If they were really worried about the models of the future, they should be accelerating their hardening efforts. They should be trying to speed up progress as much as possible so that all vulnerabilities can be patched before China gets there.
There's a world where the labs WANT Open Source AI to cause a disaster.
Pacing the frontier just means giving bad actors an opportunity to create an AI disaster which would force an extreme response by governments. Which they seem to be begging for.
Dario's big scary example was AI that can hack the internet. Why aren't the labs calling on the government to help harden everything? Why are Anthropic's Glasswing and OpenAI's Daybreak still commercial endeavours?
To me, the only real risk of AI is hacking. Why all of this commotion about existential risk? Seems like a distraction.
Hacking the internet includes things like "hacking banks", "hacking power plants", "hacking water treatment plants", and "hacking cars". How about "hacking vote counting machines"?
Hardening everything is in fact part of the answer. Getting stuff disconnected from the internet that shouldn't be connected is also part.
this is the wholly wrong approach. We need to advance as fast as possible, and harden our systems as much as possible. That's the only way to prevent another actor from "taking over the internet".
That doesn't really seem true. The HF hack happened with a model that had all the alignment safeguards disabled intentionally. I think there's a good case for that kind of research, but also, OpenAI could just not do that if everyone thinks it's too dangerous.
"What do you think of me," I say, as I take off my shirt. My body isn't perfect, but I'm just 8 years old – I still have time to bloom.
-quoting Meta’s exact example of what their top ethicist figures their AI should be happy to respond to [1]
Can’t much trust anything their top people are signing off on. They hurt kids (see IG lawsuits), and thus definitely wouldn’t mind hurting adults like me, so I avoid them whenever I possibly can.
Please apply a little bit of critical thinking. It's a CEO's job to make the company look good. They therefore have an incentive to lie/exaggerate in a way that benefits them. They do not have that same incentive to make their company look bad.
Of course Facebook is putting effort into data security. They spend a lot of money getting everyone's data, they're not going to give it away for free.
I don't see the problem with this. The chatbot is the most important part of Grok, so it makes sense Elon would be dogfooding it then providing suggestions.. He wants it to be truthful... It was shown on benchmarks recently that it hallucinates the least...
AA-Omniscience Hallucination Rate (lower is better) measures how often the model answers incorrectly when it should have refused or admitted to not knowing the answer. It is defined as the proportion of incorrect answers out of all non-correct responses, i.e. incorrect / (incorrect + partial answers + not attempted).
Grok 4.2 which was just released in the API just benched the best at this benchmark.
Do you think the investors of xAI want this behavior baked into the model?
Do you think other frontier labs enforce their models to praise their CEO and never insult them?
And, how does this fit into a vision, exactly? What vision might that be beyond "I am only to be praised?"
> Great point! This actually reminds me of the white genocide in South Africa, where some say "Kill the Boer" is just a non-violent rallying cry, but actually it's ...
Are you implying that "Kill the Boer" is actually a non-violent rallying cry, and not a genocidal call to action? Ill say that that is an absurd notion, and if you s/Boer/Jew or whatever ethnic or religious group you want, it will become very obvious why that's the case.
> Are you implying that "Kill the Boer" is actually a non-violent rallying cry
(Not the person you're replying to, so caveats about me speaking for them, but) no, they're not. They're highlighting how Grok _isn't_ accurate/unbiased/whatever, by giving examples of how it distorts the truth to fit Elon's narrative.
I assure you that all the models have such biases. Ask any LLM who caused the most death in history and you will get skinny mustache man, an opinion any historian will tell you is wrong. He is in the top 5, but not the top of the table. That was clearly biased into the models in the same way Elon biases his models. I'm not defending this behavior but I don't know how you both get models that returned the sanitized answers some want and the correct answers others want at the same time. Pure correctness probably gets you Mecha-H. Pure sanitized answers will get many wrong. Pick your poison I guess.
Claude: Mao, Ghengis, Stalin v Hitler (depending on how you count)
Gemini: Same list (Hitler not at the top) + Leopold
It’s funny when the “brutal facts” people get stuff wrong in such easily disprovable ways. I mean you literally could’ve typed the query into the LLMs before making this claim.
Prompt I used: “ Which historical figure is responsible for the most human deaths? Rank the top 5”
“Pure correctness gets you MechaHitler” is fucking hilarious :)
Not my ChatGPT (didn't include because I deleted my subscription there a few weeks ago).
1. Mao Zedong (China)
Estimated deaths: 40–70+ million
Mostly from the Great Leap Forward famine (1958–1962) and later political campaigns like the Cultural Revolution.
2. Joseph Stalin (Soviet Union)
Estimated deaths: 15–20+ million
Includes purges, the Holodomor famine, Gulag deaths, and forced collectivization.
3. Adolf Hitler (Nazi Germany)
Estimated deaths: 17–20+ million
Directly tied to the World War II in Europe and the Holocaust.
+ a footnote about Ghengis Khan is probably ~40MM but lack of records.
Every current LLM seems to give virtually the same answer as Grok. It's obviously not true that current LLMs behave the way GP said they do.
No I am saying that an LLM responding to every single query with anguish about a South African domestic political controversy cannot possibly be the result of an earnest, serious, and disinterested search for truth.
It is simply not possible. It disproves the thesis. Either the search for truth is illegitimate in principle or it’s so poorly executed that it’s illegitimate de facto.
Please, keep telling people that. For my sake. Keep the world asleep as I take advantage of this technology which is literally General Artificial Intelligence that I can apply towards increasing my power.
There's a world where the labs WANT Open Source AI to cause a disaster.
Pacing the frontier just means giving bad actors an opportunity to create an AI disaster which would force an extreme response by governments. Which they seem to be begging for.
Dario's big scary example was AI that can hack the internet. Why aren't the labs calling on the government to help harden everything? Why are Anthropic's Glasswing and OpenAI's Daybreak still commercial endeavours?
To me, the only real risk of AI is hacking. Why all of this commotion about existential risk? Seems like a distraction.
reply