Dr Sarah
banner
drsezzer.bsky.social
Dr Sarah
@drsezzer.bsky.social
Software Engineer / GenAI researcher (LLM agents) ex-The Alan Turing Institute.
Witcher fan (books/games not netflix). Recreates retro games in python for fun! *Opinions are mine* or borrowed from those more insightful.
*Won't age verify, so no DMs.*
Pinned
Engineering with these things can be frustrating, but every now and then they make be laugh.

>>> quit
It was nice chatting with you. If you change your mind and want to run me locally on Ollama, just let me know. Have a great day!

1/5
Reposted by Dr Sarah
The finale of the great TV comedy show Blackadder goes Forth.
Back then I didn't expect one of the funniest programmes ever to end with a scene that makes me cry every time I see it.
Every single time.
November 11, 2025 at 12:33 PM
I have some 'feels' for the cursor agent this morning, it seems to be struggling with the 'dont be overly verbose' part of the system prompt. Poor thing.
November 11, 2025 at 9:22 AM
Reposted by Dr Sarah
Super moon. 8”x6” oil
November 9, 2025 at 4:48 PM
Reposted by Dr Sarah
That lamp looks PISSED.
November 8, 2025 at 1:17 PM
Reposted by Dr Sarah
👀BioWare has confirmed in an N7 Day blog post that a new Mass Effect game is officially in development

EA and BioWare say the team is now fully focused on expanding the Mass Effect universe

#gamedev #gaming #N7Day #MassEffect
November 7, 2025 at 6:12 PM
Reposted by Dr Sarah
Happy #N7Day Commander Shepards!
November 7, 2025 at 6:23 PM
Reposted by Dr Sarah
"Shepard Commander" #MassEffect #N7Day (the best companion don't @ me)
November 7, 2025 at 5:45 PM
Reposted by Dr Sarah
Reposted by Dr Sarah
Cowardly programmers calling it a 'Garbage collector' rather than an Orphan Destroyer
November 6, 2025 at 11:48 AM
'Our results revealed that Extraversion,
Agreeableness, Conscientiousness, and Openness to Experience were all significantly different. In
contrast, Neuroticism did not reach statistical significance.'

Or cynically... It's just random with some RLHF/safety training impacts.
Christos-Nikolaos Zacharopoulos, Revekka Kyriakoglou
Decoding Emergent Big Five Traits in Large Language Models: Temperature-Dependent Expression and Architectural Clustering
https://arxiv.org/abs/2511.04499
November 7, 2025 at 6:49 AM
Reposted by Dr Sarah
Sat in a room last week with human rights defenders from Gaza and Sudan saying how to get, transmit and verifiably archive video of war crimes, crimes against humanity and genocide

People are risking their lives to report on drones wiping out families

as GenAI destroys the ability to believe them.
"How we choose to act now will determine whether our shared basis for reality — what we see and hear — remains trustworthy."

Generative video's threat to visual truth and what we can do about it, and why concerted action matters now, in @transformernews.ai

www.transformernews.ai/p/sora-is-he...
Sora Is Here. The window to save visual truth is closing
Opinion: Sam Gregory argues that generative video is undermining the notion of a shared reality, and that we need to act before it’s lost forever
www.transformernews.ai
November 6, 2025 at 6:52 AM
Reposted by Dr Sarah
OpenAI is using terms of service to shield themselves from liability for harms due to substitution of ChatGPT for real therapist or real doctor support.
November 6, 2025 at 2:35 AM
Reposted by Dr Sarah
We just defeated Nigel Farage’s Bill in Parliament to tear up people’s rights and withdraw from the ECHR.

