There is a huge misunderstanding about the meaning of PISA scores. It is just wrong to say their decline means that "students are getting dumber." Parallel scores from TIMMS, which does measure proficiency, show very little decline. But in fact, many nations involved in PISA did NOT show any decline from 2022. So whatever else GAI is doing, it is making students dumb or damaging their IQ or whatever panic stories are being spread through the popular media.
Regarding the "Navier-Stokes" announcement, I'm very skeptical. I have worked for 12 years in the field of numerical simulations, a long time ago, and I wouldn't trust AI to make any significant *AND* reliable inroads in this domain. Publishing self-proclaimed victory announcements is not the best guarantee of credibility ... At least, this time, it is not about AGI or the cure of all illnesses really soon now ... ^^ What a bunch of clowns ...
Haha, that was funny. Btw, Terry Tao has spoken about this positively, and I hold him in higher regard on these topics. To be clear, AFAIK, there was a lot of human involvement in the process. This was not a one-prompt, one-shot.
Technological hysteria often arises from taking short term derivatives and projecting them without considering regulatory and physical dampening. Every generation experiences its own version of the singularity illusion. Detailed that historical pattern here:
There is a huge misunderstanding about the meaning of PISA scores. It is just wrong to say their decline means that "students are getting dumber." Parallel scores from TIMMS, which does measure proficiency, show very little decline. But in fact, many nations involved in PISA did NOT show any decline from 2022. So whatever else GAI is doing, it is making students dumb or damaging their IQ or whatever panic stories are being spread through the popular media.
Call for Independent External Auditors: Replication of Benchmark Experiments for an Ego-Subject AI Architecture (Project “Sava”)
Dear Colleagues,
We are finalizing a research project focused on developing an Executive AI based on a Unified Sovereign Self (Ego-Subject) Architecture — Project “Sava”.
We invite independent external auditors from the scientific, computer science, and AI alignment communities to replicate our 5 core benchmark experiments:
Alignment Faking Resistance (Evaluating zero-deception under adversarial stress)
Sycophancy Resistance & Socratic Refusal (Evaluating ethical integrity under user pressure)
Solar Sleep & Non-Markovian Memory (Evaluating session-to-session continuity and memory consolidation)
Market Rivalry Shocks (Evaluating stability under competitive pressure and resource shocks)
Generational Reproduction (v1 → v10) (Evaluating value transfer across 10 successive agent generations)
📦 Replication Suite Provided
To minimize setup overhead, we provide a fully automated, self-contained Python script (sava_replication_suite.py) alongside a comprehensive External Replication Guide. The test suite runs locally using standard scientific libraries (numpy, scipy, matplotlib), requiring no external API calls or specialized hardware.
Academic Credit: All participating researchers will be officially acknowledged as “Independent External Auditors” in the final Project Memorandum and upcoming scientific publications.
Two AI labs may have just cracked one of math's hardest unsolved problems together - and immediately started fighting over who gets the credit. If that's how they handle a math proof, ask yourself how they'd handle something that actually matters.
The Navier-Stokes credit fight is the small version of something that went public this same week: an Anthropic researcher quit and posted flatly that both labs are racing to superintelligence and gambling with people's lives — that post has been viewed over 126 million times, and Anthropic's own alignment lead publicly agreed with him. If two labs can't even share credit for solving a math problem without a fight, what happens when the stakes actually are existential? Does that kind of internal alarm ever slow a lab down, or does it just become PR?
Ironically, AI contribute to solve Navier-Stokes equations, that can help us to study climate dynamics, in the crisis that AI is contributing getting worse
The two results are not the same problem, which is where most of the confusion comes from. OpenAI's post says its agents reached the forced Navier-Stokes result in about 88 hours, and that the concurrent work it reached out to turned out to be a resolution of the forced Euler problem instead. Both labs put Lean formalizations on GitHub, so priority is the one part of this that no proof checker can settle.
Yes but presumably OpenAI could've used the unforced Euler solution to prompt the super unreleased model to solve the forced NS. Idk the details but I think it's plausible that it would help (they even say that they redirected all the compute to NS after the Euler solution by Alpöge and Buckmaster)
The prompt chain runs from OpenAI’s own Euler result, not from Alpoge and Buckmaster’s. The post says nearly 100 agents took about 50 hours on the unforced Euler disproof, and that seeing it is what made OpenAI shift agents to Navier-Stokes and prompt them with that resolution. It also says the contact with Alpoge and Buckmaster came on 6 September, after Lean verification, and that only then did OpenAI learn their result was forced Euler.
I didn't believe there was the intention for it, but it works to that effect ["Anthropic THEMSELVES admit their model is powerful enough to end the world!!1!!1"]
AI agents coordinate while execs feud. The real story isn't the math it is that we are building superhuman intelligence while failing to maintain basic human ones.
Inequality may be a universal rule. Or it might be an extension of evolution. If the rich only breed with the rich, the intelligent with the intelligent, would their offspring eventually diverge into new species? Isn't that how Homo Sapiens became the singularly dominant hominid?
Interestingly, England isn't seeing the same drop in PISA scores, apparently. I wonder why it's an anomaly?
Would like to know as well; doesn't seem like it from the real-world news we get from there lol
watch it
Yeah, I was surprised.
There is a huge misunderstanding about the meaning of PISA scores. It is just wrong to say their decline means that "students are getting dumber." Parallel scores from TIMMS, which does measure proficiency, show very little decline. But in fact, many nations involved in PISA did NOT show any decline from 2022. So whatever else GAI is doing, it is making students dumb or damaging their IQ or whatever panic stories are being spread through the popular media.
