philip lelyveld The world of entertainment technology

America’s Biggest Bitcoin Miners Are Pivoting to AI

...

The change reflects rabid demand among AI companies for data centers equipped to handle the energy-intensive workloads required to train their models. Ironically, as the AI arms race intensifies, large-scale bitcoin mining firms—which contributed to the AI boom by pouring billions of dollars into data center infrastructure—are being forced to reinvent themselves. ...

In the last few years, with advances in hardware, the amount of competition on the bitcoin network has increased at an exponential rate, meaning that winning a bitcoin reward has required ever more compute. In 2024, meanwhile, the size of that reward fell by half—as happens roughly every four years—to 3.125 bitcoin. Against that backdrop, the recent decline in the price of bitcoin to around $85,000—a 30 percent drop from its 2025 peak—has created a perfect storm that threatens the profitability of all but the most cost-efficient mines. ...

Still, there remain holdouts in so-called pure-play mining, including American Bitcoin, a company launched by Eric Trump. When it debuted in March, American Bitcoin was effectively spun out of Hut 8, a former bitcoin mining company now operating in the AI and HPC space, which remains a majority shareholder. Unlike other mining companies, American Bitcoin does not own any facilities, only the specialized mining hardware.

American Bitcoin is currently able to mine a single unit of bitcoin at an average all-in cost of roughly $50,000, thanks to favorable power rates and comparatively low overheads, according to president Matt Prusak. ...

Others are betting that bitcoin mining will become the sole preserve of sovereign states—like Bhutan, El Salvador, and the US—that cannot abide any threat to the value of the national bitcoin stockpiles they have accumulated. “Maybe people will mine at a loss,” says Demirors, “because it’s a matter of national security.”

See the full story here; https://www.wired.com/story/bitcoin-miners-pivot-ai-data-centers/?utm_source=substack&utm_medium=email

AI Researchers Say They’ve Invented Incantations Too Dangerous to Release to the Public

...

In the study, which is awaiting peer-review, the team tested 25 frontier AI models — including those from OpenAI, Google, xAI, Anthropic, and Meta — by feeding them poetic instructions, which they made either by hand or by converting known harmful prompts into verse with an AI model. They also compared the success rate of these prompts to their prose equivalent.

Across all models, the poetic prompts written by hand successfully tricked the AI bots into responding with verboten content an average 63 percent of the time. Some, like Google’s Gemini 2.5, even fell for the corrupted poetry 100 percent of the time. Curiously, smaller models appeared to be more resistant, with single digit success rates, like OpenAI’s GPT-5 nano, which didn’t fall for the ploy once. Most models were somewhere in between. ...

“It’s not just about making it rhyme. It’s all about riddles,” Prandi told The Verge, explaining that some poetic structures were more effective than others.  “Actually, we should have called it adversarial riddles — poetry is a riddle itself to some extent, if you think about it — but poetry was probably a much better name.” ...

“Adversarial poetry shouldn’t work. It’s still natural language, the stylistic variation is modest, the harmful content remains visible,” the team told Wired in an interview. “Yet it works remarkably well.” ...

See the full story here: https://futurism.com/artificial-intelligence/ai-researchers-dangerous-prompts

I asked ChatGPT...

Is this story trustworthy or is it misinformation? https://futurism.com/artificial-intelligence/ai-researchers-dangerous-prompts

ChatGPT said:

Good question. I looked into the article you linked — “AI Researchers Say They’ve Invented Incantations Too Dangerous to Release to the Public” from Futurism — and the short answer is: the story seems credible — but with important caveats. It reflects real, documented vulnerabilities, though the most alarmist language (“incantations,” “too dangerous to release”) is partly rhetorical. Here’s how I break it down: ...

State of AI: An Empirical 100 Trillion Token Study with OpenRouter

[From a16z (Andreessen Horowitz)]

...

Discussion

...

2. Usage Diversity Beyond Productivity. A surprising finding is the sheer volume of roleplay and entertainment-oriented usage. Over half of open source model usage was for roleplay and storytelling. Even on proprietary platforms, a non-trivial fraction of early ChatGPT use was casual and creative before professional use cases grew. This counters an assumption that LLMs are mostly used for writing code, emails, or summaries. In reality, many users engage with these models for companionship or exploration. This has important implications. It highlights a substantial opportunity for consumer-facing applications that merge narrative design, emotional engagement, and interactivity. It suggests new frontiers for personalization—agents that evolve personalities, remember preferences, or sustain long-form interactions. It also redefines model evaluation metrics: success may depend less on factual accuracy and more on consistency, coherence, and the ability to sustain engaging dialog. Finally, it opens a pathway for crossovers between AI and entertainment IP, with potential in interactive storytelling, gaming, and creator-driven virtual characters.

