ShashanK🤺 retweeted
A lot of people don't know the difference between a CVE and a zero day and are freaking out on finding cve trackers seeing thousands of vulnerabilities, attributing everything to AI.
Existence of a CVE is a good thing. You could see thousands of these even before on the trackers and it's a good thing if they are increasing in numbers there post AI. It means a vulnerability was publicly disclosed and catalogued and most probably has a patch available before the disclosure was made public. And with the help of AI, MORE of these are being caught and fixed in time.
Zero days are when flaws aren't known. It doesn't mean they aren't being exploited. In fact they are being exploited even more because most people don't know about the flaw and thus don't have a fix. While the bad guys know and will keep using them.
eMudhra and php cases both followed responsible disclosure protocols where mitigations were prepared. But php was marred by a leak and breaking of embargoes by humans manually. There is no evidence of AI initiated hacks in either.
Existence of a CVE is a good thing. You could see thousands of these even before on the trackers and it's a good thing if they are increasing in numbers there post AI. It means a vulnerability was publicly disclosed and catalogued and most probably has a patch available before the disclosure was made public. And with the help of AI, MORE of these are being caught and fixed in time.
Zero days are when flaws aren't known. It doesn't mean they aren't being exploited. In fact they are being exploited even more because most people don't know about the flaw and thus don't have a fix. While the bad guys know and will keep using them.
eMudhra and php cases both followed responsible disclosure protocols where mitigations were prepared. But php was marred by a leak and breaking of embargoes by humans manually. There is no evidence of AI initiated hacks in either.
3
4
67
2,941
yeah the scapegoating has just begun
started with electricity/energy, turned out that data centers actually reduce rates by distributing the infra cost
AI will overwhelm systems and will be blamed for many known institutional/human failures. (im not saying that is always orcing function for many fields
also when tech folks talk about the PR/comms problem of AI, they don't take this into account. creating a drug won't fix it. it's just getting started.
of course, it's not like they can embrace it and ask the world to tolerate. anyways, i think this is the main comms problem to be solved by the labs rather than making it loved via cute mascots, branding and what not.
for eg: IIRC even 1 year after @AndyMasley started writing about water, env impact of AI, no lab was explicitly making their own effort to dispel all these myths.
the fact that data centers would reduce electricity rates in many states should also have been known beforehand but instead it got a lot of coverage in MSM only this year and even that's not enough.
big comms gaps exposed repeatedly.
started with electricity/energy, turned out that data centers actually reduce rates by distributing the infra cost
AI will overwhelm systems and will be blamed for many known institutional/human failures. (im not saying that is always orcing function for many fields
also when tech folks talk about the PR/comms problem of AI, they don't take this into account. creating a drug won't fix it. it's just getting started.
of course, it's not like they can embrace it and ask the world to tolerate. anyways, i think this is the main comms problem to be solved by the labs rather than making it loved via cute mascots, branding and what not.
for eg: IIRC even 1 year after @AndyMasley started writing about water, env impact of AI, no lab was explicitly making their own effort to dispel all these myths.
the fact that data centers would reduce electricity rates in many states should also have been known beforehand but instead it got a lot of coverage in MSM only this year and even that's not enough.
big comms gaps exposed repeatedly.
A lot of what's happening today in security world is misattributed to AI. Both in terms of AI causing misconfig/bad code, and in terms of finding vulns.
People have forgotten meltdown, spectre, heartbleed, shellshock, log4shell and more. All pre-AI.
Why are YOU not using agents to find and patch out issues before they get exploited?
1
0
4
208
ShashanK🤺 retweeted
same thing with "personalization" btw.
easy way to spot midwits is those who keep blabbering about personalization and democratization. VCs and mediocre thinkers do this a lot.
personalization of movies, games, music blah blah blah blah.
need to break personalization down into more specific functions/vocabulary for each category and determine when they're useful.
easy way to spot midwits is those who keep blabbering about personalization and democratization. VCs and mediocre thinkers do this a lot.
personalization of movies, games, music blah blah blah blah.
need to break personalization down into more specific functions/vocabulary for each category and determine when they're useful.
mainly talking about AI.
applies to internet, mediums etc.
im guilty of it as well when thinking about the internet.
but democratization + quality controls can co-exist.
even democracies have that.
same with "free speech". whole tech twitter used that term like idiots to justify what Elon did to twitter.
opening the floodgates is not democratization.
