Lowd Code

I build software, agentic AI automation, and ML for my clients.
Let's build together → |

[Lowd Code @LowdCode]

Cloudflare now answers a page 10 milliseconds faster at the median. The speedup showed up on the live network, not a lab run.

[Lowd Code @LowdCode]

Source: https://t.co/R0rN3N9smC

[Lowd Code @LowdCode]

Databricks starts tens of millions of machines a day. Each one used to wait about five seconds for its network rules. That wait is now 125 milliseconds.

[Lowd Code @LowdCode]

Source: https://t.co/XkXv4kABGk

[Lowd Code @LowdCode]

English Wikipedia now holds more than seven million articles. Anyone can open one. The catalog stays public.

[Lowd Code @LowdCode]

Source: https://t.co/gwQkk1jSkk

[Lowd Code @LowdCode]

NASA's Hubble has taken over 1.7 million looks at the sky. The pictures live in a public archive. Anyone can open one.

[Lowd Code @LowdCode]

Source: https://t.co/rY7u2FeHUO

[Lowd Code @LowdCode]

Roblox's cache now answers 1.38 billion queries a second at peak. It sits in front of live game storage so play does not stall.

[Lowd Code @LowdCode]

Source: https://t.co/OZPhyhKJdE

[Lowd Code @LowdCode]

Intercom's support agent now fully resolves over two million conversations a week. That is live customer support at a help desk company, not a demo.

[Lowd Code @LowdCode]

Source: https://t.co/mWyol39OT0

[Lowd Code @LowdCode]

DeepMind put over 200 million protein shapes in a public catalog. One structure used to take years. The method still holds. Anyone can look.

[Lowd Code @LowdCode]

Source: https://t.co/AK5apZ8W7b

[Lowd Code @LowdCode]

Airbnb now records 50 million samples a second from its live systems. That is how they watch the product for guests and hosts. The store holds 2.5 petabytes.

[Lowd Code @LowdCode]

Source: https://t.co/xU4rpwdPzo

[Lowd Code @LowdCode]

Uber scaled agent usage 7x. More than 70% of pull requests now come from local or cloud agents. Holding the model fixed, cost per session fell 52% from the June peak.

[Lowd Code @LowdCode]

@TonyZ278264 Great question and I do not know. I wonder how long first token takes.

[Lowd Code @LowdCode]

Apple's biggest on-device model is 20 billion parameters. It only wakes 1 to 4 billion per request. The rest stays in flash. That is how a phone holds the big model.

[Lowd Code @LowdCode]

Source: https://t.co/kmzdHMX4Ll

[Lowd Code @LowdCode]

A warehouse robot has now worked more than 65,000 hours in real factories. Agility says Digit has run that time across nine customer sites. They also announced more than $300 million in Digit v5 orders, but those are multi-year and still tied to milestones.