Vyom

Al has gotten really good at understanding what we say, but not how we say it. a transcript can't tell the difference between a genuine "i'm fine" and a frustrated one. @hume_ai is tackling exactly that with an audio-native model that understands tone, emotion and pauses. this

we gave AI agents their own computers and immediately made them come into the office. there’s even a ping pong table. @getskydive lets you build a team of agents, give them work over Slack or iMessage, and see what they’re doing in this 3D office. i would absolutely spend too

the meeting starts in 4 minutes and your context is spread across 3 apps. good luck. in this demo, one text to Robyn pulls Gmail, calendar and Granola notes into a briefing: 3 meetings, 8 action points, and 2 things due before 10am. that last part is what sold me on the idea. t

imagine giving a new hire your login and saying "just don't open anything sensitive." that's the setup when an Al agent inherits all your access. @mintive gives each agent its own identity, scoped permissions and a record of every call. access can change depending on who's aski

that's the Al demo i want to see. @snaptrude's agent lets the architect decide where to cut, then updates the layout, coordinates the structure, plumbing and wiring, and redoes the drawings. everything stays editable. getting a first design is one thing. being able to change yo

“what are you looking for?” if i knew that, i wouldn’t have 400 saved outfit posts. scene by @daletlabs generates a shopping feed around your taste, with real products inside the images. see an outfit you like, tap the pieces, buy across brands in one checkout. the bit in this

i want my laptop to understand what i’m looking at without uploading my entire life to a server. Liquid AI’s LFM3 Omni takes video, audio and text, then talks back. 2.2B parameters, built to run on device. and you can fine-tune it on your own data with one GPU. that’s the part

the wildest part is buried in the thread. Axiom says its AI found counterexamples to a claim inside a 15,000 page proof, then proved a replacement result to fix that step. that’s what makes AxiomProver interesting to me. finding where an argument breaks and doing the maths to r