Unbound - Deep Tech & Space Insights | No 275

The future rarely arrives without complications.

Artificial intelligence is challenging the rules of creative ownership. Ethereum is discovering that upgrading a decentralized network means protecting billions of dollars in existing infrastructure. An AI-powered research system is claiming a breakthrough in quantum computing mathematics. And the Hubble Space Telescope is revealing that some of the stars surrounding our galaxy may not have originated here at all.

Four stories. Four different frontiers. One underlying question:

How do we move forward without losing what matters along the way?

Welcome to Issue 275 of Unbound.

Australia is becoming a battleground over one of artificial intelligence's most consequential questions: Should AI companies be allowed to train on copyrighted creative work without first obtaining permission?

The International Federation of the Phonographic Industry (IFPI), which represents recording industry organizations worldwide, has warned Australia's government against weakening copyright protections to accommodate AI development.

At the center of the dispute is whether AI companies should be allowed to use copyrighted material for training without negotiating individual licenses.

Some technology companies favor systems that permit training unless creators explicitly opt out. Anthropic, for example, has proposed an opt-out approach.

The music industry sees a dangerous precedent: If AI developers can obtain permissive training rights in one jurisdiction, creative work worldwide could be affected.

Australia's government has not adopted the changes under discussion. Prime Minister Anthony Albanese has emphasized protecting creators, while officials have spoken of meaningful control and fair compensation.

The conflict extends far beyond music.

Authors, artists, photographers, filmmakers, and publishers all face the same fundamental question: What happens to the economic value of original work when generative AI can learn from enormous collections of it?

For developers, access to diverse training data can improve model capabilities and reduce licensing friction.

For creators, uncompensated training raises concerns about ownership, competition, and whether the people producing valuable work will share in AI's benefits.

The outcome could help shape international expectations around licensing, consent, and AI training data.

Australia's policymakers must decide how to balance investment in AI with protections for creative industries.

Watch for proposed legislative language, licensing arrangements, and whether the government favors permission-based training, opt-out mechanisms, or another approach.

Any resulting policy could influence debates beyond Australia, although its international legal effects would depend on the details.

The future of AI cannot be built on an assumption that creativity has no cost.

This debate is not simply about protecting yesterday's business models. It is about creating sustainable incentives for tomorrow's original work.

A future in which AI becomes more capable while human creators lose economic leverage would be an extraordinary technological achievement built on a deeply flawed foundation.

Innovation and compensation do not have to be enemies. But building systems that respect both will require more than promises.

Original reporting:
The Guardian — Weakening AI copyright law will spell disaster for creativity

Two competing proposals are attempting to solve an uncomfortable problem: What happens when making Ethereum more efficient threatens applications that cannot be rewritten?

Ethereum developers are debating how to protect existing smart contracts from changes associated with the planned Glamsterdam upgrade.

Part of the upgrade's broader objective is to improve scaling and make the network's pricing better reflect computing and storage costs.

But proposed changes under EIP-8037 and EIP-8038 could increase the gas required for certain operations.

For example, the base gas charge for creating a contract would rise from 32,000 to 183,600 under EIP-8037, excluding additional costs.

That matters because some existing contracts rely on fixed gas assumptions. If those assumptions become invalid, previously functional transactions could fail.

Two competing approaches have emerged:

EIP-8488, proposed by Leo Glisic, would preserve the legacy gas schedule for existing contracts while applying newer pricing to contracts using updated instructions.

Giulio Rebuffo's proposal would introduce separate gas accounting, allowing legacy execution behavior while charging transactions according to updated prices.

Both approaches remain drafts. Neither has been selected as the final solution.

Smart contracts underpin decentralized exchanges, lending markets, custody arrangements, and other blockchain applications.

Many are intentionally immutable.

Immutability builds trust by limiting arbitrary changes, but it also makes compatibility problems hard to fix.

An upgrade that inadvertently breaks established contracts could disrupt applications and potentially put user funds at risk.

This is a central challenge for decentralized infrastructure: How do you improve a system without invalidating the assumptions its users depend on?

Developers must evaluate the proposals' security, implementation complexity, and impact on Ethereum's long-term economics.

Additional testing will be essential before any changes reach the live network.

A previously planned test was postponed amid security concerns, demonstrating why compatibility analysis matters as much as performance improvements.

Neither draft should be treated as approved or deployed.

The hardest part of scaling Ethereum may not be making it faster. It may be preserving trust while changing the rules.

Ethereum's strength comes partly from the expectation that deployed contracts will behave predictably.

That creates a constraint traditional software companies rarely face to the same degree: They can often force an upgrade. Ethereum has to negotiate one across an interconnected, decentralized ecosystem.

The winning solution will need to do more than protect backward compatibility. It must preserve confidence that the network can evolve responsibly.

Original reporting:
CoinDesk — Two rival Ethereum proposals aim to stop Glamsterdam breaking old contracts

An AI-assisted research system has produced a formally checked proof of a longstanding conjecture. If the result holds up under independent scrutiny, it could mark an important milestone for both mathematics and machine-assisted discovery.