2
1
1
75
ShashanK🤺 retweeted
also it seems like the chrome sidePanel API is finally getting properly utilized lol
this is a pretty big L for Google. for context: the tab sharing feature ships with Chrome. you can trigger it with Alt+G and it opens up in a sidebar. it's pretty neat.
but afaik they've barely promoted it, you never see all the Gemini folks here talk about it. like the gmail emo, im not sure if Gemini in chrome can do it.
and im not sure if computer use comes with it. maybe it does.
Argon just doesn't matter when even after shipping something they haven't promoted it well.
0
1
0
56
related rant: a decent chunk of STEM/explainer Youtubers have made some really bad content decisions in the last few years.
and often this deviation, bias occurs when the topic moves from an innocent sciencey topic/project to something involving an industry. this trend became more apparent to me this year.
i don't have the link but there was some tweet regarding this a few months back. basically, many STEM enthusiasts, academics, activists trad/msm stem outlets who treat science topics as just these cute projects and the moment something becomes feasible i.e approaches an industrial level, they become suspicious, treat it as evil. no concept of trade-offs, very zero-sum brains.
there's probably a better way to frame this but basically you can see this everywhere. space, environmentalism/climate change, self-driving, energy.
space example: there's a cabal of ISS romantics who see it as this symbol of great aerospace, scientific achievement and international collaboration. i read about this recently and it was an expensive mess. good symbol for normies but not something which should be seen as a benchmark to emulate.
as ISS reaches end of life many of them are mad. intl cooperation (esp with europe) is code for bureaucracy and high costs.
climate change: where do i even begin. just take US/UK policies. leftists (the climate champions) have contributed a lot towards it by restricting housing (more sprawl->more cars, more resource usage, less feasible public transit etc.) and energy infra. just 2 basic examples.
greenpeace killed a GMO rice (golden rice) which would've saved 100,000 children. 100 Nobel laureates signed a letter opposing that but no effect. shows that activists have more power than Nobel winners and letter signers are not live players. europe loves to sign letters.
greenpeace also worked hard on opposing nuclear energy everywhere. they've done more damage than climate change deniers imo.
many of these old STEM outlets, magazine will effusively talk about some impractical lab project but shudder to praise a widely deployed technology by a company.
and often this deviation, bias occurs when the topic moves from an innocent sciencey topic/project to something involving an industry. this trend became more apparent to me this year.
i don't have the link but there was some tweet regarding this a few months back. basically, many STEM enthusiasts, academics, activists trad/msm stem outlets who treat science topics as just these cute projects and the moment something becomes feasible i.e approaches an industrial level, they become suspicious, treat it as evil. no concept of trade-offs, very zero-sum brains.
there's probably a better way to frame this but basically you can see this everywhere. space, environmentalism/climate change, self-driving, energy.
space example: there's a cabal of ISS romantics who see it as this symbol of great aerospace, scientific achievement and international collaboration. i read about this recently and it was an expensive mess. good symbol for normies but not something which should be seen as a benchmark to emulate.
as ISS reaches end of life many of them are mad. intl cooperation (esp with europe) is code for bureaucracy and high costs.
climate change: where do i even begin. just take US/UK policies. leftists (the climate champions) have contributed a lot towards it by restricting housing (more sprawl->more cars, more resource usage, less feasible public transit etc.) and energy infra. just 2 basic examples.
greenpeace killed a GMO rice (golden rice) which would've saved 100,000 children. 100 Nobel laureates signed a letter opposing that but no effect. shows that activists have more power than Nobel winners and letter signers are not live players. europe loves to sign letters.
greenpeace also worked hard on opposing nuclear energy everywhere. they've done more damage than climate change deniers imo.
many of these old STEM outlets, magazine will effusively talk about some impractical lab project but shudder to praise a widely deployed technology by a company.
