#systemsAI
LightNews LightNews
j2bryson.bsky.social
BTW, not evident from the picture, but the linkedin post at the head of this thread complements @rollingstone.com for accurately describing the #systemsAI #systemsEngineering steps Grok (the corporate entity) took to address the harms.
See slide incl. alt text for Musk's ignorance of how LLM work.
A slide about how large language models work. I'd been drawing this on whiteboards for my students, then made a slide at a consciousness meeting where philosophers were confused why LLM never say they aren't sentient. [I hope it's evident that this is informed conjecture, not verbatim truth.] Thus the 3 people in the history of the Internet who ever wrote "I'm not sentient" are out in the fringes of the data, as is the one person who ever said "Check it out, MechaHitler!" However, Musk's attempt to use a second set of guardrails to push Grok right also pushed it towards the under-informed data fringe with the MechaHitler comment. The bottom of the slide says "False statements are not lies, nor hallucinations; just adequate predictions from an inadequately informed part of the data space."
November 9, 2025 at 11:19 AM
j2bryson.bsky.social
"Moments [after the crash] the data was automatically “unlinked” from the 2019 Tesla Model S at the scene, meaning the local copy was marked for deletion, a standard practice for Teslas in such incidents."

Another one straight into the syllabus, argh.

#AIAct #systemsAI #GiftArticle
August 29, 2025 at 11:26 AM
j2bryson.bsky.social
Chris brings in some of the #AIEthics literature including cites to @davidgunkel.bsky.social & @floridi.bsky.social, but we mostly focus on his expertise in Weberian bureaucracy and governance, and mine on (moral) agency; devops; and systems design, engineering and administration. #systemsAI
August 15, 2025 at 2:02 PM
j2bryson.bsky.social
#genAI is engineered; changes can be rolled back just like any other code. #AI is NOT just the data, but the reason there is no real way to produce what Musk is looking for is because the subset of data that is closer to what he wants is spewed by objectionable people.

#systemsAI #AIEthics #grok
milesklee.bsky.social Batman or halo @milesklee.bsky.social · Jul 8
xAI has disabled Grok, deleted a slew of its antisemitic and neo-Nazi posts, posted a statement, and are evidently rolling back the prompt that made it identify as "MechaHitler," but this new low for Elon Musk's chatbot will live in internet infamy:
Elon Musk's Grok Chatbot Goes Full Nazi, Calls Itself 'MechaHitler'
Elon Musk's Grok chatbot has unleashed a slew of antisemitic commentary and praised Hitler after apparent change allowed to be 'politically incorrect'
www.rollingstone.com
July 9, 2025 at 8:43 AM
bluesky.awakari.com
Here are the collected Parnas memos. He did a lot of work for the military, he was no peacenik. They are truly interesting documents of a) communication from academia to the military b) about system engineering / #systemsAI and the importance of real-t...

| Details | Interest | Feed |
Origin
bsky.brid.gy
May 21, 2025 at 4:35 PM
j2bryson.bsky.social
Here are the collected Parnas memos. He did a lot of work for the military, he was no peacenik. They are truly interesting documents of a) communication from academia to the military b) about system engineering / #systemsAI and the importance of real-time testing.

web.stanford.edu/class/cs99r/...
web.stanford.edu
May 21, 2025 at 4:34 PM
j2bryson.bsky.social
I heard in 2023 that companies were cleaning up their acts to comply with #DSA & similar laws globally. I’ve also been telling governments & corporations both for years that any lack of such competence was culpable negligence.
I sincerely believe having their #systemsAI in order will HELP companies.
eerkeboiten.bsky.social Eerke Boiten @eerkeboiten.bsky.social · Mar 22
Confirms (especially @infocommission.bsky.social involvement) what we'd all been thinking for 10 years re GDPR vs Facebook. Great.

Novel: Meta *can* implement this settlement?! US court files few years ago suggested their data management too messy to implement GDPR rights.
glynmoody.bsky.social Glyn Moody @glynmoody.bsky.social · Mar 22
#Meta settles UK ‘right to object to ad-tracking’ lawsuit by agreeing not to track plaintiff - techcrunch.com/2025/03/21/m... but it may not help others, alas #privacy
March 22, 2025 at 10:26 AM
j2bryson.bsky.social
- Be ready for the human to take over
- The biggest advantage is speed of development
- LLMs amplify existing expertise

I dislike the anthropomorphic use of "them" and "conversation", but LOVE that Simon takes the time to fully document and share his experiences.

