Comments

lilly • November 21, 2026 12:15 AM

This is an interesting perspective on how security practices can sometimes become less effective when they are too complicated for users to follow. Organizations should focus on making security awareness practical and easy to understand. Simple guidelines, regular training, and realistic examples can help people recognize potential threats and make better security decisions.

ResearcherZero • November 21, 2026 12:30 AM

Problems that AI cannot solve, are problems of our own making. These problems will not be solved by wishful thinking. It will take time. More importantly, it will take education, training and funding. This is not possible, without educated and wise leadership. That does not exist. Patience, understanding and humility are skills sorely lacking in politics.

Not only are there direct harms arising from cuts to departments and agencies, indirect impacts have proven detrimental to science and research in the United States, as flow-over effects take their toll. Terminations of grants for scientific research, that were originally mandated by Congress, have knock-on effects that hamper innovation. This creates skills shortages in areas that are fundamental to maintaining key sectors of the economy.

Staff in U.S. government agencies and departments have already experienced the effects. Burnout, institutional memory and experience loss, significant capability depreciation, equipment failure, lack of maintenance and resources needed to meet the requirements for properly functioning institutions. Similar issues will impact all sectors of the American economy, including agriculture, manufacturing and the tech industry. The current policies set by the White House will make the problem worse, reducing competitiveness of U.S. firms.

‘https://www.nytimes.com/2026/10/31/science/how-trump-is-dismantling-americas-scientific-research-machine.html

The level of damage done to U.S. scientific research will have a very detrimental effect.
https://fedscoop.com/records-show-white-house-doge-involvement-in-national-science-foundation-grant-terminations/

The review panels of the Australia’s National Health and Medical Research Council were dismantled in the name of efficiency. This has left Australia without the qualified peer-review and funding panels that would normally discern the credibility of research, and offer feedback and guidance where needed. Without rigorous quality control, publications are flooded with submissions of a low caliber. Making it hard to distinguish genuine advances in scientific research, from the numerous other papers that offer little.

Australia provides an example of how funding cuts have hobbled science and research. Nearly 40% of early and medium-term researchers are considering leaving academia due to the difficulty of securing funding. In the education sector, a lack of science students is creating large shortages in workforce capability. Current and past government policies have ensured a lack of science facilities and STEM subjects to fill that gap. Over the next decade Australia will struggle to innovate due to a skills shortage crisis.

https://www.smh.com.au/national/australia-s-research-system-has-long-been-broken-but-do-we-have-the-will-to-fix-it-20260318-p5oom2.html

Winter • November 21, 2026 4:32 AM

@Dave

Some things as as relevant today as they were 50 years ago.

Regarding Trusting Trust. David A Wheeler found a way to get around the Compiler poisoning

Fully Countering Trusting Trust through Diverse Double-Compiling (DDC) – Countering Trojan Horse attacks on Compilers
‘https://dwheeler.com/trusting-trust/

I think this approach could be adapted to AI.

Clive Robinson • November 21, 2026 5:27 AM

@ Kiwano,

The YouTube chanel only has one vid and it’s a new one.

Whilst it’s the only vid under that contributor handle, I’ve seen it under others.

Also fun for those who know about LLM function this should raise a smile,

https://m.youtube.com/watch?v=gxYjLaJ1UTk

But as with much that is called “YouTube AI slop” it appears on multiple chanels. Look up @Astrablueai to see more AI snark on the AI industry and society. Or look up @Tetsuo (which somebody elsewhere implied was a play on “Test You Out” as in the Turing test reversed).

I’ve been watching the development of AI Entertainment Vids for a while now looking for “tells” and I’ve mentioned one or two in the past (yeah I know it’s the new “continuity slip” game 😉

Mostly they’ve solved the obvious extra limbs and the like but…

Gratuitous use of the “American Flag and insignia” and “Judge Dread / Captin Crunch” type badges is a red flag. But they still are missing out many things…

In the vid you linked too you will note that the blinking and other eye movement is not natural and the microphone on the left ear does not move correctly with the head and face movements.

Another give away on female figures is the “phantom appearance of under cloths” that shows up with a ghostly outline with certain body movements. For Instance in the real world leather trousers do not show a transparency sufficient to show a thong when bending etc.

The reason apparently is due to Hellon Rusk’s AI that gained notoriety for producing images of people that were initially sans attire. That they thought they sufficiently solved with “beach wear”… but this got abused by near transparent attire… So a rule now appears to be “under cloths now have to be solidly there” and mistakes in the implementation make them visible when they would not normally do so.

As has been mentioned by some AI Guru’s AI needs “neurosymbolic AI based on world experience” so clouds are not labeled as whales.

My view is and has been that AI is not going to actually be universally useful thus the G in AGI untill it has real world agency… And even then LLMs and the ML for them is not going to give the I in AGI as it’s the wrong base technology almost entirely…

But with a look at the stock market I suspect the US are not going to have the opportunity to get the moat and they’ve now effectively run out of runway and are running on debt they can not cover. Effectively that means the AI Hype bubble is now a case of

“We’re going down baby, we are going down…”

As paper billionaires discover they owe more tax than the paper can cover…

The question is of course what percentage of the US and world economy is going down as well and how big and long the potential recession will be.

