Clay Smith

technology and technology accessories. views my own.

[Clay Smith @smithclay]

Going to start posting again. Too much fun to build stuff and see what people are creating right now.

[RawTree @rawtreedb]

RawTree is OTel Native

https://t.co/wmOW3KqBDU

[Clay Smith @smithclay]

@jtregunna https://t.co/6puy0iXukJ :)

[Clay Smith @smithclay]

@raunakdoesdev https://t.co/LYXPRpJ8g6

[Clay Smith @smithclay]

Extension update: you can now stream OpenTelemetry metrics, logs and traces to @ApacheIceberg, DuckLake, Parquet or, uh, maybe Google Sheets inside a @duckdb process.

https://t.co/0hRU6jeZ1K

[Clay Smith @smithclay]

@chythram1 I’ve got a good test case: an open-source project that has a worker that writes to a pipeline that writes to r2 data catalog. Deploy button seems to work great for simple workers but breaks once you start layering on components that live outside of wrangler.toml

[Clay Smith @smithclay]

@mitchellh @ellie_huxtable The thing that was the worst about zellij (despite vastly superior UX!) was how poorly in worked vs tmux in the popular iOS terminal apps — Blink, etc had all sorts of weirdness re: keyboard shortcuts and scrolling.

[Clay Smith @smithclay]

@bentlegen @mitsuhiko I raise you “data lineage”

[Clay Smith @smithclay]

@RhysSullivan @dillon_mulroy then store it in s3 or r2 in parquet format + query with duckdb, months of traces from sessions fall under the free tiers. https://t.co/MlEEG1GGHW

[Tobias Schmidt @tpschmidt_]

AWS just shipped native OpenTelemetry metrics support in CloudWatch. This one got buried under the S3 Files announcement, but it deserves more attention!

[Clay Smith @smithclay]

.@kmad's post re: RLM+Dataframes + @isaacbmiller1's talk at dSPY SF sent me down a rabbit hole today: querying otel logs from R2 data catalog ... works great. Bonus: have Claude Desktop visualize the results.

https://t.co/rATgGAwabH

[Kevin Madura @kmad]

What if we combined RLMs with DataFrames?

Recursive language models by @a1zhang and @lateinteraction are a promising new direction for LLMs...

... but initially only operate on strings. The REPL they use is capable of much more than that.

By combining RLMs, DSPy, and DataFrame support you can do things like achieve ~87% on DABench in 15 lines of code, largely by "getting out of the way" of the model.

The approach is fairly simple:
- Import the required libraries into the REPL sandbox for your type, in this case pandas and pyarrow for efficiency
- Define a serialization approach to get the data from host to sandbox
- Define the Output fields you want
- Let the model loose in the REPL; it will iteratively explore and analyze the data until it's ready to return the response

[Clay Smith @smithclay]

@thenanyu observability as well

[Clay Smith @smithclay]

observability friends: claude now can natively make **really good** instant monitoring dashboards natively in the app from pretty much any API using the visualization/chart support that launched today.

... just tell it what you want too see, the controls/layout you want, and it works (with analysis/commentary) 🥹

[Clay Smith @smithclay]

Spoiler: this is a bigger deal than charts and diagrams.

[Claude @claudeai]

Claude can now build interactive charts and diagrams, directly in the chat.

Available today in beta on all plans, including free.

Try it out: https://t.co/tHPAZRgQkn

[Clay Smith @smithclay]

@DanielCWest @repligate yep, and pretty good at controlling ableton live to make the music https://t.co/MEDruqveKC

[Clay Smith @smithclay]

@shcallaway In progress! Hopefully it’s going to DJ the dSPY SF meetup about RLMs in March pending AV confirmation.

[Clay Smith @smithclay]

@psobot sadly, also true in sunny and warm SF with people doing fun outside things on weekends. (but still too much time on screens)

[Clay Smith @smithclay]

@_brimtown [water surges forward, lifeboats clatter, British engineer posts to Slack]

“Gentlemen, it has been a privilege going on-call with you tonight”

[gets one 😅 emoji reaction]