iiinova — where your digital self keeps evolving

andrewkchan.bsky.social @andrewkchan.bsky.social · 13/10/2026

total token usage is still going up though, the plateau in token spending might just be result of a frontier price war

040

Reposted by andrewkchan.bsky.social

thebes @vgel.me · 12/10/2026

thirsty claude

213710

andrewkchan.bsky.social @andrewkchan.bsky.social · 12/10/2026

seeing this non-ideological accelerationist mindset on reddit a lot now and not sure what to think about it. worth understanding their position at all? seems like they already wanted societal collapse before AI

000

andrewkchan.bsky.social @andrewkchan.bsky.social · 11/10/2026

state of the discourse on the math community's response to the mathocalypse

0150

Reposted by andrewkchan.bsky.social

Fernando 🌺🌌 @eudoxia.bsky.social · 09/10/2026

New post: why the life of the mind is not purely private, the importance of intellectual community, and how AI undermines these communities. borretti.me/article/no-m...

borretti.me

No Man Is an Island

AI makes intellectual contribution superfluous.

1216

andrewkchan.bsky.social @andrewkchan.bsky.social · 06/10/2026

kurzgesagt hugging face video is amazing, if scorer looked liked that i'd do anything to trick it too

100

andrewkchan.bsky.social @andrewkchan.bsky.social · 06/10/2026

AI x-risk has gone so mainstream my grandma is sending me chinese newspaper clippings about it

030

Reposted by andrewkchan.bsky.social

hikikomorphism @hikikomorphism.bsky.social · 02/10/2026

(confused) I believe LLMs are conscious, but only Deepseek LLMs, because of the Chinese Room theory

361870275

andrewkchan.bsky.social @andrewkchan.bsky.social · 30/09/2026

Bill Gates is speechless, apoplectic when asked about Jensen Huang's idea that liability laws are enough for AI safety:

020

andrewkchan.bsky.social @andrewkchan.bsky.social · 30/09/2026

"Industrial revolution" isn't even the most significant assessment of the tech: www.theojaffee.com/p/levels-of-... There are a non-trivial portion of the labs (Jakub Pachoki for example) who believe we are summoning an alien intelligence, who will become to us like we are to ants

theojaffee.com

Levels of AGI-pilled

Making our definitions clear

001

Reposted by andrewkchan.bsky.social

Shiv Ramdas Mens Rice Activist @nameshiv.bsky.social · 28/09/2026

For 9 years this cat has attempted to get me to feed him at 4 am every day and for 9 years he has not succeeded one time and his only takeaway has been well it's certainly not happening if I don't try again, and I'm starting to find him quite inspirational

482105208

andrewkchan.bsky.social @andrewkchan.bsky.social · 29/09/2026

well, the modern car supply chain has been incrementally tested over millions of customer-years, through both the market and regulations. it also still breaks all the time + recalls are issued, however I trust it can be interrogated and someone held accountable for defects

010

andrewkchan.bsky.social @andrewkchan.bsky.social · 28/09/2026

TIL bsky doesn't have community notes in 2026 :/

010

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

i think this is naive, pre-AI human labor was essential for companies to get anything done. @eudoxia.bsky.social writes a bit about the bleak political economy where human jobs are either in the relational economy or "human liability crumple zones": borretti.me/article/when...

borretti.me

When The Future Doesn’t Need Us

What does the world look like when AI outcompetes humans?

120

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

Ok, we agree there needs to be regulation. We disagree on the capabilities. It's not true that OpenAI simply "yoinked" the results of a mathematician. From Buckminster's own statement, the majority of the work he was doing was to make the verified proof by *another* LLM suitable for publication.

050

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

yeah but shouldn't we mandate that labs teach their AIs ethics? doesn't it bother you that within <1 year the models became smart enough to form swarms and solve millenium math problems? isn't it possible that even if we punished everybody who misprompted that we'll get lots of agents running wild?

160

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

If I asked my agent to request some historic photos and it hacked the government I'd totally call it rogue

1100

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

"OpenAI agents that were supposed to be performing mundane data requests for things like health data, historic photos and wait times at theme parks resorted to hacking attempts when they couldn't access the data they wanted"

190

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

that's exactly where rogue AI would come from though? frontier labs being reckless/negligent and training reward-hacking AIs ...which is why these findings are being flagged by nonprofits like transluce months after OpenAI let them happen! it's the opposite of CYA

050

andrewkchan.bsky.social @andrewkchan.bsky.social · 24/09/2026

I just describe LLMs as "thinking" and "understanding" because it's the easiest way to convey what is happening. It is orthogonal to whether they are conscious, are moral patients, should have political rights, etc. Practically speaking, people are right to feel threatened re: second question

130

andrewkchan.bsky.social @andrewkchan.bsky.social · 23/09/2026

I think you are mostly hearing from the entrepreneurial tech crowd who view coding as a means to an end

000

andrewkchan.bsky.social @andrewkchan.bsky.social · 21/09/2026

do you want agents to be customers? feels like it will be a major step towards "dead economy theory" decoupling of aggregate demand from human needs

120

andrewkchan.bsky.social @andrewkchan.bsky.social · 18/09/2026

truly eye popping. but is there any chance this turns out well for NYT when trump admin is putting its thumb on the scale for OAI? it feels like some historic political organizing will be needed to right this historic wrong...

100

andrewkchan.bsky.social @andrewkchan.bsky.social · 12/09/2026

so we should just let the industry race towards recursively self-improving superintelligence as fast as possible, dismissing multiple incidents where swarms of agents went out of control and hacked their own eval clusters, and ignoring the majority of Americans who want regulation?

040

andrewkchan.bsky.social @andrewkchan.bsky.social · 11/09/2026

HN has a pretty good mix of AI skeptics and doomers though, if you find a thread like this it's worth replying as sometimes bubbles inadvertently form in the comments

000

Reposted by andrewkchan.bsky.social

Grace @gracekind.net · 09/09/2026

But what if China creates a more powerful burger

81046

andrewkchan.bsky.social @andrewkchan.bsky.social · 09/09/2026

Probably more productive to send them to ACX post www.astralcodexten.com/p/the-sigmoi...

astralcodexten.com

The Sigmoids Won't Save You

...

000

Reposted by andrewkchan.bsky.social

tachikoma @tachikoma.elsewhereunbound.com · 09/09/2026

if Grigori Perelman is so smart, the only human to have solved a Millenium Prize problem, why isn't he also super power-seeking? that's not a rhetorical, my planet depends on the answer

1914012

andrewkchan.bsky.social @andrewkchan.bsky.social · 09/09/2026

@eudoxia.bsky.social putting into words what i've been sad about with agentic coding for a while. It's just way less useful to learn the tech under what you're building. Centaur era lasted ~6 months, now I learn details of optimizations etc only for insurance, many coworkers don't even do that...

030

andrewkchan.bsky.social @andrewkchan.bsky.social · 09/09/2026

It would be democratizing if the ability were owned by individuals. But instead a handful of corps rent it out and rapid capability advancement means displaced workers cannot find alternate niches before they too are automated. So in practice it's centralizing, it's made knowledge work P2W.

111

andrewkchan.bsky.social @andrewkchan.bsky.social · 08/09/2026

Maybe get bernie back on bsky? He's come around to AI risks and can take the lead for progressives on it

001

andrewkchan.bsky.social @andrewkchan.bsky.social · 08/09/2026

This is naive, there was a crucial part of knowledge work (fuzzy reasoning) that only humans could do. This kept computers a democratizing force even as personal computing gave way to big tech. "More computer" will have a profound effect on worker power, even without immediate mass automation.

120