PWV Portfolio Feed

interchange standards are not data storage standards, i shout, about both iso 8601 and 6709

i checked who the original poster on twitter was, it's some crypto zealot, unsurprising

the vibes are right of the original post but i feel like we're doubling the dunning effect here. maybe i'm tripling it, but like, sound off in the comments computer historians, but this explanation is bullshit?

sorry but like, this isn't right?
iso 8601 is a spec for text representations of datetime. it doesn't have an epoch and certainly not one that is 1850.
85 cobol does have dates stored from an epoch, but it's either 1601 or 1582

For further reduce latency, we've deployed GPUs in North America and Europe, with more regions coming soon. Cloudflare workers route your requests to the nearest data center, ensuring a responsive experience worldwide.

To meet our strict latency targets, we leveraged speculative decoding. This technique parallelizes token generation by using the input as a reference, significantly speeding up predictions without sacrificing quality.

We fine-tuned Zeta using Direct Preference Optimization (DPO), teaching it both good and bad edit patterns. This allows us to deliver high-quality predictions that avoid common pitfalls, even in tricky edge cases.

Our evals for Zeta use a larger LLM to assess if Zeta's suggested edits match our intent. This allows us to validate predictions even when outputs vary, ensuring consistent quality across different coding styles.

Zeta predicts edits at arbitrary locations in your code. It rewrites larger chunks around the cursor, enabling natural, context-aware predictions that go beyond simple completions.

As you work, Zed predicts your next edit. Apply it with a simple 'tab' press, or chain multiple edits by pressing 'tab' repeatedly. It's a natural extension of the Zed experience, saving you time and keystrokes.