19 earthquakes before 7:15am today in San Ramon. Our community WhatsApp was blowing up.
So I dug into a peer-reviewed study of the 2015 swarm and compared it to what's happening now:
#SanRamon#earthquake#BayArea#Dublin#TriValley
Sent @Muse a YouTube link to my daughter's 10U softball game live-stream. It watched the game, followed her jersey number and provided very good insights Didn't know I needed this :)
Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter. This is the biggest jump we've made so far on coding and agentic work. Try it in Muse Code and our API.
Next up 🍉 and Muse Spark open weights releases coming soon.
(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
Grew up watching Sachin make batting look like poetry. Now we watch Bumrah make bowling look like dark arts. Indian cricket fans are truly blessed to witness legends in different eras! 🏏
Not to take anything away from Samson and the batting heroes — but this World Cup win belongs to Bumrah. He was the difference. 🇮🇳🐐
Congratulations Team India!!
This is genius. Been doing this manually on an M4 Pro Mac Mini with LoRA on MLX . ~50 experiment iterations over days that
autoresearch could've done overnight.
Painful lessons an agent would've found in hours:
- LoRA scale matters more than rank (20→2 killed accuracy despite 4x rank)
- grad_accumulation silently 4x'd effective epochs — overfitting before first eval
- "obvious wins" (mask_prompt, balanced data) both made things worse
The 5-min fixed budget is the key insight instead of babysitting 45-min eval cycles. Excited
to try this with MLX on Apple Silicon.
New in Cowork: scheduled tasks.
Claude can now complete recurring tasks at specific times automatically: a morning brief, weekly spreadsheet updates, Friday team presentations.
.@bcherny love what y'all are building with Claude Code! Small feature wish: would be amazing to schedule background tasks on a timer while CC is running. I built a agent that syncs team chats, code reviews, and project trackers into
persistent memory — works great but sync is manual. A simple every 30min: run https://t.co/g7lbjBA3o0 would make it a true always-on second brain. Background commands exist today but can't be scheduled. Would be a game changer! cc @AnthropicAI
@anshnanda Use fusion of keyword/grep search and embedding bases semantic match. I use openai text embedding model to generate store embedding local db.
@rmasand Same. And at least in 2015 it stopped after a month. We're approaching month 3 and this morning was the worst yet — 19 quakes before most people's alarms went off. On the bright side, every past swarm here eventually quieted down. Hoping this one will too.
19 earthquakes before 7:15am today in San Ramon. Our community WhatsApp was blowing up.
So I dug into a peer-reviewed study of the 2015 swarm and compared it to what's happening now:
#SanRamon#earthquake#BayArea#Dublin#TriValley
Full paper (free): https://t.co/ecspxXa6t9
Key takeaway: These swarms are caused by fluids pushing through small cracked rock — a completely different process from what causes major earthquakes. 10+ swarms since 1970, none led to a big one.
I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!
"Can you find the IP for my Tesla Wall Connector?"
Claude Code checked the ARP cache, looped through IPs, hit /api/1/vitals on each one, and found it at 192.168.x.x
No nmap installed. No router access. Just clever API probing.
This thing is wild!
@bcherny@AnthropicAI
🎯 India looks strong with 60% win probability. Watch out for Virat Kohli vs Kyle Jamieson battle! 🏏🔥
#INDvNZ#ind#nz#Cricket#CricChanakya
Check detailed predictions at https://t.co/VSaBjZwo08