David Röthlisberger

Co-founder of

[stb-tester @stb_tester]

Stb-tester is exhibiting at #IBC2025 this week! Drop by our stand (14.B57) to see how our private, on-device AI models can be adopted incrementally in your existing test scripts.
https://t.co/gBhL6fbp63

[David Röthlisberger @drothlis]

What happened to #10FOOT? 😟

[David Röthlisberger @drothlis]

ChatGPT-4o can find the focus from a screenshot of the UI, but still has no clue what left & right mean:

[David Röthlisberger @drothlis]

16 months since https://t.co/JOIudYqPbG and LLMs still don't know left from right.

[stb-tester @stb_tester]

Celebrating 10 years of Stb-tester:
https://t.co/OYq8skHLGq

[David Röthlisberger @drothlis]

@hugs @StooCrock Maybe I’m being too sceptical, but this reads to me like very little value compared to a human tester, and it hints at extensive human validation/triage of the results.

[David Röthlisberger @drothlis]

@hugs @StooCrock “As of January 2024, DragonCrawl executes the core-trip flow in 5 different cities once per night, and also for the Rider and Driver Android apps before releasing them to our customers.”

[stb-tester @stb_tester]

Stb-tester's new "Smart Navigation" functions:
https://t.co/oTXMREu1DS

[David Röthlisberger @drothlis]

@mvandemar @SullyOmarr @vercel Exactly! It found *a* relevant part of the codebase and implemented a hack/workaround, which is still impressive I suppose, but claiming that it understands the entire codebase, or what's causing the actual issue, is a bit of a stretch.

[David Röthlisberger @drothlis]

Primer: facebook's HTMX from 2010
https://t.co/ck27w0s5mE

[The Guardian @guardian]

The pet I’ll never forget: Simón the street dog was my loyal companion – and a daring solo explorer https://t.co/WIxDWmdGMg

[David Röthlisberger @drothlis]

This reminds me of Vandewiele et al's 2020 metastudy that found FIFTY PERCENT of ML papers trained on a particular electrohysterography dataset had train/test set data leakage: https://t.co/N29oylAghZ

[Santiago @svpino]

Andrew Ng is one of the most recognized researchers in the world.

In 2017, a team he led published a paper with a huge mistake. 11 days later, they had to publish a correction.

Here is what happened:

[David Röthlisberger @drothlis]

@raphaelmilliere @DeanBuono @ilyasut @GaryMarcus GPT-4's no better:

[David Röthlisberger @drothlis]

@raphaelmilliere @DeanBuono @ilyasut @GaryMarcus No further comment.

[David Röthlisberger @drothlis]

@jakomelli Thanks for the feedback, it was such a niche use-case that I did wonder if anyone would ever find it useful!

[David Röthlisberger @drothlis]

@hugs Whaaaat!? Surely they use the same javascript engine, HTML/CSS parsing & rendering, etc... right?

[David Röthlisberger @drothlis]

@eltishworld https://t.co/YDuLsoyCYz is very good.

[David Röthlisberger @drothlis]

In the @AppleTV latest update (tvOS 16.2 20K362) the "Reduce transparency" and "Increase contrast" accessibility settings appear to be doing the opposite.

[David Röthlisberger @drothlis]

@jamesshore You’ll get faster (immediate!) feedback, but I’d still want separate contract tests for infrequent edge cases.

[David Röthlisberger @drothlis]

@jamesshore So far I have dealt with this by running my contract tests frequently (every 10 minutes). Your runtime telemetry sounds kinda like moving the contract testing into the production code itself - interesting.