Actually there is something very interesting about Billy Evans. After marrying Holmes he was the head of Special Projects at Luminar Technologies, a LiDAR startup that infamously collaborated with YouTuber Mark Rober last year on autonomy safety tests comparing Luminar’s sensors ed to Luminar’s stock pumping and Tesla going down 5% in one day after they published a video where a Tesla Model 3 drove through a solid wall while claiming Tesla FSD (a separate, advanced, paid product) was operating. It was not activated in the video.
Which I didn’t think would be a relevant detail until he talked about reviving Theranos 2.0 with AI and lasers and literally bought the Texas vanity plate “LASERS”.
Luminar Technologies declared bankruptcy almost exactly when filming for You Can See Everything wrapped, after all its deals fell through and its flagship customer (Volvo) pulled out citing a “failure to meet contractual obligations.”
0
0
2
103
yeah. personally i think some of these solutions aimed at individuals is just an indicator of poor policy, ability to get things done like building some simple solar farms.
making balconies ugly isn't getting any country to energy abundance.
great post: solarchase.substack.com/p/2026-thread-…
making balconies ugly isn't getting any country to energy abundance.
great post: solarchase.substack.com/p/2026-thread-…
0
0
1
47
also it seems like the chrome sidePanel API is finally getting properly utilized lol
this is a pretty big L for Google. for context: the tab sharing feature ships with Chrome. you can trigger it with Alt+G and it opens up in a sidebar. it's pretty neat.
but afaik they've barely promoted it, you never see all the Gemini folks here talk about it. like the gmail emo, im not sure if Gemini in chrome can do it.
and im not sure if computer use comes with it. maybe it does.
Argon just doesn't matter when even after shipping something they haven't promoted it well.
0
1
0
56
this is a pretty big L for Google. for context: the tab sharing feature ships with Chrome. you can trigger it with Alt+G and it opens up in a sidebar. it's pretty neat.
but afaik they've barely promoted it, you never see all the Gemini folks here talk about it. like the gmail emo, im not sure if Gemini in chrome can do it.
and im not sure if computer use comes with it. maybe it does.
Argon just doesn't matter when even after shipping something they haven't promoted it well.
but afaik they've barely promoted it, you never see all the Gemini folks here talk about it. like the gmail emo, im not sure if Gemini in chrome can do it.
and im not sure if computer use comes with it. maybe it does.
Argon just doesn't matter when even after shipping something they haven't promoted it well.
Next up: using ChatGPT right within a Chrome tab got a major upgrade – the full power of ChatGPT is now in your browser.
Your entire ChatGPT history is right there, and you can ask question in Chat or get things done in Work – side by side while you’re on a site.
0
0
3
194
ShashanK🤺 retweeted
ultimately the "problem" as i see it is that math academia credit assignment is built on the proof-standard economy assuming proofs were scarce, now proofs are not scarce and that currency is undergoing hyperinflation
the question is to what extent the AI labs ought to stabilize the theorem economy during this difficult transition period to an economy built on something else (exposition? understanding in some more concrete form?)
is this an obligation they have to math academia? or something it would be nice of them to do but is supererogatory
the question is to what extent the AI labs ought to stabilize the theorem economy during this difficult transition period to an economy built on something else (exposition? understanding in some more concrete form?)
is this an obligation they have to math academia? or something it would be nice of them to do but is supererogatory
1
1
5
82
ShashanK🤺 retweeted
@teortaxesTexsproportionate focus on financial profitability actually holds back the development of public goods like infrastructure where vast majority of economic value is not captured in the market transaction but positive societal externalities (or mitigating negative ones) that feed into the broad economy. In such an environment, it (somewhat counter-intuitively) makes sense to reduce financial profitability (lower fares) if it drives more network usage. This is why pure market-based approaches have largely failed for public infrastructure.
2) You can tie in the general concept of reducing physical distance with Geoffrey West's work 👇 on the superlinear scaling laws driven by social interaction network density (specifically as a city's population increases, economic outputs like GDP/wages/innovation rise at a power-law exponent of ~1.15.
