[
Date Prev][Date Next][
Thread Prev][Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [tsf-dev] TSF process feedback - part 2
|
Daniel,
For what it’s worth (and yes, I myself am deeply unsatisfied with “AI”, including the fact that the current hype train is for things that are definitely ‘A’, but not one bit ‘I’):
Stop calling it AI. I always use the term LLM.
How-to-use-TSF example repo: https://anotherdaniel.github.io/tsftemplate/
LLM Agent and Skill definitions based on that approach: https://github.com/AnotherDaniel/tsfagent
Example project/output generated by that agent, working with requirements and TSF quality tracking: https://anotherdaniel.github.io/psa-ng/index.html
Thanks for the links.
My 2 cents: even if a more rigorous context might keep the stochastic parrot more “honest” - and it’s up to real experts to assess the quality artifacts generated there
My suggestion is for a checking, not generation.
I'm sure people are using other tools for generation.
- I think it is a ridiculous expectation that “all AI output that is important will be reviewed by humans”.
A database of 1,397 legal decisions in cases where generative AI produced
hallucinated content – typically fake citations
https://www.damiencharlotin.com/hallucinations/
Developers have never liked other people’s code - now we think they will love, understand and work with 100x the amount of AI-slopped content? They will go through the motions if you force them, but I’m not expecting any depth here...
Ignoring the problem will not make it go away.
Regards,
Daniel
Daniel Krippner
Open Source Technologist
M +49 172 833 1416<tel:+491728331416>
daniel.krippner@xxxxxxxx<mailto:daniel.krippner@xxxxxxxx>
ETAS GmbH, ETAS/EAC
Borsigstraße 24, 70469 Stuttgart, Germany
www.etas.com<file:///Applications/Microsoft%20Outlook.app/Contents/Frameworks/EmailRendererKit.framework/Resources/www.etas.com>
Managing Directors: Dr. Thomas Irawan, Nicolet Eglseder, Mariella Minutolo
Chairman of the Supervisory Board: Dr. Walter Schirm
Registered Office: Stuttgart, Registration Court: Amtsgericht Stuttgart, HRB: 19033
From: tsf-dev <tsf-dev-bounces@xxxxxxxxxxx> on behalf of Derek M Jones via tsf-dev <tsf-dev@xxxxxxxxxxx>
Date: Tuesday, 5. May 2026 at 11:34
To: Paul Sherwood <paul.sherwood@xxxxxxxxxxxxxxx>
Cc: Derek M Jones <derek@xxxxxxxxxxxx>; tsf developer discussions <tsf-dev@xxxxxxxxxxx>
Subject: Re: [tsf-dev] TSF process feedback - part 2
Paul,
> Is it actually hybrid? The location says "in person"
Click 'reserve a spot'
Only 967 online spots left!
Who is the potential customer, i.e., the people paying with their attention
and work time?
That's quite unusual as a definition of "customer", but ok; I'm sticking with my previous answer...
It's one definition that fits the open source model.
The major established users are going to stay with the established tools,
those that they have always used.
Yes I expect they will, until they (the people and the tools) are superseded.
Is the market big enough to make it commercially worthwhile
creating better tools?
I suspect that several startups are spending VC money writing
test agents, with conformity checking/generation happening
as a side-effect.
Grok's responses to a few basic questions
https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fx.com%2Fi%2Fgrok%2Fshare%2F33b5581aa03f493284b661326197609f&data=05%7C02%7Cdaniel.krippner%40etas.com%7Cc199a92fdb244b34957a08deaa897b2e%7C0ae51e1907c84e4bbb6d648ee58410f4%7C0%7C0%7C639135704624204447%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=YGWA8ko%2FmqsaO7bLU88K8IKRaQTubEvZnodqjF1Fv7A%3D&reserved=0<https://x.com/i/grok/share/33b5581aa03f493284b661326197609f>
I'm not going there :)
No hallucinated references. The ones I checked existed.
Writing conformance statements is a skill that takes
practice to learn and become good enough.
LLMs are very good at analyzing sequences of words.
Why not provide a skills assistant that:
...
This is an interesting idea - I just need to conquer my instinctive suspicion that LLMs are mostly snake-oil.
There is certainly lots of snake-oil salesmen and many
of the claims are very overblown.
As an assistant, LLMs are great.
Just don't depend on them doing everything, which is
what the sales pitch claims.
Simplistic indeed. The first point it makes is mostly incorrect, as far as I can tell. I could also argue with a lot of
the criticisms, but frankly the thought of having to debate with any of the LLM services just makes me depressed.
As somebody who has spent a lot of time doing this stuff,
my main complaint is that the LLM missed a lot of issues.
A more detailed and specific question will fix some of this.
The issues found were valid for text claiming to be a conformance
statement.
--
Derek M. Jones Evidence-based software engineering
blog:https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fshape-of-code.com%2F&data=05%7C02%7Cdaniel.krippner%40etas.com%7Cc199a92fdb244b34957a08deaa897b2e%7C0ae51e1907c84e4bbb6d648ee58410f4%7C0%7C0%7C639135704624229267%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=uN4M9jt3uipZslmOyj37LG6UgCMB6jI%2Bzpc%2B2IpY1zQ%3D&reserved=0
_______________________________________________
tsf-dev mailing list
tsf-dev@xxxxxxxxxxx
To unsubscribe from this list, visit https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Faccounts.eclipse.org%2F&data=05%7C02%7Cdaniel.krippner%40etas.com%7Cc199a92fdb244b34957a08deaa897b2e%7C0ae51e1907c84e4bbb6d648ee58410f4%7C0%7C0%7C639135704624247423%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=28unOnIERn1kgUCUNq%2Fbsavw9v%2FuPi7ZbEBI4GlFfMA%3D&reserved=0<https://accounts.eclipse.org/>
_______________________________________________
tsf-dev mailing list
tsf-dev@xxxxxxxxxxx
To unsubscribe from this list, visit https://accounts.eclipse.org
--
Derek M. Jones Evidence-based software engineering
blog:https://shape-of-code.com
