Anthropic: "In a joint study with the UK AI Security Institute and the Alan Turing Institute, we found that as few as 250 malicious documents can produce a "backdoor" vulnerability in a large language model—regardless of model size or training data volume. "

https://www.anthropic.com/research/small-samples-poison

#cybersecurity #backdoor #datapoisoning #llm #ai

Anthropic: "In a joint study with the UK AI Security Institute and the Alan Turing Institute, we found that as few as 250 malicious documents can produce a "backdoor" vulnerability in a large language model—regardless of model size or training data volume. "

https://www.anthropic.com/research/small-samples-poison

#cybersecurity #backdoor #datapoisoning #llm #ai

@neil This is going to be the tip of a pin stuck in an iceberg.

I'm frankly amazed at how long it is taking for criminals to weaponise #LLMs like this. Even my spam email is still trivial to spot.

As with cybersecurity, regulation is non-existent and governments don't care or are made up of people who only know what lobbyists bother to tell them.

I despair...

...and keep working on #AutonomiDweb

#LLM

Where our society is heading? Probably.

#ai #llm #chatgpt

Where our society is heading? Probably.

#ai #llm #chatgpt

#Barcelona people, here's a rare opening for a software engineer + machine learning, LLM work, and all that jazz

https://jobs.apple.com/en-us/details/200615333-0215/applied-ml-engineer-localisation-release-engineering?team=SFTWR

#swiftlang #llm #ml #job #apple

Over 40 years, we were collectively told to give tax cuts to rich people.

And we were told that if we did that, wealth would trickle down and everyone would be better off.

Over 40 years, pretty much everything got cut to fund these tax cuts.

Schools. Hospitals. Public housing. Public transport. Universities. Roads projects. Mental health services. Welfare payments.

People literally went homeless or starved, so rich people could get tax cuts.

Because the wealth would trickle down.

Eventually the eroding of public goods caused social dislocation.

So governments further cut those public goods to fund more police and prisons. To continue giving tax cuts to rich people.

But they said the wealth would trickle down.

Eventually the climate started changing because of the amount of toxic fossil fuel pollution in the atmosphere.

So governments chose to keep the tax cuts rather than fund infrastructure to reduce emissions.

(Many of those billionaires getting tax cuts made their money selling toxic fossil fuels.)

And as the oceans and atmosphere warmed, the bushfires, droughts, hurricanes, cyclones, floods, and droughts got worse.

But they said the wealth would trickle down.

Eventually people were getting pissed off at the dire state of the world.

The rich misdirected that anger at immigrants!

And First Nations!

And trans people!

And neurodivergent people!

Anyone but the billionaires who got the tax cuts.

So governments chose to keep the tax cuts. (For the rich. Everyone else got new tariff taxes.)

But they said the wealth would trickle down.

So did the wealth trickle down?

Well...

A group of billionaires saw this kinda cool tech demo.

It predicted the next pixel of an image, based on the colour patterns of every image on the internet.

It also predicted the next word in a sentence, based on an analysis of every piece of writing on the internet.

The rich decided that this clearly showed that a sentient computer was just around the corner.

The problem was these tech demos needed servers with a lot of GPUs to work.

So the rich took all the money they got from those tax cuts.

And they bought GPUs.

Millions and millions and millions and millions of GPUs.

All the tax cuts? All the underfunded schools? All the draconian welfare cuts? All the public housing shortages? The delays in funding clean energy.

In the end, it didn't trickle down.

And instead of all the public goods it could have bought...

...We'll be left with millions and millions and millions of GPUs in a landfill.

#ChatGPT #Claude #AI #LLM #capitalism #socialism #business #politics #Nvidia

Claude Code’s context is like a lithium battery. It reserves 22.5% for autocompactification and you never want to go past about 80% to avoid accidentally triggering that.

Compactification = full frontal lobotomy. Much better restarting.

#LLM #AI #Claude #VibeCoding.

@derpostillon hats geschafft. #Gemini benutzt ihn jetzt auch als vertrauenswürdige Quelle.

https://www.der-postillon.com/2017/05/laserpointer.html

#AI #LLM

Claude Code’s context is like a lithium battery. It reserves 22.5% for autocompactification and you never want to go past about 80% to avoid accidentally triggering that.

Compactification = full frontal lobotomy. Much better restarting.

#LLM #AI #Claude #VibeCoding.

@derpostillon hats geschafft. #Gemini benutzt ihn jetzt auch als vertrauenswürdige Quelle.

https://www.der-postillon.com/2017/05/laserpointer.html

#AI #LLM

Deloitte refunds Aussie gov after AI fabrications slip into $440K welfare report

Big Four consultancy billed Canberra top dollar, only for investigators to find bits written by a chatbot

"....Deloitte has agreed to refund part of an Australian government contract after admitting it used generative AI to produce a report riddled with fake citations, phantom footnotes, and even a made-up quote from a Federal Court judgment...."

https://www.theregister.com/2025/10/06/deloitte_ai_report_australia/

#hallucination #ai #llm #chatbot

Deloitte refunds Aussie gov after AI fabrications slip into $440K welfare report

Big Four consultancy billed Canberra top dollar, only for investigators to find bits written by a chatbot

"....Deloitte has agreed to refund part of an Australian government contract after admitting it used generative AI to produce a report riddled with fake citations, phantom footnotes, and even a made-up quote from a Federal Court judgment...."

https://www.theregister.com/2025/10/06/deloitte_ai_report_australia/

#hallucination #ai #llm #chatbot

This research reveals LLMs are opaque vessels for political bias, engineered by undemocratic corporate entities for profit.

Their ability to subtly shape ideology and discourse presents a profound societal threat, poised to influence mass opinion and exacerbate structural conflicts like class warfare.

We must urgently confront their potential as potent, unregulated tools of ideological control.

https://arxiv.org/html/2410.01810v1

#llm #technology #politics

Remember my #issue with same hard- and software but one device doesn't support the native resolution of an external TFT?

I could fix it today with the help of #Claude #AI using its #Opus 4.1 #LLM: https://karl-voit.at/2025/09/14/fix-wrong-34-TFT-resolution/

As so often, it was actually a layer-8-issue. I unintentionally got a Debian testing kernel on the host that worked with the TFT. After using the testing kernel on the other host, the TFT now works on both machines.

However, I now need to find out how to *not* get testing packages for my systems where I don't want it ... different issue.

@alienghic @gnomon thanks for your help! 🙇

#publicvoit

#Barcelona people, here's a rare opening for a software engineer + machine learning, LLM work, and all that jazz

https://jobs.apple.com/en-us/details/200615333-0215/applied-ml-engineer-localisation-release-engineering?team=SFTWR

#swiftlang #llm #ml #job #apple

Dear LLM, please make me a pet site about cats. Make sure to include recipes!

#ai #slop #aislop #llm

Funny, people always forget to fill out all the details on their website leaving the "Lorem ipsum", the LLM engines are replicating that mistake on their machine generated slop pages

#aislop #ai #slop #llm #loremipsum