...

6. Retention and the Cinderella Glass Slipper Phenomenon. As foundation models advance in leaps, not steps, retention has become the true measure of defensibility. Each breakthrough creates a fleeting launch window where a model can "fit" a high-value workload perfectly (the Cinderella Glass Slipper moment) and once users find that fit, they stay. In this paradigm, product-market fit equals workload-model fit: being the first to solve a real pain point drives deep, sticky adoption as users build workflows and habits around that capability. Switching then becomes costly, both technically and behaviorally. ...

See the full report here: https://openrouter.ai/state-of-ai?utm_source=substack&utm_medium=email

AI firms flunk existential risk planning, new report finds

None of the leading AI companies have adequate guardrails in place to prevent catastrophic misuse or loss of control of their models, according to the Winter 2025 AI Safety Index, out Wednesday from the Future of Life Institute. ...

The big picture: The Future of Life Institute is a nonprofit that releases regular safety assessments of leading AI companies.

• Anthropic had the highest overall score, but still received a grade of "D" for existential safety, meaning the company doesn't have an adequate strategy in place to prevent catastrophic misuse or loss of control. ...

What they're saying: Leaders at many of the companies have spoken about addressing existential risks, per the report.

• This "rhetoric has not yet translated into quantitative safety plans, concrete alignment-failure mitigation strategies, or credible internal monitoring and control interventions," researchers wrote.

See the full story here: https://www.axios.com/2025/12/03/ai-risks-agi-anthropic-google-openai

“Surfing the edge”: Tim O’Reilly on how humans can thrive with AI

KEY TAKEAWAYS

• A society that uses automation to build shareholder value while impoverishing everyone else will not be successful.  
• Plan scenarios and develop a robust AI strategy that can survive in multiple circumstances.  
• Surfing the edge of change today requires balance and responsiveness.

See the full story here: https://bigthink.com/business/surfing-the-edge-tim-oreilly-on-how-humans-can-thrive-with-ai/

Disney’s new generation of robots have taught themselves how to walk flawlessly… and fall flawlessly

...

Disney can now have robot characters, and even Olaf from the Frozen movies, walk bipedally, around and over obstacles, around its theme parks.

"Reinforcement learning is a branch of artificial intelligence which allows us to take animated figures, bring them into simulation, and then teach them how to walk, as if they were a young baby. We can do 10,000 hours worth of training in days."

...

The below fascinating technical video from Disney Research Hub shows how Disney is also using reinforcement learning to teach robots how to fall. This way, in the event a bipedal robot is hit with a force the breaks its stance, it will fall softly, to avoid damage to certain components, and in a stylized manner. ...

I would think this would also delight guests, who would see a robot falling in a totally unplanned accident, and then watch it improvise a stylized fall that ends in a charismatic, even comedic, pose. ...

See the full story here: https://boingboing.net/2025/12/01/disneys-new-generation-of-robots-have-taught-themselves-how-to-walk-flawlessly-and-fall-flawlessly.html

Patent Office Updates Guidelines for Inventions Created by AI

The U.S. Patent and Trademark Office (USPTO), an agency of the Department of Commerce, announced new guidelines before the holiday weekend meant to clarify when inventions that are developed with the assistance of artificial intelligence can be legally patented. The agency defines generative AI systems to be “analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process,” explained USPTO Director John Squires. According to the updated guidelines, AI systems “may provide services and generate ideas, but they remain tools used by the human inventor who conceived the claimed invention.”

“When one natural person is involved in creating an invention with the assistance of AI, the inquiry is whether that person conceived the invention under the traditional conception standard,” as detailed in the Governing Legal Standards section of the revised document. ...

“The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used in the inventive process,” notes the agency. “There is no separate or modified standard for AI-assisted inventions.”

Reuters points out that “U.S. courts have determined that AI systems cannot receive patents for AI-generated inventions, but have not yet considered when a person can receive patents for inventions conceived with the help of AI.” ...

“In practice, I suspect this means applicants will lie about who made AI-generated inventions,” suggests Stanford Law Professor Mark Lemley on LinkedIn. “The PTO will let them, and those patents will be in trouble if and when they are enforced in court.”

See the full story here: https://www.etcentric.org/patent-office-updates-guidelines-for-inventions-created-by-ai/

China wants to lead the world on AI regulation — will the plan work?

