Connect with us

AI Research

Why the Gap Always Wins

Published

on


View a PDF of the paper titled Murphys Laws of AI Alignment: Why the Gap Always Wins, by Madhava Gaikwad

View PDF
HTML (experimental)

Abstract:We prove a formal impossibility result for reinforcement learning from human feedback (RLHF). In misspecified environments with bounded query budgets, any RLHF-style learner suffers an irreducible performance gap Omega(gamma) unless it has access to a calibration oracle. We give tight lower bounds via an information-theoretic proof and show that a minimal calibration oracle suffices to eliminate the gap. Small-scale empirical illustrations and a catalogue of alignment regularities (Murphy’s Laws) indicate that many observed alignment failures are consistent with this structural mechanism. Our results position Murphys Gap as both a diagnostic limit of RLHF and a guide for future work on calibration and causal preference checks.

Submission history

From: Madhava Gaikwad [view email]
[v1]
Thu, 4 Sep 2025 23:03:25 UTC (375 KB)
[v2]
Wed, 10 Sep 2025 14:35:38 UTC (370 KB)



Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

AI Research

Oracle’s AI-Powered Sales Growth Threatens Elon Musk’s Reign as World’s Richest Man

Published

on


Oracle is on an all-time roll — and the prophecy written in its stock price is that founder Larry Ellison may become the wealthiest man on the face of the planet.

The company reported its latest earnings results after the bell on Tuesday, including an astounding projection that revenue from its cloud computing will increase 77% to $18 billion this year and could rise as high as $144 billion annually by 2030, thanks in large part to four multibillion-dollar deals with three different customers. By Wednesday afternoon, The Wall Street Journal reported on one of those marquee deals: a five-year, $300 billion agreement to supply cloud computing power to OpenAI.

Picks and Shovels

Oracle said its remaining performance obligations (RPOs), the contracted revenue that has not yet been realized, such as the OpenAI deal that the WSJ reported will begin in 2027, are now sitting at $455 billion. That’s up 359% from just a year ago, inspiring quite literal words of awe from Wall Street analysts during the company’s earnings call: Guggenheim’s John DiFucci said he was “blown away,” while Brad Zelnick of Deutsche Bank said, “We’re all kind of in shock, in a very good way.”

In other words, Oracle has become a favorite supplier of picks and shovels amid the artificial intelligence gold rush. And CEO Safra Catz says that’s because it’s different from cloud computing rivals such as Microsoft, Google and Amazon: “Some of our competitors, they like to own buildings … That’s not really our specialty. Our specialty is the unique technology, the unique networking, the storage — just the whole way we put these systems together.”

More importantly, Oracle’s RPO figure is an indication that the AI industry is preparing to spend big to keep the gold rush from going bust anytime soon:

  • In a June filing, OpenAI revealed it was on track to generate just $10 billion in revenue this year, leagues away from the roughly $60 billion it will be paying Oracle under the deal. Meanwhile, sources told the WSJ earlier this year that OpenAI’s Sam Altman recently informed investors the company won’t be profitable until 2029 (and it’s still sorting out its for-profit structure).
  • Oracle is similarly putting growth over profit; the company’s debt-to-equity ratio has soared to 427%, with S&P data showing its $21 billion in operating cash flow in the 12 months through August was eclipsed by its $27 billion in capital expenditures. 

Margin Call: The AI moment has not only been very good to picks-and-shovels companies like Oracle but also to financiers on Wall Street funding the massive debt splurge. Not everyone is drooling, though. In a note published Wednesday seen by MarketWatch, D.A. Davidson analyst Gil Luria flagged that “at best,” Oracle is running its AI cloud compute services “at single-digit operating margins, if not serving compute at a loss in some instances,” a far cry from the 50% margins of its legacy business. All the same, Oracle’s share price leaped 36% by the close of trading Wednesday, and was sufficiently high during much of the trading day that Ellison’s fortune briefly eclipsed that of Elon Musk, according to Bloomberg. Oracle’s was the only single-day gain greater than 25% for a company worth at least half a trillion dollars, according to Dow Jones data. ​​It’s also good for the company’s best single-day stock performance since 1999 … or right before a certain early digital age bubble started to burst.



Source link

Continue Reading

AI Research

Aston University and University of Leeds win £3.4m for AI research network

Published

on


A joint project from Aston University and the University of Leeds to set up an artificial intelligence (AI) tools research network has won £3.4m from the Research England Development Fund.

The four-year funding will go towards developing a network to assess AI tools available for PhD research and promote responsible AI uptake and innovation by researchers.

The Artificial Intelligence Researcher Development Network Plus (AI.RDN+) is being led by Aston University’s Professor Phil Mizen, professor of sociology and policy at Aston University, and University of Leeds’ Dr Hosam Al-Samarraie, associate professor in digital innovation design, and Professor Arunangsu Chatterjee, dean of digital transformation. The network will also include the eight Midlands Innovation research universities and the 12 members of the Yorkshire Universities consortium.