Physical Superintelligence (PSI) announced that an internally developed AI research system had produced a proof of the Aaronson–Ambainis conjecture, a problem dating to 2008.

The conjecture addresses a foundational question in quantum computing:

When can quantum computers achieve dramatic speed advantages over classical computers?

The problem concerns quantum algorithms operating on random inputs without useful special structure.

The conjecture implies that, in a particular mathematical model, quantum computers cannot obtain exponential query-complexity advantages on typical random inputs; the corresponding classical overhead can be bounded polynomially.

PSI says its system used multiple instances of an undisclosed Anthropic model and reached a complete proof during its third day of operation.

The proof was formalized in Lean, a mathematical proof assistant that checks logical arguments step by step.

The paper and code have been published, and PSI reports successful machine-checking, including verification with independent checker implementations.

However, the work has not yet undergone journal peer review, and independent expert scrutiny remains important.

Two potentially significant breakthroughs are at stake.

First, quantum computing.

The result would help clarify the importance of exploitable mathematical structure in producing exceptional quantum speedups.

It does not mean quantum computers are ineffective, nor does it undermine established algorithms such as Shor's algorithm, which exploit highly specific mathematical properties.

Second, AI-driven research.

Producing a sophisticated mathematical proof is different from answering questions or summarizing existing knowledge.

A verified result of this kind could strengthen the case for AI systems as genuine research collaborators, particularly when paired with tools that rigorously check their work.

Researchers will need to examine the mathematical arguments, independently reproduce the formal verification, and assess whether the formal statement accurately captures the original conjecture.

Broader scientific acceptance depends on that scrutiny.

If confirmed, the achievement could accelerate interest in AI systems that generate mathematical ideas, construct proofs, and work alongside researchers on difficult open problems.

The most exciting implication is not that AI can answer harder questions. It is that AI may help discover answers humanity has never had.

That represents a meaningful shift in the relationship between humans and intelligent systems.

But we should distinguish a machine-checked formal argument from complete independent scientific validation.

The strongest future for AI-powered discovery will combine machine-generated insight, rigorous formal methods, and human scientific judgment.

If this work survives scrutiny, it could become an early landmark in that emerging model of research.

Original reporting:
TokenPost — Physical Superintelligence Says AI System Proved Quantum Computing Conjecture

Additional research: PSI's published proof and technical explanation

An extraordinary image of an ancient star cluster offers something more valuable than a beautiful view of space: a glimpse into the Milky Way's complicated history.

NASA and the European Space Agency have released a Hubble Space Telescope image of Messier 30, also known as M30, a globular star cluster approximately 27,500 light-years from Earth in the constellation Capricornus.

Globular clusters are dense, roughly spherical collections containing tens of thousands to millions of stars.

Many are extraordinarily old, with estimated ages of about 11.5 to 13.5 billion years.

That makes them valuable records of early-universe conditions.

M30 is particularly intriguing because it belongs to a population of globular clusters whose orbital direction runs opposite to the Milky Way's rotation.

One possible explanation is that some of these clusters formed outside our galaxy and were subsequently captured as the Milky Way grew.

The image combines Hubble observations across ultraviolet, visible, and near-infrared wavelengths.

Our galaxy did not necessarily form as a single, isolated structure.

Astronomers believe that interactions, mergers, and the accumulation of smaller systems helped shape galaxies over billions of years.

Ancient star clusters preserve chemical and dynamical evidence of that history.

By studying their stellar populations, orbital motions, and composition, scientists can investigate how the Milky Way assembled itself.

Hubble's multi-year observations also let researchers measure extremely small changes in stellar positions and movements.

Astronomers will continue comparing cluster motions, stellar ages, and chemical properties to reconstruct the Milky Way's formation history.

Long-baseline Hubble observations and complementary measurements from other observatories can help distinguish stars that formed within our galaxy from populations that may have arrived through ancient mergers.

M30's orbit is a clue, not definitive proof of an extragalactic origin.

Sometimes the most important discoveries are not about finding something new. They're about realizing that something familiar has a completely different origin story.

M30 reminds us that galaxies are not static collections of stars.

They are evolving systems shaped by interactions stretching across billions of years.

The night sky may look permanent, but the Milky Way's history is written in motion.

And some of the stars we call our own may once have belonged to another galaxy.

Original source: NASA — M30's Sparkling Stars

Further reading: NASA — Hubble Learns Galactic History From Stars

Progress does not arrive in a vacuum.

AI needs access to knowledge, but creators need sustainable incentives. Ethereum needs to scale, but existing contracts need to remain reliable. Machine-assisted mathematical discovery needs room to advance, but extraordinary claims still demand rigorous verification.

And above it all, ancient star clusters remind us that even the structures we think we understand have histories far more complicated than they appear.

This week's stories point toward a shared lesson:

The most important breakthroughs aren't simply about making systems more powerful. They're about understanding the foundations those systems depend on.

The future is being written across code, creativity, mathematics, and the cosmos.

Stay curious. Stay ahead. Stay Unbound.

Unbound — Deep Tech & Space Insights

Exploring the breakthroughs reshaping technology, science, and our understanding of the universe.