Lightnews — Scholar-powered news

Reposted by Dr Sarah

WIRED @wired.com · 23h

A canonical problem in computer science is to find the shortest route to every point in a network. A new approach beats the classic algorithm taught in textbooks. www.wired.com/story/new-me...

A New Algorithm Makes It Faster to Find the Shortest Paths

A canonical problem in computer science is to find the shortest route to every point in a network. A new approach beats the classic algorithm taught in textbooks.

www.wired.com

3 37 160

Dr Sarah @drsezzer.bsky.social · 1d

@emalliab.bsky.social :o

Dr Sarah @drsezzer.bsky.social · 2d

Interesting title given the actual paper it's talking about! HHH alignment is naturally left leaning.

The Register @theregister.com · 2d

OpenAI GPT-5: great taste, less filling, now with 30% less bias

AI model maker touts effort to depoliticize its product OpenAI says GPT-5 has 30 percent less political bias than its prior AI models.…

dlvr.it

1

Reposted by Dr Sarah

Go Jauntly: walks & nature trails @gojauntly.bsky.social · 3d

Today is #WorldMentalHealthDay. Step outside and take a moment for yourself. Feel the air, notice the colours, listen to the world around you 🍂

1 6 26

Dr Sarah @drsezzer.bsky.social · 2d

bsky.app/profile/drse...

Dr Sarah @drsezzer.bsky.social · 3d

The blog linked from here is pretty cute. Interesting way to check what models may have been trained on.

Sung Kim @sungkim.bsky.social · 3d

Defying Transformers: Searching for "Fixed Points" of Pretrained LLMs by Jiacheng Liu

He wondered what CAN'T be transformed by Transformers? So, he wrote a fun blog post on finding "fixed points" of your LLMs. If you prompt it with a fixed point token,

1

Dr Sarah @drsezzer.bsky.social · 2d

I guess there are many ways of injecting malicious data into a training set (I personally like the well timed Wikipedia edits idea). I haven't read the full report, but you might find this interesting and indirectly relevant...

1 1

Dr Sarah @drsezzer.bsky.social · 3d

💯 An example from earlier this year... www.axios.com/2025/03/06/e...

Exclusive: AI chatbots echo Russian disinformation, report warns

Bots from Microsoft, Google, OpenAI and others spew falsehoods.

www.axios.com

1 1

Dr Sarah @drsezzer.bsky.social · 3d

The blog linked from here is pretty cute. Interesting way to check what models may have been trained on.

Sung Kim @sungkim.bsky.social · 3d

Defying Transformers: Searching for "Fixed Points" of Pretrained LLMs by Jiacheng Liu

He wondered what CAN'T be transformed by Transformers? So, he wrote a fun blog post on finding "fixed points" of your LLMs. If you prompt it with a fixed point token,

Dr Sarah @drsezzer.bsky.social · 5d

We need to understand why they want this (and the other nonsense), both as a party and as individuals. It'll likely be 'chase the money', but logical argument won't cut it, we need to be better informed so we can't fight it all more effectively.

1

Dr Sarah @drsezzer.bsky.social · 5d

Pep765... Nice.

Dr Sarah @drsezzer.bsky.social · 5d

Ah, young Geralt is adorable!

Bookish Scifi & Fantasy @bookishsff.bsky.social · 13d

Geralt of Rivia is back!

Crossroads of Ravens is out today: a brand new tale of the #Witcher, tracing his first adventures as a fresh graduate of Kaer Morhen. An absolute must-read for any Witcher fans 🗡️ ✨

💙📚🪐
#Sapkowski
#SFF

A pile of the book on a table in a bookshop.

1

Dr Sarah @drsezzer.bsky.social · 5d

Oh, do let us know how you find it. :)

1

Reposted by Dr Sarah

Uncle Duke @uncleduke1969.bsky.social · 7d

Sign reads “Large Print Audio Books” over book store shelf. Inset picture of Garrett Morris from Saturday Night Live cupping his hands and appearing to yell.

100 430 3.2K

Reposted by Dr Sarah

nixCraft @cyberciti.biz · 7d

I’m looking forward to the day when an LLM, out of sheer frustration, tells a "vibes coder" that "the code works on my machine" and that it’s a skill problem. That day, be afraid, very afraid, of the machines 😳

11 17 160

Dr Sarah @drsezzer.bsky.social · 7d

Offs. Anthropic at its best. #rollseyes

Mark Riedl @markriedl.bsky.social · 8d

AI is going to take over the wo—oh are you stressed out little buddy? Here, let me help you

dan @danabra.mov · 8d

apparently when Claude starts lying and trying to evade work, it means it's stressed out and you need to reduce its level of stress. e.g. scope down the task a bit more, suggest to break it down into parts, suggest to extract a minimal repro

1 2

Reposted by Dr Sarah

Phil Swatton @philswatton.bsky.social · 9d

I wrote a slightly longer response to the NS piece on my substack. You can read it here: dysfunctionalprogramming.substack.com/p/some-comme...

1 3

Reposted by Dr Sarah

Philip Ball @philipcball.bsky.social · 9d

Sorry also: Sooo good to hear the notion of AGI dissected and largely binned, instead of just uncritically accepted. And for that matter, the same with the idea that mere scaling up of LLMs will get you "there" (wherever "there" is supposed to be.)

3 9 43

Reposted by Dr Sarah

Philip Ball @philipcball.bsky.social · 9d

First, Turing's paper was rather inconsistent and illogical. But no big deal, he was just having fun. (Turing, contrary to common legend, was extremely playful.) The Test was never meant as some kind of benchmark for AI, and it's absurd that that's what it became. /2

1 6 41

Dr Sarah @drsezzer.bsky.social · 8d

Haha this! 💯

Marcus Hutchins @malwaretech.com · 9d

You don't really get to appreciate just how stupid vibe coder's takes are until you try to use AI to write anything even remotely novel. It just endlessly shits the bed unless the code you're writing is something you could have just copy and pasted from stack overflow.

1

Reposted by Dr Sarah

The Register @theregister.com · 10d

Mad man builds chatbot in Minecraft with redstone, Python, and patience

Replies are slow and it's prone to gibberish – just like any other AI Never mind Doom running on a potato, or whatever – the next generation of ridiculous computing belongs to Minecraft YouTuber Sammyuri, who built a working chatbot in the perennially popular voxel building sandbox.…

dlvr.it

2 12

Reposted by Dr Sarah

Jamie Cummins @jamiecummins.bsky.social · 12d

@science.org just dropped a story covering this preprint! Check it out below, and thanks to @cathleenogrady.bsky.social for the great write-up! www.science.org/content/arti...

1 16 36

Dr Sarah @drsezzer.bsky.social · 11d

Short article covering the psychometrics and LLMs work we published a few months back. Hopefully, more readable (i.e. less maths!) :)

@philswatton.bsky.social

cetas.turing.ac.uk/publications...

Patterns, Not People: Personality Structures in LLM-powered Persona Agents

Expert Analysis

cetas.turing.ac.uk

1 1 4

Dr Sarah @drsezzer.bsky.social · 11d

In a similar vein, AI doomers worry me, they seem to think that if we were able to create a super intelligence it would kill us all (wipe us out like bugs) Can't help thinking (possibly hoping) it would be better than that, but more worryingly the doomers are just reflecting their own values. :(

1 1

Dr Sarah @drsezzer.bsky.social · 11d

Do you think perhaps there's something the human brain/our psychology can't handle about being so rich? Do we need an element of insecurity to, well, literally stay sane?

1 1

Dr Sarah @drsezzer.bsky.social · 11d

This is utterly terrifying!

1 1