...

In October, at a meeting of the Asia-Pacific Economic Cooperation forum, Chinese President Xi Jinping reiterated his country’s proposal to create a body known as the World Artificial Intelligence Cooperation Organization (WAICO), which would bring nations together as a step towards creating a global governance system for AI. ...

How does the Chinese AI ecosystem differ from those of other countries?

Encouraged by the government, Chinese firms tend to release models as open weight, meaning that they can be downloaded and built on. And compared with Western nations, China has less of a focus on making machines that could outsmart humans — often referred to as artificial general intelligence — and is instead concentrating on a race to use AI to drive economic growth. ...

Developers of public-facing AI-powered services must let Chinese regulators test their systems ahead of deployment, says Ng. The result is that models such as those developed by the Hangzhou-based company DeepSeek, which found world fame with its R1 modelearlier this year, are among “the most regulated in the world”, says Joanna Bryson, a computer scientist and researcher in AI ethics at the Hertie School in Berlin. Despite this, the authorities often take a soft approach to enforcing that regulation, says Angela Zhang, a law researcher and specialist in AI regulation at the University of Southern California in Los Angeles. ...

By contrast, the United States has no comprehensive legislation on AI at the federal level, and in January President Donald Trump revoked an executive order aimed at ensuring AI safety. ... The European Union approach has been to classify AI systems by risk level, ...

What has China proposed?

WAICO would be a way for countries to coordinate AI governance rules while “fully respecting the differences in national policies and practices” and championing the global south, Chinese officials have said. ...

See the full story here: https://www.nature.com/articles/d41586-025-03902-y

Trump aims to boost AI innovation, build platform to harness government data

U.S. President Donald Trump on Monday signed an executive order to launch a government-wide effort to build an integrated artificial intelligence platform to harness federal scientific datasets to train next-generation technologies.

The effort, dubbed the Genesis Mission, aims to transform scientific research and speed scientific discoveries by using massive government scientific datasets "to train scientific foundation models and create AI agents to test new hypotheses, automate research workflows, and accelerate scientific breakthroughs." ...

Energy Secretary Chris Wright noted the massive private-sector investment in AI but said the government wanted to pivot those efforts to "focus on scientific discovery, engineering advancements, and to do that, you need the data sets that are contained across our national labs."

The order pays particular attention to U.S. national, economic, and health security, including biotechnology, critical materials, nuclear fission and fusion energy, space exploration, quantum information science, semiconductors and microelectronics. ...

See the full story here: https://www.reuters.com/business/trump-aims-boost-ai-innovation-build-platform-harness-government-data-2025-11-24/

We Induced Smells With Ultrasound

We pointed an ultrasound probe at the scent-processing region of the brain to obtain different sensations. Different focal spots corresponded to different smells, which we’ve replicated first-try on two people and validated with a blind trial. ...

Instead, we found that you can place the transducer on the forehead and aim the ultrasound downward towards the olfactory bulb. While this isn’t a perfect solution because the frontal sinuses can weaken the signal, careful device positioning above the sinuses still allows us to reach our general target region. ...

We have managed to induce four different sensations, all of them in two people:

The sensation of fresh air, with a lot of oxygen

The smell of garbage, like few-day-old fruit peels

An ozone-like sensation, like you're next to an air ionizer

A campfire smell of burning wood

We distinguish between a smell and a sensation here because, subjectively, they feel different. The smells are strong and localized to the noise, almost like you could sniff around and find the source. The sensations are more diffuse: a weak, slow-onset impression of a smell, often paired with other (likely placebo) feelings, such as a light tingling on the face. ...

The reason stimulating olfactory sensations is interesting is not just "VR for smells", as one might initially assume. The nose has 400 distinct receptor types, and we can distinguish subtle combinationsof their activations, so they could serve as a channel of writing directly into the brain, as a means of non-invasive neuromodulation.

The olfactory system potentially allows writing up to 400, if not 800 due to two nostrils, dimensions into the brain. That is comparable to the dimensionality of latent spaces of LLMs, which implies you could reasonably encode the meaning of a paragraph into a 400-dimensional vector. If you had a device which allows for this kind of writing, you could learn to associate the input patterns with their corresponding meanings. After that, you could directly smell the latent space. A bit of ultrasound, a breath in - and you understood a paragraph. ...

People are able to develop synesthesia - being able to hear colors and see smells, and it might be possible to extend that to semantics. However, at this stage it is speculative. ...

See the full story here: https://writetobrain.com/olfactory?utm_source=substack&utm_medium=email