AI.RDN+ will carry out an extensive consultation with PhD researchers, supervisors, examiners and research-enabling staff, such as professional services staff and technicians, to understand how publicly available AI tools are used and how its challenges are identified and negotiated. Using this information, AI.RDN+ will create a resource base with guidance on what AI tools are available, how they can be used, and identify case studies of best practice.

The network will also create training resources for all stakeholders, co-creating and testing guidance, training, and professional and career development resources with the 20 universities in the Midlands Innovation and Yorkshire Universities consortia.

Professor Mizen said: “The Artificial Intelligence Researcher Development Network Plus will provide detailed knowledge of the uptake and impact of publicly available AI tools across the doctoral ecosystem and use this to co-create much-needed information, resources and professional and skills training opportunities. Our project is a unique opportunity to build knowledge and capture innovation, and to use this to build the resources needed for the ethical and responsible use of AI in doctoral research.”

Professor Chatterjee added: “This award reflects the importance of collaboration across universities to understand and shape the role of AI in research. By working alongside Aston and partners across the Midlands and Yorkshire, we can bring together complementary expertise and perspectives.

“I am particularly pleased to see Dr Hosam Al-Samarraie leading this work for Leeds, bringing both expertise and vision to the partnership. Together, through this network, we can build shared resources and approaches that ensure AI adoption in doctoral research is innovative, ethical and delivers real benefit for researchers and society.”

The eight Midlands Innovation universities are: Aston University, Cranfield University, University of Birmingham, University of Keele, Loughborough University, University of Leicester, University of Nottingham, and University of Warwick.

The 12 Yorkshire Universities are: University of Bradford, University of Leeds, Leeds Beckett University, Leeds Trinity University, Leeds Conservatoire, Leeds Arts University, University of Hull, University of Huddersfield, University of Sheffield, University of York.

AI.RDN+ also has the active support of a range of expert organisations, including Jisc, UK Council for Graduate Education, Vitae, and the National Centre for Universities and Business. It is also supported by two other Research England-funded projects – the Next Generation Research SuperVision Project, on the future of doctoral supervision, and Prosper, researching the professional and career development of post-doctoral researchers.

UKRI Research England shapes research and knowledge exchange in English universities and distributes over £2bn to universities in England every year.

Register now to receive our free daily business news emails…



Source link

Continue Reading

AI Research

The Blogs: Artificial Intelligence is getting worse, not better | Moshe-Mordechai van Zuiden

Published

on


The future lies behind us

The hype is still fueled by false prophets and wishful ‘thinkers.’

Many invested trillions, so new investors must be lured all the time.

I said it before, and I’ll say it again: AI is neither; it’s a collection of human stupidity. It will investigate how to colonize the galaxy instead of how to clean up Earth. It will tell us that all can be solved without addressing the real problem: human irrationality. It’s predicting the past almost perfectly.

Its only creativity is with the truth, but it’s still lagging far behind Trump.

It’s ruining grammar checkers with its promotion of floweriness.

Its texts, full of irrelevant details, increasingly steal of our time.

Its music is still boooooring, its visuals increasingly unreal.

Who needs AI when porn and social media already killed off our brains?

After brainstorming, we must carefully check the craziness we came up with. We don’t have the personnel to check all of AI’s craziness.

Its development is promising and will always stay that.

If AI ever were to become serious, it would cost so much energy that we’ll cook the oceans and go extinct in one giant global heatwave.

On AI, cut your losses.


You may find more controversial writings on Amazon or my own blog.

