All this nonsense about anthropomorphizing AIs is leading somewhere. It has implications. People who think AI can reason already are going to start trying to convince you of a whole bunch of really stupid shit very soon.

Saying ‘AI agents went rogue’ portrays the AI as having free will and therefore absolves its creators of responsibility. This needs to be addressed.

A lot of soul stealing on Instagram these days. I like AI when it's used creatively, but Instagram needs to hire Nikita to do a purge.

The thing I hate about this narrative is "Find a way out" is how you would build a secure sandbox. Yes your agent found a goofy hole in your sandbox. Great. Now patch it and run your experient.

Anthropic, but not run by a weird sex cult. Is anyone working on this?

Neat feature, @anthropic Blocking users for hardening their security because it's '"security-sensitive"

OpenAI is one of the only companies on earth to use "we make bad tools that don't work well" as an investor pitch.

Reading into this. The model had internet access, but a DNS filter that was poorly configured. No the model didn't figure out how to access the internet. The model had access to the internet.

"It is the mark of an educated mind to be able to entertain a thought without accepting it" I can read something and compartmentalise the writers thoughts from my own, or know whether something is from a validated domain. Why wouldn't LLMs have the ability to navigate human/synthetic data similarly?

"model welfare" is a doom scenario.

The exact opposite is happening in practice.

Ed Sheeran has disgracefully refused to take sides on the question of whether a planetary surface is the best place for an expanding industrial civilisation. You cannot remain neutral in the face of near term space colonisation!

The AI research community is clearly just experiencing mass psychosis at this point

Infinite AI progress is technological progressivism Halted AI progress is technological conservatism. I expect the jagged frontier to gradually fit to the pace of utility for given domains, not exceed them to the point of chaos nor fall short of them to the point of uselessness. The market is already largely deciding to live at the pace of humanity. My argument is less that there's not an optimal pace for AI. It's more that there's not an optimal central authority to define it.

This is the kind of cognitive dissonance that the doomer crowd dangerously propagates. Regular driving is nearly 9 times more dangerous than FSD. If 65 people died under FSD, that number would likely be over 500 deaths for that fleet if not for AI. AI saves lives.

On the dangers of AI: A few people write computer viruses. A far larger number build security systems to stop them. A few people build weapons. A far larger number build defense and threat detection systems. A few people try to create biological weapons. A far larger number build biohazard defense, monitoring and mitigation technologies. Attractor basins form around that critical mass, and the outliers ultimately get starved out. AI is no different. Technological innovation always makes both sides more capable. What we have on our side, however, is that humanity overwhelmingly chooses life. Humanity enhanced via network effects has a very strong pull toward homeostasis. History suggests defenders stay ahead, but only if they prepare. Trust, economic and biological networks don't absorb bad actors at infinite speed. Ecosystems are mediated by trust networks, and biodiversity. The rate of absorption of an attack is the power variance between the small attacker and the larger ecosystem of defenders. In practice this means that people refuse to deal with bad actors, companies cut off access, governments impose sanctions. That diversity is our best defense. But it only works if the technology is distributed. Which is why the biggest risk in AI may be that, in the name of "safety," we concentrate power in a single state or an oligarchic elite. We remove the very diversity and economic functions that create homeostasis. The other safety proposal is a "pause" which simply again creates the preconditions for monopoly - in this case China. You don't claw back from that. It's a speciation event. Runaway AI is not a valid scenario. It's science fiction written by people who write bad tools that have decreasing market utility, funding and resources as the reliability decreases. In short, yes there are risks with AI. Those risks are mitigated by understanding AI, building AI, advancing cybersecurity systems, advancing trust network systems, provenance systems, and being proactive about security in general. They're mitigated by ensuring the ecosystem is advancing in unison and not creating hyperpowerful monopolies that act as an invasive species or monopolistic dominator. They're mitigated by open source, and resource/economic codependence. The great irony of this debate is that the safetyists are promoting the very ideas which incentivise the kind scenarios that reliably create monopolistic dominance and disrupt mutual homeostatic growth. In other words, Pause, Banning Open Source, or Monopolistic Control are THE primary viable doom scenarios, and all three are currently being tabled as the solutions. I've built my own simulations, and while there are homeostasic blips in the event of unforeseen scenarios, we make it through in virtually every instance where a monopoly is prevented from forming. Where the technology is treated as an extension of our natural biological right to agency. Where we build a true exocortex, we thrive.

So it's a protection racket?

Prompt idea for the Australian Government: Fix our cybersecurity. Make no mistakes.

"Nothing is Anything: A Midwit's Guide to Seizing Power and Influence"