I think that is the real long-term experiment here: What happens to productivity when 50-60M people in the Pearl or Yangtze deltas live within a "1 hour commuter zone" - compare to equivalent Tokyo at ~30-33M. The experiment is a new form/level of urbanization intensity.
x.com/GlennLuk/statu…
2) You can tie in the general concept of reducing physical distance with Geoffrey West's work 👇 on the superlinear scaling laws driven by social interaction network density (specifically as a city's population increases, economic outputs like GDP/wages/innovation rise at a power-law exponent of ~1.15.
I think that is the real long-term experiment here: What happens to productivity when 50-60M people in the Pearl or Yangtze deltas live within a "1 hour commuter zone" - compare to equivalent Tokyo at ~30-33M. The experiment is a new form/level of urbanization intensity.
x.com/GlennLuk/statu…
9/ The ideas in Scale revolve around the idea that just like you can predict certain characteristics of biological organizations based on their size, you can also predict certain characteristics about cities based on their size, signalling some natural realtionship between the 2.
3
12
70
4,604
same thing with "personalization" btw.
easy way to spot midwits is those who keep blabbering about personalization and democratization. VCs and mediocre thinkers do this a lot.
personalization of movies, games, music blah blah blah blah.
need to break personalization down into more specific functions/vocabulary for each category and determine when they're useful.
easy way to spot midwits is those who keep blabbering about personalization and democratization. VCs and mediocre thinkers do this a lot.
personalization of movies, games, music blah blah blah blah.
need to break personalization down into more specific functions/vocabulary for each category and determine when they're useful.
mainly talking about AI.
applies to internet, mediums etc.
im guilty of it as well when thinking about the internet.
but democratization + quality controls can co-exist.
even democracies have that.
same with "free speech". whole tech twitter used that term like idiots to justify what Elon did to twitter.
opening the floodgates is not democratization.
2
1
1
75
ShashanK🤺 retweeted
10 remarks on AI and math that can be true all at once:
1. The recent progress is stunning and exciting.
2. The Quasi-Riemann Hypothesis is arguably Fields-medal level.
3. Mathematics is being transformed and there is no way back.
4. OpenAI doesn't give a fuck about the future of mathematics.
5. Half-baked unreadable preprints claiming major results can effectively stall mathematical progress.
6. Some mathematicians are responding with naive, overly defensive comments.
7. The social-media wankers framing this as "OpenAI tramples gatekeeping" have no experience of mathematical research.
8. The product of mathematics is clarity and understanding. Not theorems, by themselves.
9. Mathematics is a critical infrastructure of human cognition. The human need for understanding isn't obsolete and it can't be offloaded to machines.
10. We haven't "solved" mathematics and we won't get there before the Sun burns out.
1. The recent progress is stunning and exciting.
2. The Quasi-Riemann Hypothesis is arguably Fields-medal level.
3. Mathematics is being transformed and there is no way back.
4. OpenAI doesn't give a fuck about the future of mathematics.
5. Half-baked unreadable preprints claiming major results can effectively stall mathematical progress.
6. Some mathematicians are responding with naive, overly defensive comments.
7. The social-media wankers framing this as "OpenAI tramples gatekeeping" have no experience of mathematical research.
8. The product of mathematics is clarity and understanding. Not theorems, by themselves.
9. Mathematics is a critical infrastructure of human cognition. The human need for understanding isn't obsolete and it can't be offloaded to machines.
10. We haven't "solved" mathematics and we won't get there before the Sun burns out.
159
465
2,979
158,507
mainly talking about AI.
applies to internet, mediums etc.
im guilty of it as well when thinking about the internet.
but democratization + quality controls can co-exist.
even democracies have that.
same with "free speech". whole tech twitter used that term like idiots to justify what Elon did to twitter.
opening the floodgates is not democratization.
applies to internet, mediums etc.
im guilty of it as well when thinking about the internet.
but democratization + quality controls can co-exist.
even democracies have that.
same with "free speech". whole tech twitter used that term like idiots to justify what Elon did to twitter.
opening the floodgates is not democratization.
democratization seems like an irreproachable catch-all term for dismantling of standards (good and bad)
lot of folks defend the dismantling of good standards using democratization as a shield
0
0
1
134
ShashanK🤺 retweeted
Taking a short break from LLM research to go back to my roots.
