>>> But inference engines are now out of fashion, because they rely (at least, for now) on human curated knowledge bases, and lack the facile fluency of #LLMs.

I do not believe that we can achieve #ArtificialGeneralIntelligence #AGI without a semantic layer comprising at least an explicit ontology, a semantic model, and an auditable, sourced, set of inference rules.

I think #LLMs are probably a dead end. And I think they are distracting from technologies which could be genuinely useful.

#LLM-Generated Invariants for Bounded Model Checking Without Loop Unrolling

https://dl.acm.org/doi/10.1145/3691620.3695512

"We investigate a modification of the classical Bounded Model Checking (BMC) procedure that does not handle loops through unrolling but via modifications to the control flow graph (CFG). A portion of the CFG representing a loop is replaced by a node asserting invariants of the loop[...]."

(continues next toot...)

2/3

@gerrymcgovern it will end in a financial crisis. Now we are in good humor and smiling, but when the crisis will hit, many people will be crying and complaining that nobody told them about the issues. They will be loosing their money and some billionaires will cash in, profiting from the losses of the many.
Hey people, we are telling you since years that #LLMs don’t work as you believe. We are the nobody experts in #AI who have told you about the issues!

AI Is Homogenizing Our Thoughts (June 2025):

https://www.newyorker.com/culture/infinite-scroll/ai-is-homogenizing-our-thoughts

"With the #LLM “you have no divergent opinions being generated,” Kosmyna said. She continued, “Average everything everywhere all at once—that’s kind of what we’re looking at here.”

#AI is a technology of averages: [#LLMs] are trained to spot patterns across vast tracts of data; the answers they produce tend toward consensus, both in the quality of the writing [...] and in the calibre of the ideas."

1/2

This article explains it quite nicely. An LLM is not an artificial mind but rather a zillion-dimensional map of the abstract vector space of all the texts on which it has been trained. Every possible text is a single one-dimensional path through that space. The LLM treats your input as the beginning of a path through that landscape and tries to continue it.

Likewise, an image generator is a zillion-dimensional map of all the images in its training data set. It doesn't know anything about the world, it just tries to find the coordinates of a point in the vector space of all possible images that matches your prompt.

https://www.psychologytoday.com/us/blog/the-digital-self/202507/ais-hidden-geometry-of-thought

If we ever see a real artificial mind, some kind of LLM will probably be a small but significant component of that, but the current wave of machine learning will most likely come to a grinding halt very soon because of a lack of cheap training data. The reason why all of this is happening now is simple: The technologies behind machine learning have been around for decades, but computers weren't fast enough and didn't have enough memory for those tools to become really powerful until the early 2000s, and around the same time, the Internet went mainstream and got filled with all kinds of data that could be datamined for training sets. Now there is so much synthetic content out there that automated data mining won't work much longer, you need humans to curate and clean the training data, which makes the process slow and expensive. I expect to see another decades long AI winter after the commercial hype is over.

If you look for real intelligence, look at autonomous robots and computer game NPCs. There you can find machine learning and artificial neural networks applied to actual cognitive tasks in which an agent interacts with its environment. Those things may not even be as intelligent as a rat yet, but they are actually intelligent, unlike LLMs.

Patrick Hall, who teaches #data #ethics & #MachineLearning at George Washington University, said he's not surprised #Grok ended up spewing toxic #content, given that the #LLMs that power #chatbots are initially trained on unfiltered online data.

“It's not like these language models precisely understand their system prompts. They're still just doing the statistical trick of predicting the next word.”He said the changes to Grok appeared to have encouraged the bot to reproduce toxic content.
#tech

@josephcox wrote

Turns out if you just use a bunch of big words, you can trick LLMs into telling you how to (redacted) or hack an ATM.

That's a totally predictable result, sadly. LLMs don't do higher symbolic reasoning, just sophisticated word association.

Their manual alignment/safety training includes probing with harmful prompts ("red teaming," IIRC), but it's easy to bypass those guardrails if you're creative: for example, I tricked Google Gemini into giving reckless medical "advice" by adopting the language and tone of a physician consulting about a patient rather than a patient asking about their own health.

Basically, any highschool improv kid should be able to beat their protections by playacting different roles and personalities.