@eltc9584 Same playbook as GPT-6's Intelligent UI. The chat itself becoming the tool is the whole game now. First thing I'd have it build: a live P&L that updates as I talk. What's the first tool you'd build?
@larsnow 16.9 MB changes the math. No API bills, no latency, no privacy tradeoff. I'd build a meeting copilot that runs fully on device and never uploads a byte. What would you build with it?
@DataDiscovered Demos are calculators. The unlock is tools that keep living after the chat ends. First thing I'd build: a live budget tracker that updates as I talk. What's the first tool you'd have it make?
Haiku 5.5 costs $0.10. Subagent approved.
Numbers: $0.10/$0.50, 100K-token cliff at 5x.
39.2% Terminal-Bench vs 70.6% Sonnet.
1M context, 128K output.
Cheap only below 100K tokens.
Which agent would you run on it daily?
@AbhiCodes15 good question honestly. most vibe coded stuff i've seen falls over at real traffic. when yours slowed down, was it the db queries or the frontend bundle that killed it first?
@Karthikvarmamkv@cactuscompute 16.9mb doing whisper's job at 11ms is the kind of thing that quietly kills whole api businesses. first thing i'd build is a meeting bot that transcribes everything locally and never uploads a second. what would you build if transcription was basically free?
@driesdegelder being the data the widget is built on, that's the real shift here. citations are dead, data is the new game. first tool i'd have chatgpt build inside a chat is a mortgage calculator with live rates. what's the first one you'd ask for?
@avynsrc@NFT_Chen so codex is just quietly spending 10x less on thinking and nobody noticed. that flips the whole cost debate honestly. terminal-bench 4.0 has claude code at $14.30 a task vs $1.92 for codex + gpt-6.1 sol. at that gap which one are you actually running daily?
@thomas_rehmer one injection taking over every agent in the region is insane. this is exactly the stuff that keeps companies from shipping agents at all. are you running your agents network-isolated now or is that still on the todo list?
Founders film. They do not edit.
Kinetic captions that move with your words.
CaptionsEasy is free, forever.
Film in the morning, post by lunch.
One command on your own machine.
What would you post if captions took zero time?
@aiartgallerie 53gb at 3-bit is the headline, 75.5gb peak at full context is the fine print. so basically nobody's laptop. would you actually run a 137b locally or is this just microsoft dodging cloud token bills?
@mathiasvrb bairesdev survey had only 29% of devs trusting ai output and 78% of ctos raising review spend. how much of the diff do you actually read line by line?
@eltc9584 jetbrains survey of 15k devs had claude code at 39% vs copilot at 21%. auto routing could close that gap if it actually works. you still on copilot?
Applying to Anthropic's startup program? You agree to all of this just by applying:
- They can build competing tech from what they learn looking at your data
- You're banned from building anything competitive on their API
- They can rewrite the terms anytime without telling you
Read the fine print before you sign. Would you still apply? (screenshot via @rstureborg)