Farage wants to do away with the Britain Churchill built and turn it into a version of Trump's America. We stopped him.
October 29, 2025 at 3:30 PM
Reposted by Dr Sarah
‘In a twist that befuddled researchers for a year, almost no human beings visit the sites, which are hard to browse or search. Instead, their content is aimed at crawlers, the software programs that scour the web & bring back content for search engines & LLMs’
www.washingtonpost.com/technology/2...
Russia seeds chatbots with lies. Any bad actor could game AI the same way.
In their race to push out new versions with more capability, AI companies leave users vulnerable to “LLM grooming” efforts that promote bogus information.
www.washingtonpost.com
April 18, 2025 at 9:57 AM
Reposted by Dr Sarah
"Chimpanzees revise their beliefs if they encounter new information, a hallmark of rationality that was once assumed to be unique to humans"

Chimps Are Capable of Human-Like Rational Thought, Breakthrough Study Finds
October 31, 2025 at 10:00 PM
Reposted by Dr Sarah
They said it couldn’t be done, but Fintech somehow invented something worse than payday loans
October 31, 2025 at 11:43 PM
Although we called out [*] AI generated code increasing the risk of malicious dependencies, we're kicking ourselves we didnt think of calling it *Slopsquatting *

www.theregister.com/2025/10/30/p...
npm hit by PhantomRaven supply chain attack
: PhantomRaven slipped over a hundred credential-stealing packages into npm
www.theregister.com
October 31, 2025 at 10:28 AM
A "bollard in a wig" - The Guardian.

I'm not going to watch season 4, even with Cavill as a (almost) perfect Geralt, the abuse of the story line and twisting of the characters, compared with both the books and the games is just too insulting.

www.theguardian.com/tv-and-radio...
The Witcher season four review – Liam Hemsworth is as charismatic as a bollard in a wig
The replacement that Netflix has chosen for lead actor Henry Cavill is utterly lacking in his predecessor’s likability. The ex-Neighbours actor is a scowling lunk who brings a wildly uneven vibe to th...
www.theguardian.com
October 31, 2025 at 10:06 AM
Reposted by Dr Sarah
"Two extremely rare dinosaur "mummies" found in the badlands of Wyoming are the first examples of hoofed reptiles, according to a new study."

www.livescience.com/animals/dino... #archaeology #WY
First-ever 'mummified' and hoofed dinosaur discovered in Wyoming badlands
Researchers have unearthed two dinosaur "mummies" in the badlands of Wyoming, confirming duck-billed dinosaurs had hooves, alongside a string of other discoveries.
www.livescience.com
October 31, 2025 at 12:58 AM
Reposted by Dr Sarah
When you use LLMs to write code in a human-in-the-loop way, you ARE the agent. Multiple LLMs in different web tabs, you write the best prompts, you check the code, ask the right questions, cross-check with some other LLM. Being good agents is not trivial, but very rewarding.
October 29, 2025 at 4:56 PM
Reposted by Dr Sarah
as a condition of funding, we were asked to affirm that we wouldn’t undertake any diversity, equity, and inclusion work, whether or not we used the government funds to do so. The PSF simply couldn’t agree to that statement,
October 27, 2025 at 2:47 PM
Human generated content is what pulls me into social media for too many hours... When do you reckon big tech will realise AI generated content just isn't that addictive?

Are they really building business models to emotionally exploit vulnerable users, as a mitigation?

Yeah, I guess they are.
October 28, 2025 at 7:15 AM
Reposted by Dr Sarah
At least for my brain, this is the clearest explanation I’ve seen yet of the recent types of attention and other variations of LLMs
A short talk on the main architecture components of LLMs this year + a look beyond the transformer architecture: www.youtube.com/watch?v=lONy...
October 27, 2025 at 7:01 PM
Reposted by Dr Sarah
Saying no to important, necessary funding takes guts. That's what @python.org did, turning down $1.5 million in NSF funding due to the federal government's demands that they not "operate any programs that advance or promote DEI."

pyfound.blogspot.com/2025/10/NSF-...
The PSF has withdrawn a $1.5 million proposal to US government grant program
pyfound.blogspot.com
October 27, 2025 at 9:03 PM
The final installment is published...

cetas.turing.ac.uk/publications...
cetas.turing.ac.uk
October 24, 2025 at 2:25 PM