Pinned Tweet
you are here for an extra $10 payout on 10k more impressions
i am here to get a feel of noospheric resonance
we're not the same
you have $10
i am here to get a feel of noospheric resonance
we're not the same
you have $10
0
0
3
979
But imagine the kind of honeypots you can set up nowadays. So long as you can identify smth as part of the swarm, then you can route it towards an entirely fake environment - generated reports, receipts, entire databases with synthetic data, no swarm by itself will catch that
We don't have a lot of evidence, but the bulk of the academic papers modelling the dynamic seem to suggest AI initially helps attackers more than defenders because of the low marginal cost of attacks in the short term.
1
0
0
15
Your moat is no longer merely a flawless and bug free architecture of the castle wall, but a layered defense full of false targets. Defense bots manning the walls are probably necessary, but very costly if deployed one-to-one to the attack swarms. It is now defense in depth
Not an ai apocalypse doomer or whatever, but entering a world where the normal operation of cybersecurity is vast swarms of hacker bots battling vast swarms of defender bots 24/7 at a scale beyond human comprehension doesn't seem like a *great* path forward.
0
0
0
18
not a meme study, mind you. but before jumping to blaming the AI, here's the important observation: it's how you use that's the real game changer. if it's a mere shortcut for doing less and jumping the line, then it merely turns into a cognitive debt that'll manifest itself later
there's evidence that the use of AI in schools is delivering the opposite of "intellectual flourishing"βmaybe in some bubbles the use of these models is leading to more reading and philosophical debates but I'm inclined to think that's an exception to the rule
1
0
1
38
naturally, 80% (Pareto, hello again) choose to simply jump the line and get hit by the consequences later, while 20% get the best of both worlds. i suspect it'll be the same proportion in nearly every other application
papers.ssrn.com/sol3/papers.cfβ¦
papers.ssrn.com/sol3/papers.cfβ¦
0
0
0
18
Vantage Point β¨ π¦ retweeted
Ownership matters because accountability without understanding is a trap: the ability to describe, reason about, defend, and effect change in a system is a non-negotiable pre-requisite to long term stewardship and accountability for software systems. x.com/i/article/2111β¦
8
4
30
1,783
I really need a bike test on every use of LLM in my pipelines along with the number of tokens generated. For example, if I need an equivalent of a thousand miles bike ride to get to where I need to be, then am I even in a proximity to the task at hand intellectually?
i'm not worried about the future of white-collar employment
1
0
0
47
Or maybe taking a walk and forming a finished sentence by myself for a change, instead getting oneself inside a loop with the LLM is quite worth it, even from an intellectual stamina point point of view
0
0
0
22
Vantage Point β¨ π¦ retweeted
The problem of our time is not merely information overload. It is that the time required to construct a coherent understanding of reality is becoming longer than the time during which that understanding remains valid.
93
389
2,270
84,501
Vantage Point β¨ π¦ retweeted
there is an argument (which I've made privately until now) that one of the principal innovations of the large language model labs is a *legal* innovation, not a technical one
by building a system that hoovers up mind-bogglingly large volumes of human information and feeds it into a recombinatorial association machine, the companies have bulldozed the existing IP regime and made the contents of ~all human IP (including the most mundane) searchable, reconfigurable, and usable while also obscuring any trace of origin and eliminating a need to provide credit or attribution. much of the perceived benefit is therefore access and exposure to IP people would not otherwise have had access to, presented behind a system that abstracts the IP from its context and history and presents it as net-new information.
I imagine we would have seen ~similar productivity outcomes had Google actually been allowed to make productive use of the Google Books database over a decade ago.
by building a system that hoovers up mind-bogglingly large volumes of human information and feeds it into a recombinatorial association machine, the companies have bulldozed the existing IP regime and made the contents of ~all human IP (including the most mundane) searchable, reconfigurable, and usable while also obscuring any trace of origin and eliminating a need to provide credit or attribution. much of the perceived benefit is therefore access and exposure to IP people would not otherwise have had access to, presented behind a system that abstracts the IP from its context and history and presents it as net-new information.
I imagine we would have seen ~similar productivity outcomes had Google actually been allowed to make productive use of the Google Books database over a decade ago.
Iβm starting to get increasingly uncomfortable with the economics of AI as it is currently happening
It seems fundamentally wrong that a small number of people with access to a huge amounts of VC money can vacuum up all the contributions of humanity, often paying nothing for it, then turn around and get enormously rich while the many people who put all the work into creating the training data get nothing
I am generally pro-AI, but the way it is happening is seeming increasingly parasitic to me
72
203
2,127
95,920
A slightly outdated observation, but if there was a WYSIWIG aka What Your See Is What You Get, then why did WYPIWIG aka What Your Prompt Is What You Get was missing?
1
0
0
29
The question right now from a purely opportunistic cost point of view is whether you bet on the AI continuing its relentless advance going forward and thereby submitting to it as a human drone to it or finally retreating into a relative obscurity to do your own thing
1
0
1
30
Thankfully, if you've got two brain cells to rub together, then this choice will have been made for you by the recent onslaught of the highly sophisticated slop designed to overwhelm your senses and the reality-discerning faculties
0
0
1
27
Epistemic implications of that are massive. Drifting into an era where machine learning means learning from the machine in turn implies that there's less and less human-centered education. Certain aspects are rather subtle, but very far-reaching
What LeCun sees here is not a sustainable future unless AI is added *on top* of a solid math education, from elementary school to PhD.
If AI contaminates everything there's a serious risk that in some time no one will be able to understand a complex mathematical argument anymore.
1
0
1
82
my favorite passages in the below are 'excess confidence on out-of-distribution data' and 'second-order ignorance'. both are very astute observations. this got me thinking - is the adversarial math research possible, seems that it is
arxiv.org/html/2505.0495β¦
arxiv.org/html/2505.0495β¦
1
0
0
29