TechTakes

Wow, the story is so much more troubling when you lie about it!

This tied into a hypothesis I had about emergent intelligence and awareness, so I probed further, and realized the model was completely unable to ascertain its current temporal context, aside from running a code-based query to see what time it is. Its awareness - entirely prompt-based - was extremely limited and, therefore, would have little to no ability to defend against an attack on that fundamental awareness.

How many times are AI people going to re-learn that LLMs don't have "awareness" or "reasnloning' in a sense humans would find meaningful?

i don't understand the "safety" angle here. if chatgpt can output authoritatively-looking sentence-shaped string about pipebombs, then it's only because similar content about pipebombs is already available on wide open internet. if model is closed, then at worst they would have to monitor its use (not like google blocks any similar information from showing up). if model is open, then no safeguards make sense in the first place. i guess it's more about legal liability for openai? now they can ignore it with all these bills about "ai safety" gone (for now)

"The world is finite and kids are infinite, especially African kids." Jfc. Anyway goes to show just how white supremacist the whole "save the children" idea is.

I know the only intended message there is "I am a big racist", but what kind of dumb fuck adage is "the word is finite, kids are infinite". You're not even trying mother fucker

Is it a crime to enjoy a succulent Chinese AI?

I've previously discussed the concept of model collapse, and how feeding synthetic data (training data created by an AI, rather than a human) to an AI model can end up teaching it bad habits, but it seems that DeepSeek succeeded in training its models using generative data, but specifically for subjects (to quote GeekWire's Jon Turow) "...like mathematics where correctness is unambiguous,"

That sound you hear is me pressing F to doubt. Checking the correctness of mathematics written as prose interspersed with equations is, shall we say, not easy to automate.

OpenAI can't simply "add on" DeepSeek to its models, if not just for the optics. It would be a concession. An admittal that it slipped and needs to catch up, and not to its main rival...

I actually disagree here. I think Ed underestimates how craven and dishonest these people are. I expect they'll try to quietly integrate any efficiency improvements they can get from it and bluster through any investor questions about it. Their hope at this point has to be that more hardware is still better and that scaling is still gonna be the thing to make fetch happen. This again isn't a revolutionary new structure, even if it is a significant improvement over anything Saltman and co have been doing.

https://www.wheresyoured.at/deep-impact/ And the companion piece on his blog.

What I didn't wager was that, potentially, nobody was trying. My mistake was — if you can believe this — being too generous to the AI companies, assuming that they didn’t pursue efficiency because they couldn’t, and not because they couldn’t be bothered.

This isn't about China — it's so much fucking easier if we let it be about China — it's about how the American tech industry is incurious, lazy, entitled, directionless and irresponsible. OpenAi and Anthropic are the antithesis of Silicon Valley. They are incumbents, public companies wearing startup suits, unwilling to take on real challenges, more focused on optics and marketing than they are on solving problems, even the problems that they themselves created with their large language models.

It's wild that Yudkowsky saw a binary choice of "nuclear holocaust" and "superintelligence" and chose "nuclear holocaust" in the first place.