OK, I changed it back to the original headline from the site: "Are brain-scanning hard hats a safety dream—or a privacy nightmare?"
3,428 karma · joined June 6, 2011
Good quality article. Provides a nice summary of the issues, and also what kind of improvement is expected, given certain changes.
https://sciencefictionspod.substack.com/p/episode-91-entangl...
But even in the most conservative reading of the evidence, there is a lot of amazing ecology going on here. And the economic analogy seems apt. I buy her argument that it's worth investing a lot more to explore these interactions.
Stepping back - evolution has had a vastly greater amount of time to deal with economic & game theoretic situations than we have. Learning more about what type of strategies have survived the test of time could be very important. Especially as we are currently at the beginning of an explosive change in our own ecosystem (and I ain't talking about biodiversity or climate change...)
https://www.nytimes.com/2026/10/13/opinion/homelessness-unho...
And if you deem creating an account using a fake email is too high a price to pay for quality human journalism:
Calling this a "fix" is ambitious, since these measures don't address the long-term issues. But these do seem like sensible measures with a big positive impact. It demonstrates that there's a lot we can do that doesn't require radical change.
And if you deem creating an account using a fake email is too high a price to pay for quality human journalism: https://archive.is/VlFlc (also you will miss the videos, which in this case are worth the price of admission, IMHO)
paper: >Generation of a transgenic cephalopod https://www.biorxiv.org/content/10.64898/2026.09.23.753964v1
I must say I am still amazed and mystified at how cephalopods to camouflage so well, especially because they must mimic a perspective much different from the one they have access to from their eyes (or skin).
>If a human did this, it would be an instant Fields Medal, no questions asked.
- Alex Kontorovich, chair of the department of mathematics at Rutgers
https://www.nytimes.com/2026/10/07/science/microbiome-evolut...
Interesting. This increased my credence that newborns are conscious / sentient (which had been somewhat low).
Interesting to consider how to properly draw the line on when to allow this.
It argues that psychological counseling should be a key step, and is not in Canada. If true, the point seems strong.
Please, someone correct me in my thinking, since I should hope this Princeton professor of math is more intelligent than I:
He asserts:
Today ML results are not "creative" but come from:
>a search guided by ideas developed by human mathematicians, but far more exhaustive than any search they could carry out unaided.
Therefore:
>There are many risks associated with the uses and misuses of AI. The extinction of mathematics as a human endeavor is not among them.
How does behavior today guard against risks tomorrow? Has he considered the trend from imagenet to today, or from LLMs to today?
Are there any hurdles that prevent against ML models being "creative" (especially as creativity can be thought of as a search process)?
A couple of years ago the goalposts were something like:
"It sounds like the LLM is saying something meaningfull, but it actually doesn't"
Today the goalpost is apparently:
"LLM solved one of the hardest outstanding math problems, but it did it by assuming a constraint. We now, post-hoc, think that was a bad constraint to allow"
Now, I'm not going to assert that human math is obsolete based on this. But asserting that there is no risk of it becoming obsolete based on this evidence does not seem to me to be "intelligent" let alone worthy of ivy league tenure.
I'm assuming this is the same guy who wrote the Nexus series, which was amazing BTW.
A Way Out of the A.I. Arms Race? https://www.nytimes.com/2026/09/30/world/europe/ai-arms-race...
For many reasons, Betteridge's law applies to the NYT piece.
But nevertheless, it worth encouraging all efforts to address the systems dynamic.
and the core of their recommendation, transparency, is key towards improving the quality of the outcome. The more open, broadly distributed, and cheap ML gets, the better.
This video highlights why. He hits a number of key points about what we need to do:
1) develop better models (e.g. ones focused on program synthesis w/ verification) 2) Reduce the concentration of economic value (e.g. promote open source models). 3) Avoid deliberately creating AGI. The benefits of ML without it are still massive. Doubt this is feasible game theoretically long-term. But worth promoting, since it's valuable even if we just slow things a bit.
I'm happy he's getting some attention, and wish he got more.
There needs to be a dramatic acceleration in profit. Yet, the headwinds seem to be increasing: OW models, Infrastructure-building limitations, liability...
https://en.wikipedia.org/wiki/The_Dictator%27s_Handbook
Autocrats succeed to the degree they can gain power while having to please as few other people as possible.
Only working together can we build a cheap toaster: https://en.wikipedia.org/wiki/Thomas_Thwaites_(designer)
I buy the Sapiens argument that our dominance is due to being more social, than being "smart" individually.
On the other hand big dysfunctional group can hobble the effectiveness of even its best members (I bet I don't need to cite an example for you).
But the opportunity is far less constrained than this. The space of human societies is limited to agents who all have similar structures and relaitvely fixed "selves" and goals. ML agents can be as diverse as we like, can change their own nature on a dime, and can have any type of self "identify" or goals imaginable. They are not constrained by any "human nature" to only cooperate in certain ways, but can dynamically pursue any strategy at all.
Perhaps there is a natural convergence or local equilibrium with more stable and uniform identities. But I'd bet on the other side - that there will be an insanely diverse and dynamic society that will be able to coordinate in ways that we can neither compete with nor conceive of. Cryptic message boards are just the first drop in the deluge.
I'm skeptical this bill is actually a positive step in the overall campaign. That said, I think the more discussion about the concept, the better.
Paper: https://www.econometricsociety.org/publications/econometrica...
This seems on the same spectrum with research showing the biggest predictor of success in children is social connections with the more affluent: https://www.sciencenews.org/article/friendships-rich-people-...
I generally buy this, and think the class or networkworking view is a much more productive one to combat inequality, vs. tribal identity distinctions.
At some level of course, we are a social species and those who acquire better connections will always outperform, often for reasons we want to endorse (e.g. they will create better connections among ideas).
But networks can also serve as self-perpetuating systems that impede their intended function. e.g. base nepotism, or Science advancing "one death at a time." There's a lot of opportunity to reduce these effects.
While this seems overhyped in terms of it's applicability (wraps around DNA vs. actually editing it), it's still really cool.
The ancestor of CRISPR was developed by viruses over 4 billion years ago.
It's crazy how early life arose (the crust only formed 4.5 billion years ago). And one can only imagine how much insane stuff happened that we can never glimpse.
Gift link: https://www.nytimes.com/2026/09/18/science/tilcayo-new-cat-s...
>Liminal Capital, an investment firm, estimated in August that about a third of jobs appear to be more exposed to substitution than augmentation by A.I. In those occupations, new hires among workers ages 22 to 25 have fallen by a third since 2021, even as firings have barely budged.
Their paper: https://www.liminal-capital.com/documents/liminal-ai-jobs-im...
I'm skeptical of the quality/bias of this, given the source. That said, the story makes sense to me at a high level. Even without the current ML expansion, general trends in automation and maturity & consolidation of the tech sector seemed to be generating a similar effect.
note this was GPT-5.2