MM is a prolific and creative writer and thinker, previously a daily blog contributor to the TOI. He often makes his readers laugh, mad, or assume he’s nuts—close to perfect blogging. He’s proud that his analytical short comments are removed both from left-wing and right-wing news sites. None of his content is generated by the new bore on the block, AI. *
As a frontier thinker, he sees things many don’t yet. He’s half a prophet. Half. Let’s not exaggerate. Or not at all because he doesn’t claim G^d talks to him. He gives him good ideas—that’s all. MM doesn’t believe that people observe and think in a vacuum. He, therefore, wanted a broad bio that readers interested can track a bit what (lack of) backgrounds, experiences, and educations contribute to his visions. *
This year, he will prioritize getting his unpublished books published rather than just blog posts. Next year, he hopes to focus on activism against human extinction. To find less-recent posts on a subject XXX among his over 2000 archived ones, go to the right-top corner of a Times of Israel page, click on the search icon and search “zuiden, XXX”. One can find a second, wilder blog, to which one may subscribe too, here: https://mmvanzuiden.wordpress.com/ or by clicking on the globe icon next to his picture on top. *
Like most of his readers, he believes in being friendly, respectful, and loyal. However, if you think those are his absolute top priorities, you might end up disappointed. His first loyalty is to the truth. He will try to stay within the limits of democratic and Jewish law, but he won’t lie to support opinions or people when don’t deserve that. (Yet, we all make honest mistakes, which is just fine and does not justify losing support.) He admits that he sometimes exaggerates to make a point, which could have him come across as nasty, while in actuality, he’s quite a lovely person to interact with. He holds – how Dutch – that a strong opinion doesn’t imply intolerance of other views. *
Sometimes he’s misunderstood because his wide and diverse field of vision seldomly fits any specialist’s box. But that’s exactly what some love about him. He has written a lot about Psychology (including Sexuality and Abuse), Medicine (including physical immortality), Science (including basic statistics), Politics (Israel, the US, and the Netherlands, Activism – more than leftwing or rightwing, he hopes to highlight reality), Oppression and Liberation (intersectionally, for young people, the elderly, non-Whites, women, workers, Jews, LGBTQIA+, foreigners and anyone else who’s dehumanized or exploited), Integrity, Philosophy, Jews (Judaism, Zionism, Holocaust and Jewish Liberation), the Climate Crisis, Ecology and Veganism, Affairs from the news, or the Torah Portion of the Week, or new insights that suddenly befell him. *
Chronologically, his most influential teachers are his parents, Nico (natan) van Zuiden and Betty (beisye) Nieweg, Wim Kan, Mozart, Harvey Jackins, Marshal Rosenberg, Reb Shlomo Carlebach, and, lehavdil bein chayim lechayim, Rabbi Dr. Natan Lopes Cardozo, Rav Zev Leff, and Rav Meir Lubin. This short list doesn’t mean to disrespect others who taught him a lot or a little. One of his rabbis calls him Mr. Innovation [Ish haChidushim]. Yet, his originalities seem to root deeply in traditional Judaism, though they may grow in unexpected directions. In fact, he claims he’s modernizing nothing. Rather, mainly basing himself on the basic Hebrew Torah text, he tries to rediscover classical Jewish thought almost lost in thousands of years of stifling Gentile domination and Jewish assimilation. (He pleads for a close reading of the Torah instead of going by rough assumptions of what it would probably mean and before fleeing to Commentaries.) This, in all aspects of life, but prominently in the areas of Free Will, Activism, Homosexuality for men, and Redemption. *
He hopes that his words will inspire and inform, and disturb the comfortable and comfort the disturbed. He aims to bring a fresh perspective rather than harp on the obvious and familiar. When he can, he loves to write encyclopedic overviews. He doesn’t expect his readers to agree. Rather, original minds should be disputed. In short, his main political positions are among others: anti-Trumpism, for Zionism, Intersectionality, non-violence, anti those who abuse democratic liberties, anti the fake ME peace process, for original-Orthodoxy, pro-Science, pro-Free Will, anti-blaming-the-victim, and for down-to-earth, classical optimism, and happiness. Read his blog on how he attempts to bridge any tensions between those ideas or fields. *
He is a fetal survivor of the pharmaceutical industry (https://diethylstilbestrol.co.uk/studies/des-and-psychological-health/), born in 1953 to his parents who were Dutch-Jewish Holocaust survivors who met in the largest concentration camp in the Netherlands, Westerbork. He grew up a humble listener. It took him decades to become a speaker too, and decades more to admit to being a genius. But his humility was his to keep. And so was his honesty. Bullies and con artists almost instantaneously envy and hate him. He hopes to bring new things and not just preach to the choir. *
He holds a BA in medicine (University of Amsterdam) – is half a doctor. He practices Re-evaluation Co-counseling since 1977, is not an official teacher anymore, and became a friendly, powerful therapist. He became a social activist, became religious, made Aliyah, and raised three wonderful kids. Previously, for decades, he was known to the Jerusalem Post readers as a frequent letter writer. For a couple of years, he was active in hasbara to the Dutch-speaking public. He wrote an unpublished tome about Jewish Free Will. He’s a strict vegan since 2008. He’s an Orthodox Jew but not a rabbi. *
His writing has been made possible by an allowance for second-generation Holocaust survivors from the Netherlands. It has been his dream since he was 38 to try to make a difference by teaching through writing. He had three times 9-out-of-10 for Dutch at his high school finals but is spending his days communicating in English and Hebrew – how ironic. G-d must have a fine sense of humor. In case you wonder – yes, he is a bit dyslectic. If you’re a native English speaker and wonder why you should read from people whose English is only their second language, consider the advantage of having an original peek outside of your cultural bubble. *
To send any personal reaction to him, scroll to the top of the blog post and click Contact Me. *
His newest books you may find here: https://www.amazon.com/s?i=stripbooks&rh=p_27%3AMoshe-Mordechai%2FMaurits+van+Zuiden&s=relevancerank&text=Moshe-Mordechai%2FMaurits+van+Zuiden&ref=dp_byline_sr_book_1





Source link

Continue Reading

Trending