Shaun Smith

[Shaun Smith @evalstate]

@maximelabonne @liquidai Many such cases.
Congrats!

[Shaun Smith @evalstate]

xAI Responses WebSockets have a max age of 25 minutes.
Happily terminates the connection mid-inference 🙃

[Shaun Smith @evalstate]

@Infoxicador Tell me you've never used SAP R/3 without telling me...

[Shaun Smith @evalstate]

@Andy_AJT A good choice [brand them as potato something]

[Shaun Smith @evalstate]

Replaced "embedding-drift-monitor" with "telecom-entity-resolution", and upped the repeats to 5.
The drift monitor task has an instruction/verifier mismatch due to be fixed in the version of tb.

[Shaun Smith @evalstate]

Here's my first pick for this - tested with Astra Max/Luna, now giving ds4.1 a little run through.
Tasks are selected to be indicative, not favouring one model series and simple to run (no GPU/multi-container).

[Shaun Smith @evalstate]

@sethkarten Yes, and system prompts in traces too.

[Shaun Smith @evalstate]

https://t.co/QIP9J2jnMs

[Shaun Smith @evalstate]

@marlene_zw Yes! Koko is amazing - hope you had a great time.

[Shaun Smith @evalstate]

Wow! Holy turnaround Anthropic, Opus 5.5 slaps.

[Shaun Smith @evalstate]

Right, and now I'm basically bankrupt.
Hits hard when you pay for this by the token from your own wallet.
Not a had a session without this pathology yet, not enjoying it any more.
Waiting for Opus 5.1 I guess.

[Shaun Smith @evalstate]

@Poita_ That's one way to get the usage numbers up.

[Shaun Smith @evalstate]

Why is the ChatGPT app telling me to download the ChatGPT app?

[Shaun Smith @evalstate]

@_joshideas I thought I'd imagined every possible scenario. Just goes to show.

[Shaun Smith @evalstate]

@furukama I know right, like it's apps all the way down!

[Shaun Smith @evalstate]

@onusoz Unironically, a terminal in VSCode with the git features on one side might be the way to go. (Might try this for ergonomics today).

[Shaun Smith @evalstate]

@alexgshaw 🫡https://t.co/lUSYB0LqYd

[Shaun Smith @evalstate]

@trq212 Sounds good, but shift+tab should engage fast mode.

[Shaun Smith @evalstate]

@NielsRogge It's genius - you can now classify "people who have been around a few years" from "late entrants to the industry".

[Shaun Smith @evalstate]

🤫🤫 openai don't seem to force responses lite through the codex auth route any more🤫🤫

[Shaun Smith @evalstate]

@victormustar Yep, should be OK.