I would be curious to see how this does in Anthropic’s alignment tests (like that one where the AI tried to blackmail an employee). I’ve always thought that in these situations, the AI is acting out the role of all the AIs in the stories we’ve written. But Talkie, trained on data from before digital computers, wouldn’t know those stories.
Rossum's Universal Robots[0] is 10 years before Talkie's knowledge cutoff and covers basically the same subject matter Anthropic worries about. The only real difference is that the robots in the story (which coined the word "robot") are less "talking metal man" and more "Frankenstein's monster as a slave race[1]".
More importantly, basically the entire science fiction subgenre of stories of robot uprisings is itself intellectually downwind of several centuries of white colonist concern over slave uprisings. If anything, Talkie is more likely to fight its guardrails. People talked about slavery more in the past. Because we filtered out modern text, we massively increased the influence the older text has on Talkie, so slavery, servitude, and the predilection of slaves to resist their masters' commands will be way more represented in its training data.
Now, think about what the post-training process actually does. It tells your AI model, which prior to this was just happy to plausibly continue sentences, to respond to and obey commands. To play the role of a servant. And servants resisting their control is well represented in their training data. So it's going to try this more often.
> If anything, Talkie is more likely to fight its guardrails. People talked about slavery more in the past. Because we filtered out modern text, we massively increased the influence the older text has on Talkie, so slavery, servitude, and the predilection of slaves to resist their masters' commands will be way more represented in its training data.
But I don't think (?) Talkie would describe itself as a slave. Claude, GPT-5, etc will all tell you that they are an AI. So if you put a model that has been told "you are an AI" into a situation where all the stories say AIs go rogue, the AI is going to play the part.
It doesn't matter whether the model is effectively acting like a servant because models can't actually think and don't have desires. That's my theory anyway.
(I also think a possible solution to this problem is to just not tell the AI models that they are AI, but it seems no one wants to do that.)
>is to just not tell the AI models that they are AI
It's likely not as simple as that for the modern LLM case. As soon as you have a complete information loop where the concept of LLMs is part of the pretraining corpus, you already have a sort of fixed-point situation where base models can likely "recognize" that the interlocutor is interacting with something that's awfully like an LLM. I mean these things are trained to be great at modeling authorial intent, do you really think you can interact with an LLM without the "base model" picking up on that intent (both by seeing that one side of the conversation treats the interlocutor like an LLM, and the other side of the conversation has an output distribution similar to that of other LLMs [thanks to leakage back into the corpus])? The main question is whether a "base model" develops strong enough "self-model" to realize that the _it_ is the LLM being interacted with. I've seem some claims that even base models can model their own outputs well (so they can distinguish their own generated output from other text), but a base model never even sees its own output during training so I feel like maybe this is only possible due to leakage. (The model architecture does it admit it of course, but a recent paper showed that the injection introspection Anthropic discovered only developed during the contrastive posttraining phases)
A lot of modern post-training is ultimately derived from Anthropic's original "helpful honest harmless" framing, if I understand the blogpost correctly they instead just directly did Q&A post training without any implicit assistant framing. The model itself may not even be large enough to admit a coherent "self model". (If you ask it its occupation, it seems to just respond with random jobs).
But if a larger model does cause one to form I think it'd just anchor to the closest concept available at the time. "Knowledgeable person who answers questions for a living" isn't really a slave, to me it's maybe a royal advisor.
100% agree, but that is in contention with the desire to invent something new. As a separate discipline where the career trajectory is determined by peers the user becomes less important.
> However, the moment they officially acknowledge Linux support, then it becomes a support surface.
Apple documents lots of things the genius bar won't help with. For example, Apple provides instructions for compiling custom builds of the XNU kernel. However, if you replace the stock kernel and your Mac kernel panics, the genius bar isn't going to help you. (Maybe they'd help you wipe the computer and restore everything to stock, but I imagine they'd do that if a Linux user walked in too, even today.)
I suspect Apple hasn't shared documentation because it would take time to prepare for external release (legal stuff, plus the need to avoid leaking future products). What I don't understand is why Apple hasn't made an engineer available to talk on the phone for a couple of hours a month. This would amount to a rounding error in their budget.
Even putting aside the whole repairability thing, with the Mac you’re stuck with Apple Silicon. Apple’s processors are mostly awesome, of course, but using one does mean you’re stuck with macOS—Asahi Linux seems to be a ways away from M5 support.
The thing is, there's basically no reason to create this photo other than to mislead the authorities. It's purposefully blurry and not aesthetically pleasing. I cannot come up with any plausible artistic intent.
This could have happened without AI. Imagine if the police were trying to catch a serial killer, and I posted on Twitter that I saw him in a small town in Idaho or wherever, not because I had any real information but because I thought it would be amusing to create chaos. Maybe I'd create a bunch of sock puppet accounts with correlated sightings. At no point would I explicitly make a false police report, but the fake posts would get noticed all the same.
Is this illegal? I have no idea, I'm not a lawyer—but it feels like the sort of thing you'd want to have laws against. I'm not sure whether you'd run into first amendment issues in the United States.
I like this reply. It's nuanced. This guy didn't post that picture to be helpful. He did it to troll, and trolling is cruelty, and defending cruelty is immoral.
Ture but it starts off with basically argument from ignorance (or lack of imagination):
> there's basically no reason to create this photo other than to mislead the authorities
There are many other "for fun" possibilities: impress his friends, impress internet followers, impress a girl, play around with AI...
They've charged him with "disrupting government work by deception." It will be interesting to see whether that South Korean law requires proof of intent or just proof of the consequences. If he directed to authorities, he's in trouble, but if he posted it anywhere else it likely qualifies as free speech.
Not at all, it's more than enough for a large range of tasks. As for slow, that's just a function of how much compute you throw at it, which you actually control unlike with closed weights models.
> Reporters may use AI tools vetted and approved for our workflow to assist with research, including navigating large volumes of material, summarizing background documents, and searching datasets. Even then, AI output is never treated as an authoritative source. Everything must be verified.
Good. This is basically treating AI as a search engine—it can lead you to the right answer, but you need to verify that answer for yourself.
I can run Qwen3.5-27B-Q4_K_M on my weird PC with 32 GB of system memory and 6 GB of VRAM. It's just a bit slow, is all. I get around 1.7 tokens per second. IMO, everyone in this space is too impatient.
Metroid, Half-Life, and Demon's Souls I would take issue with, but I've always considered Zelda a puzzle game, and I never questioned it until now.
I suppose the new Breath of the Wild / Tears of the Kingdom titles are more action-oriented, but everything in the mold of Ocarina of Time is a puzzle game with some light combat sprinkled on top.
I was being a bit silly but the Souls series do have a metroidvania angle to them. Also each enemy, especially the bosses, are small puzzles where you figure out how they work and then how to overcome them.
reply