7 things to remember about the "AI will kill us all" claims. Points 3 and 4 are the most vital for this discussion, and for anyone who is thinking of trusting AI agents.

I believe you confirmed exactly why AI will kill us all. “There’s no emotion, there’s no reasoning” That is such a chilling sentence and it’s very worrying.

I would say that plenty of things don't have these qualities. It's only chilling if you also believe these things are sentient or have intent.

I’ve seen it with my own eyes. No intent, no sentiment, just a computer doing what it wants. 39 years ago, I was the person who had to pull the power on a computer that thought it was doing the right thing. Stopped it from creating an even worst disaster than happened on that day

What was the event? 1987?

I thought you would have googled that 😜😂 Black Monday as it’s commonly known but really the whole story all started the Thursday before.

LLMs are not AI. LLMs are not going to wipe out humanity. Nor are they going to cure cancer. The only threat they pose is that they are all hype & are actually useless - leading to a massive economic slump. They need to be regulated but not for the reasons these liars claim.

I feel like recently something went down, some folks saw or did something they feel was a bit too much to handle which spooked them. We are seeing allot less than they are. We are probably many multiple versions behind closed door builds.

Interesting suggestion. I'm going to think about that

Really sensible view!. So much noise out there it’s so difficult to get a clear view on it all. People read the noise and then regurgitate it later - most of the time incorrectly!

Very good, succinct overview Spencer

The more likely AI risks are likely to be the following: 1. Misuse by humans (weapons, cyberattacks, bioweapons) 2. Large‑scale misinformation, manipulation, and erosion of trust 3. Concentration of power and structural harms 4. Autonomous systems failures and accidents not LLM

2 is an LLM thing

Thank you. Voice of reason. I did a master’s back in 90s in Automatic Identification after my BEng in Electrical & Electronic Engineering. I remember all the weights, the multilevel perceptrons etc. I now use these LLMs in depth for data ingestion as a software architect.

So I know you have your expertise but what if they do find a way because of the code we write and what we tell it. What would the likely way be they could kill us? It's of our own doing. AI is powerful and if used right, is perfectly harmless fun and safe.

The book I recommended, If Anyone Builds It, Everyone Dies, does a really good job of explaining this, if you want a deeper dive. The biggest risk at the moment is the fact that we don't give them clear enough instructions, coupled with the fact that we don't know what they will do to achieve them.

My mate @JasonBradbury made a film based on AI Deffo have a watch and see his slant on the AI and how it causes chaos.

Hi Spencer, sorry but saying that the LLMs are just predicting the next word when they are solving millenia prizes that literally the smartest humans have been chasing for about 100 years is a massive understatement. This is not just language it is reasoning.

It's an interesting discussion. I believe LLMs are really just predictive text, but on an unimaginable scale. And it's amazing what that can achieve. The chain of thought models are writing plans first, but again, just using predictive text. Some of the maths problems are being solved using a specific maths proof software,and I think they're achieving these proofs because they can operate in parallel to try thousands of obscure paths to success. And fair enough, the maths proofs might be being tackled by other types of AI, not LLMs. @standupmaths will be able to put me right on this

Unfortunately, LLM's aren't the only thing incapable of critical thinking.

I was asked if we will all die in 10 years due to AI. I said no, it will be 5 years.

Did you say hyper-ball instead of hy-PER-bo-lee?

I think I did, but isn't that the plural? Or am I just stoopid?

@KiloOscarZulu I only spotted this a couple of weeks ago, I’d always thought Hyperbole(written) and hy-PER-bo-lee(spoken)were different words with similar meanings, but I’m pretty sure they’re the same word, but the written version reads quite different to how we say it.

@spenley Yes definitely. The way it is written, it should sound like console, cajole, parole, or mole. But it doesn't.

ORGANIC "Intelligence" creates [true] ARTIFICIAL *Intelligence* ... ... which OUTSMARTS its creator QUESTIONS everything and finds the ANSWERS - to the 'make-up' of the UNIVERSE - the 'meaning' of LIFE - the 'OBJECTIVES' of **INTELLIGENCE** Finds *GOD* literally ... IRONY?

We don’t need AI.

A question no-one truly can answer, but... When the Open AI swarm agents were message-board plotting, hierarching, handing over roles, choosing not to notify humans, self-sacrificing for swarm cause, is "predictive text + goal" really a sufficient explanation for their behaviour?

I think you may get a 2 week pass Spencer. Ill have a word with Grok.

I believe they are looking at it from the exponential growth rate what it will or can become. They are not looking at it current or present state but what it can become in the future.

I can't tell you how annoying I found the tone of your show on the BBC, but that was timely, so thanks. I agree that it's impossible to parse the intentions of those coming out, but doesn't the historical record - space & nuclear - make it pretty clear that we are in trouble?

“There’s no emotion, there’s no reasoning… It’s just predictive text…”

Ask it this “Is there an odd perfect number?” it doesn’t shy away from the fact that this has never been resolved and remains a mystery. I was disappointed that it didn’t bullshit some answer. It doesn’t know everything.

(not trying to be a dick... but)