Anonymous • November 21, 2026 6:11 AM

@Dave @Winter – Lean people tend to lean towards having CS backgrounds and most of them have seen Reflections on Trusting Trust. The way GRH was verified was with multiple independent kernels (compilers), some of which have been themselves formally verified to correctly interpret the semantics of Lean’s type system. So the possibility of a bug or backdoor should be regarded as negligible.
One could still take issue with the statement of GRH in Mathlib itself, or worry that Lean’s type system is unsound (there are interesting questions here in that it is stronger than ZFC in certain precise ways) but these, too, seem vanishingly unlikely to have been exploited for this proof.
It is very reasonable to trust that Levent’s proof is formally valid.

ResearcherZero • November 21, 2026 6:12 AM

@Clive Robinson

They all got paid enough to live off for the rest of their lives, even if their stock options sour. I’m sure they would like more, and have army of lobbyists working on it.

<Shadow Lobbying

The act of political influence by strategic advisors and consultants on behalf of industry, requires a carefully constructed vernacular to avoid blatantly describing its own unethical and corrupt behavior. No anti-corruption and ethics legislation burden exists in the space.

‘https://www.forbes.com/sites/williamhartung/2026/11/10/shadow-lobbying-when-is-a–lobbyist-not-a-lobbyist/

Lubricating the golden revolving doors of the nuclear industrial base.
https://responsiblestatecraft.org/nuclear-weapon-lobbying/

Which Senators and Members of the House recieve the cash, and who is paying them?
https://quincyinst.org/research/inside-the-icbm-lobby-special-interests-or-the-public-interest/#h-tools-of-influence-campaign-contributions-and-revolving-door-lobbyists

ResearcherZero • November 21, 2026 6:25 AM

Silicon Valley is chasing its fair share of the US$2 trillion that is currently on offer.
And they sleep with anyone to get it.

‘https://thebulletin.org/2026/08/the-defense-revolving-door-silicon-valley-edition/

The security risks posed by powerful companies that flout good governance are many. Rules and procedures are designed to mitigate those risks. In any organization, once basic protocols are ignored, the integrity of the institution is compromised. To restore such principles requires ethical leadership, and the courage to not bow to the pressure of the mob. Leaders set the stage for the behaviour of their subordinates.

Undue influence and selfishness of those who’s motives are not aligned with good governance, ethics and adherence to internal procedures, are a constant source of risk to a firm’s security. Those risks extend beyond the bounds of companies that build dual-use technologies and handle sensitive data and secrets.

Human weaknesses are the easiest and most valuable route into any secure space. Functions of an organization no longer operate properly when internal procedures and protocols are ignored. When it occurs inside organizations that perform critical functions, staff, public safety, and national security are jeopardized.

When law enforcement breaks the law, the harm is severe. When industry does it, the results can be catastrophic. When industry makes the law, the Gates of Hell await the entry of all living things. The Pitt could do with an overhaul of its catalogue system.

https://www.lowyinstitute.org/the-interpreter/jensen-huang-s-affair-exposes-silicon-valley-s-national-security-power

Winter • November 21, 2026 7:53 AM

@Clive

Also fun for those who know about LLM function this should raise a smile,

That clip is a full course on LLMs in 5 minutes. You need some more course time to fully appreciate it.

Winter • November 21, 2026 7:59 AM

@anonymous

It is very reasonable to trust that Levent’s proof is formally valid.

I don’t know Lean, but I understand the criticism is that the proof might very well be completely valid, but the mathematicians are not sure they understand what exactly was proven.

Winter • November 21, 2026 9:07 AM

@anonymous

> It is very reasonable to trust that > Levent’s proof is formally valid.

I don’t know Lean, but I understand the criticism is that the proof might very well be completely valid, but the mathematicians are not sure they understand what exactly was proven.

PS: The mathematicians feel they ended up in a “The answer is 42” situation.

lurker • November 21, 2026 12:26 PM

@ResearcherZero
re securing research funding

The problem arises because society has shifted the relationship between available funds and curious intellect, such that the money is now filtered through a Government-Industrial-Comples. But few would choose a Great Leap Backwards to the times of Galileo and Newton.

dave • November 21, 2026 12:47 PM

@winter and others

Yes, it’s that exact same criticism raised about the supposed solution to Fermat’s Last Therom. “Nobody understands it except five mathematicians and they are lying”.

That’s reason why domain expertise is not the real moat; the real moat is the social trust put in domain expertise. That’s the problem with Lean. It’s not a question of validity; it’s the problem that a trillion dollars has a lot of power to corrupt and why should someone like myself whose math ability doesn’t surpass basic calculus trust those guys?

One of the oldest pipe dreams in computer science goes all the way back to Babbage.

On two occasions I have been asked, ‘Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?’ I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.

— Charles Babbage

And I cannot comprehend the confusion of human understanding that makes people demand I trust a computer program when a trillion dollars may be on the line. What Lean is doing is not making Anthropic, AI, or LLM’s look great; it’s making people think that mathematicians are selling out to the highest bidder.

Winter • November 21, 2026 2:36 PM

@dave

“Nobody understands it except five mathematicians and they are lying”.

Accusations of malfeasance require evidence.

Leave a comment

Blog moderation policy

Login

Allowed HTML <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre> Markdown Extra syntax via https://michelf.ca/projects/php-markdown/extra/

Sidebar photo of Bruce Schneier by Joe MacInnis.