Introducing... Concept Lenses! A training-free way to do similarity search along whatever concept you care about.
Typically, embedding-based similarity search is constrained by whatever notion of similarity the ng. This means that if the model is primed for semantic similarity, but you actually care more about visual appearance, you'd need to switch models.
Under a Concept Lens projection, though, distance between embeddings means distance only along the concept you care about.
Blog post: jasperlu.com/blog/prompted-…
Demo: jasperlu.com/projects/conce…
Introducing... Concept Lenses! A training-free way to do similarity search along whatever concept you care about.
Typically, embedding-based similarity search is constrained by whatever notion of similarity the ng. This means that if the model is primed for semantic similarity, but you actually care more about visual appearance, you'd need to switch models.
Under a Concept Lens projection, though, distance between embeddings means distance only along the concept you care about.
Blog post: jasperlu.com/blog/prompted-…
Demo: jasperlu.com/projects/conce…
32
42
548
38,324
ShashanK🤺 retweeted
Yes, math is an aesthetic pursuit! One might say artistic/literary, something like poetry. You need a rich and deep enough ‘culture’ of ‘poets’ to do the ‘poetry’, to carry it forward. If they’re not there, the ‘poetry’ dies. Yes!
The idea that math is just ‘a set of problems to be solved’ is of course false and silly if you think about it. Which ‘problems’? Is it a list? Is the list finite or something? Once the mathematicians have finished ‘the problems’ they’re supposed to stop?
No, of course not. They made the ‘problems’! They are also continually adding to them. Every ‘problem’ on this hypothetical list of ‘problems’ was made-up by the ‘poets’ themselves, usually in abstract language and framing they created. If they aren’t there doing that, I guess there’s no more ‘problems’ either. Congratulations?
The idea that math is just ‘a set of problems to be solved’ is of course false and silly if you think about it. Which ‘problems’? Is it a list? Is the list finite or something? Once the mathematicians have finished ‘the problems’ they’re supposed to stop?
No, of course not. They made the ‘problems’! They are also continually adding to them. Every ‘problem’ on this hypothetical list of ‘problems’ was made-up by the ‘poets’ themselves, usually in abstract language and framing they created. If they aren’t there doing that, I guess there’s no more ‘problems’ either. Congratulations?
One of the funniest consequences of AI is exposing the fact that mathematicians consider math to be an aesthetic pursuit and a source of permanent unvetted sinecures rather than a set of problems to be solved
44
78
586
49,373
very nice
p.migdal.pl/equations-expl…
p.migdal.pl/equations-expl…
Navier-Stokes Equation, explained with colors
p.migdal.pl/equations-expl…
0
1
1
44
yes. please tell me how many hours have all the people spent playing those vibe-coded games set in nyc/sf etc. how many playing hours have the creators of those games logged???
literally all of them are not even optimized. they always hang. heavy threejs sites.
they get a s it. maybe a few clicks.
most of them are not art or a project (forget product). they're just a piece of content made using millions of tokens.
what @vanschneider said:
x.com/vanschneider/s…
literally all of them are not even optimized. they always hang. heavy threejs sites.
they get a s it. maybe a few clicks.
most of them are not art or a project (forget product). they're just a piece of content made using millions of tokens.
what @vanschneider said:
x.com/vanschneider/s…
Every app, game, or open-source project is becoming the equivalent of a post, reel or piece of content online.
You scroll through GitHub repos the way you scroll through a feed. You try a game for five minutes. Open an app once or twice. Glance at a new tool, understand the idea and then move on.
Software itself has entered the content economy. Everything is now a piece of content.
And most of it will get exactly what most content gets:
A few seconds of attention before the next thing arrives.
1
0
3
155