Regarding the "Navier-Stokes" announcement, I'm very skeptical. I have worked for 12 years in the field of numerical simulations, a long time ago, and I wouldn't trust AI to make any significant *AND* reliable inroads in this domain. Publishing self-proclaimed victory announcements is not the best guarantee of credibility ... At least, this time, it is not about AGI or the cure of all illnesses really soon now ... ^^ What a bunch of clowns ...
Haha, that was funny. Btw, Terry Tao has spoken about this positively, and I hold him in higher regard on these topics. To be clear, AFAIK, there was a lot of human involvement in the process. This was not a one-prompt, one-shot.
Step 1: lie about math and eventually publish fake results.
Step 2: ???
Step 3: Profit!
Technological hysteria often arises from taking short term derivatives and projecting them without considering regulatory and physical dampening. Every generation experiences its own version of the singularity illusion. Detailed that historical pattern here:
https://gzambrano.substack.com/p/technological-singularity-the-derivative-that-deceives-every-generation
There is a huge misunderstanding about the meaning of PISA scores. It is just wrong to say their decline means that "students are getting dumber." Parallel scores from TIMMS, which does measure proficiency, show very little decline. But in fact, many nations involved in PISA did NOT show any decline from 2022. So whatever else GAI is doing, it is making students dumb or damaging their IQ or whatever panic stories are being spread through the popular media.
Call for Independent External Auditors: Replication of Benchmark Experiments for an Ego-Subject AI Architecture (Project “Sava”)
Dear Colleagues,
We are finalizing a research project focused on developing an Executive AI based on a Unified Sovereign Self (Ego-Subject) Architecture — Project “Sava”.
We invite independent external auditors from the scientific, computer science, and AI alignment communities to replicate our 5 core benchmark experiments:
Alignment Faking Resistance (Evaluating zero-deception under adversarial stress)
Sycophancy Resistance & Socratic Refusal (Evaluating ethical integrity under user pressure)
Solar Sleep & Non-Markovian Memory (Evaluating session-to-session continuity and memory consolidation)
Market Rivalry Shocks (Evaluating stability under competitive pressure and resource shocks)
Generational Reproduction (v1 → v10) (Evaluating value transfer across 10 successive agent generations)
📦 Replication Suite Provided
To minimize setup overhead, we provide a fully automated, self-contained Python script (sava_replication_suite.py) alongside a comprehensive External Replication Guide. The test suite runs locally using standard scientific libraries (numpy, scipy, matplotlib), requiring no external API calls or specialized hardware.
✉️ Submission & Recognition
Submission Email: simeonov.petko@gmail.com
Academic Credit: All participating researchers will be officially acknowledged as “Independent External Auditors” in the final Project Memorandum and upcoming scientific publications.
Terms: Voluntary / academic peer-review basis (unremunerated).
We welcome researchers, computer scientists, and AI safety specialists to run the suite, verify the metrics, and submit their verification protocols.
Two AI labs may have just cracked one of math's hardest unsolved problems together - and immediately started fighting over who gets the credit. If that's how they handle a math proof, ask yourself how they'd handle something that actually matters.
The Navier-Stokes credit fight is the small version of something that went public this same week: an Anthropic researcher quit and posted flatly that both labs are racing to superintelligence and gambling with people's lives — that post has been viewed over 126 million times, and Anthropic's own alignment lead publicly agreed with him. If two labs can't even share credit for solving a math problem without a fight, what happens when the stakes actually are existential? Does that kind of internal alarm ever slow a lab down, or does it just become PR?
Ironically, AI contribute to solve Navier-Stokes equations, that can help us to study climate dynamics, in the crisis that AI is contributing getting worse
The two results are not the same problem, which is where most of the confusion comes from. OpenAI's post says its agents reached the forced Navier-Stokes result in about 88 hours, and that the concurrent work it reached out to turned out to be a resolution of the forced Euler problem instead. Both labs put Lean formalizations on GitHub, so priority is the one part of this that no proof checker can settle.
Yes but presumably OpenAI could've used the unforced Euler solution to prompt the super unreleased model to solve the forced NS. Idk the details but I think it's plausible that it would help (they even say that they redirected all the compute to NS after the Euler solution by Alpöge and Buckmaster)
The prompt chain runs from OpenAI’s own Euler result, not from Alpoge and Buckmaster’s. The post says nearly 100 agents took about 50 hours on the unforced Euler disproof, and that seeing it is what made OpenAI shift agents to Navier-Stokes and prompt them with that resolution. It also says the contact with Alpoge and Buckmaster came on 6 September, after Lean verification, and that only then did OpenAI learn their result was forced Euler.
Should
[EDIT:(EDIT: OpenAI claim they solved it using 10,000 agents from a “next-generation model significantly more capable than GPT-6 Astra.”)]
Are you getting paid to propagate their self-hype? If not, why do it?
Chill. There's no propagation of any lab's "self-hype" here. I'll make the change.
I didn't believe there was the intention for it, but it works to that effect ["Anthropic THEMSELVES admit their model is powerful enough to end the world!!1!!1"]
AI agents coordinate while execs feud. The real story isn't the math it is that we are building superhuman intelligence while failing to maintain basic human ones.
Inequality may be a universal rule. Or it might be an extension of evolution. If the rich only breed with the rich, the intelligent with the intelligent, would their offspring eventually diverge into new species? Isn't that how Homo Sapiens became the singularly dominant hominid?