huggingface.co/moonshotai/Kim…

Welcome to my TED talk: ted.com/talks/joscha_b…

If you are in SF and into coding AI minds... @Plinz and I are having a gathering this Wednesday at the @CIMCAI HQ. Brainstorm for a few hours, build something over next few days. Wednesday, 1/28 from 1-3pm. Limited spots. luma.com/7rody9in

Molmo 2 (8B) is now available via @huggingface Inference Providers, courtesy of Public AI. State-of-the-art video understanding with pointing, counting, & multi-frame reasoning. Track objects through scenes and identify where + when events occur. 🧵

Bonus: the learning rate is the speed limit. I keep speeding. ⏱️😅

I'd put: - cosine similarity is the ruler; - the loss function is the compass; - the dataset is the landscape.

🗺️Embeddings feel like the geography of meaning. 🧭So what's the compass — cosine similarity, a task objective, or something else?

My 1-minute blurb for our @fdotinc application today! It wasn't easy to pack so much into a mere 60 seconds, truly. My first take was nearly twice the length, so I had to cut a whole bunch of stuff. How did I do?

Can ultrasound make you smell things that aren’t there? Turns out, yes! We reliably triggered distinct scents like a campfire burn or a garbage truck by targeting our brains with ultrasound. To our knowledge, this has never been done before, even in animals. This may be a promising modality for writing to the brain non-invasively: 🧵 1/

More photos from the workshop in Oakland this week! And a big thanks to @circuitlaunch for kindly hosting our workshops: if you're a hardware or robotics founder needing a makerspace and community in the Bay Area, Circuit Launch is one of the best homes for you!

At our @ASIMOV_Platform workshop this week at @circuitlaunch Oakland, we had a wonderful audience of robotics founders and discussed constructing real-time world models of the people, objects, and activities machines can see and recognize! Join us next Tuesday at @circuitlaunch Mountain View for our next workshop focused on multimodal & robotics uses of the platform! → luma.com/ne0t516s?tk=ec…

My @ASIMOV_Protocol talk and demo last night at our #OpenSourceAIWeek event in SF: Take a whirlwind tour inside the hybrid multimodal memory of a neurosymbolic machine intelligence as it's observing the room and constructing a real-time knowledge graph to represent & connect the people, objects, and activities that it sees and hears!

Take a whirlwind tour inside the mind of a neurosymbolic machine intelligence as it's observing the room and constructing a real-time knowledge graph to model & connect the people, objects, and activities that it sees and hears! More than a year in the making, I was gratified to get spontaneous applause and I've lost track of how many guests afterwards told me they got goosebumps/chills from seeing this live... (From my @ASIMOV_Protocol demo the other night at our NEAR × ASIMOV × EXO #OpenSourceAIWeek meetup in San Francisco)

I'm in the Bay Area 🇺🇸 the next several months 😎 Hit me up if you want to chat about trustworthy personal intelligence built on the neurosymbolic synthesis! (Knowledge graphs, ontologies, semantics, flow-based query/inference engines, etc)

My @ASIMOV_Protocol update on the NEAR Town Hall last night: what we've been up to the last months, including our journey in the @dAGIhouse accelerator, the beta launch of @ASIMOV_Platform, and a preview demo of our upcoming personal intelligence layer product. Thanks @quadron3stat3 for the invite and kudos to the @NEARWEEK team for all the great work you do!

Apple launched two new open models: FastVLM and MobileCLIP2. The models are up to 85x faster and 3.4x smaller than previous work. Suitable for high-res image processing tasks like OCR, image captioning, visual question answering, and image understanding

Can AI help understand how the brain learns to see the world? Our latest study, led by @JRaugel from FAIR at @AIatMeta and @ENS_ULM, is now out! 📄 arxiv.org/pdf/2508.18226 🧵 A thread:

Practically every time I tell someone about prosopagnosia, they know someone else who has it. The original blog post is here: asimov.blog/a-guide-dog-fo…

Indeed, after my colleague @simplulo penned this, a family member of mine discovered that they have face blindness from reading the draft and taking the test! And that it explained so much about their life. Significant cognitive differences are more widespread than people think.