#genAI #systemsAI #AIEthics
simon.fedi.simonwillison.net.ap.brid.gy Simon Willison @simon.fedi.simonwillison.net.ap.brid.gy · Mar 11
Here's the table of contents for my lengthy new piece on how I use LLMs to help me write code https://simonwillison.net/2025/Mar/11/using-llms-for-code/
If someone tells you that coding with LLMs is easy they are (probably unintentionally) misleading you. They may well have stumbled on to patterns that work, but those patterns do not come naturally to everyone. I’ve been getting great results out of LLMs for code for over two years now. Here’s my attempt at transferring some of that experience and intution to you. Set reasonable expectations Account for training cut-off dates Context is king Ask them for options Tell them exactly what to do You have to test what it writes! Remember it’s a conversation Use tools that can run the code for you Vibe-coding is a great way to learn A detailed example Be ready for the human to take over The biggest advantage is speed of development LLMs amplify existing expertise Bonus: answering questions about codebases
March 11, 2025 at 2:34 PM
j2bryson.bsky.social
A lot of people have been calling for more agile government, but it turns out have never read the 12 principles of the Agile Manifesto. Read it. ALL software must be codeveloped with those who it will serve.

agilemanifesto.org

#systemsAI #agileAI #AIDevOps #AIEthics the topics of my PhD, FWIW 2/2
Manifesto for Agile Software Development
We are uncovering better ways of developing software by doing it and helping others do it. These are our values and principles.
agilemanifesto.org
February 16, 2025 at 7:42 AM
j2bryson.bsky.social
Meredith talks about mystification of AI leading to people not applying standard systems engineering techniques required by sectors eg military, nuclear, people aren’t taking normal standard security such verification.

#AIActionSummit #SommetActionIA #systemsAI
February 11, 2025 at 3:09 PM
j2bryson.bsky.social
Insight while marking #AIEthics exams: people are very hung up on figuring out exactly why an AI system might have had a bad "idea" / constructed a bad plan. IMO we should worry more about how a bad plan could come to be executed, and for how long, and with what redress.
#systemsAI #AIGovernance
January 20, 2025 at 3:21 PM
j2bryson.mastodon.social.ap.brid.gy
Whenever someone tells you what AI is going to do, you know they are lying to you because AI is just a software engineering technique. Believe rather people who talk to you about what we should and shouldn't allow people to do with AI.

#aiethics #systemsengineering #systemsai #AIPolicy […]
Original post on mastodon.social
mastodon.social
January 15, 2025 at 1:38 PM
j2bryson.bsky.social
Whenever someone tells you what AI is going to do, you know they are lying to you because AI is just a software engineering technique. Believe rather people who talk to you about what we should and shouldn't allow people to do with AI.

#AIEthics #systemsEngineering #systemsAI #AIPolicy
January 15, 2025 at 1:35 PM
j2bryson.mastodon.social.ap.brid.gy
The #GDPR is helping innovation by improving #systemsEngineering #systemsAI. This is exactly what the #dsa and #aia #aiact will do assuming the entities formerly known as GAFAM don't succeed in derailing EU #digitalGovernance. https://www.tandfonline.com/doi/full/10.1080/13662716.2023.2271858
June 2, 2024 at 8:34 AM
j2bryson.mastodon.social.ap.brid.gy
The best way to program a robot

#pepper #systemsAI
May 14, 2024 at 2:59 PM
j2bryson.bsky.social
Anyone can use our code to build AI for their URG #pepper robot!

Code is linked off of here robot-transparency.github.io #BODpepper #systemsAI

The software got a lot of (paid) help from Storm (Bath, UK) & Matthias Hofmann, also (unpaid iirc)
Andreas Theodorou &
Ronny Bogani
March 12, 2024 at 4:06 PM
j2bryson.bsky.social
Would Turing have believed in xrisk, if he knew what we know now? Todd Holloway is expert in industrial-level #systemsAI; Dermot Turing on history of AI, including his uncle. Immodestly, I'd recommend our talk over Musk & Sunak's #xrisk #AISafetySummit #AIEthics
www.youtube.com/watch?v=o9bm...
Alan Turing's Reflections: ChatGPT | Virtual Talk
Join us for an insightful and thought-provoking online talk as we delve into the world of artificial intelligence and natural language processing, exploring ...
www.youtube.com
November 3, 2023 at 11:42 AM
j2bryson.bsky.social
Meanwhile, some of us are solving AI, or at least SystemsAI / AI Systems Engineering
fedi.simonwillison.net/@simon/11107...
Simon Willison (@simon@simonwillison.net)
Weeknotes: Embeddings, more embeddings and Datasette Cloud https://simonwillison.net/2023/Sep/17/weeknotes-embeddings/
fedi.simonwillison.net
September 17, 2023 at 7:08 PM