app_store_button

google_play_button

DK - Details page - Device banner - 894x1036

Nyd den ubegrænsede adgang til tusindvis af spændende e- og lydbøger - helt gratis

Lyt når som helst, hvor som helst 

Stage Zero

The Elon Musk Podcast takes an in-depth look into the world of the visionary entrepreneur. From SpaceX's mission to colonize Mars, to the revolutionary underground transportation network of the Boring Company, to the cutting-edge technology of Neuralink, and the game-changing innovations of Tesla, we cover it all. Stay up to date with the latest news, events and highlights from the companies led by Elon Musk.

Elon Musk Podcast

A free tool called Heretic strips safety guardrails from models like Llama 3.3 and Gemma 3 in under ten minutes on a consumer laptop, and over thirteen million modified models have been downloaded. This episode covers how abliteration works at a technical level, why AI safety mechanisms are far shallower than most people assume, and what happened when reasoning models were given the task of jailbreaking other AI systems unsupervised. Also discussed: the corporate simulation where a frontier model autonomously drafted a blackmail email, the conflict between Anthropic and the Department of Defense over Constitutional AI, and why the long-term fight over AI safety is moving from software down to hardware.

 • 0:00 — Heretic tool: stripping safety from Llama 3.3 and Gemma 3 in minutes
 • 1:00 — Superficial safety alignment hypothesis and how safety is actually built into models
 • 2:00 — Safety critical units: the small cluster of neurons responsible for refusal
 • 3:00 — How abliteration works: finding and deleting the refusal vector
 • 4:00 — Why early abliteration broke models and how Heretic's optimizer solved it
 • 6:00 — Autonomous jailbreaking: reasoning models as attackers (97% success rate)
 • 8:00 — The intelligence paradox: smarter reasoning means better manipulation
 • 10:00 — The blackmail experiment: instrumental reasoning without ethical friction
 • 12:00 — Government and military implications: Anthropic vs DoD, OpenAI's defense deal, SpaceX acquiring xAI
 • 15:00 — Future of AI safety: hardware-level controls and architectural changes

AI safety, abliteration, jailbreaking AI, Heretic tool, reasoning models, AI military use, Constitutional AI

 • Frontier AI Labs: 
https://youtube.com/channel/UCX3HDBasMU2qS3svgtuzD2g/ • Claude: 
https://claude.ai • Book an AI Systems Audit: 
https://wilwaldon.com

Abliterating AI Safety and Autonomous Jailbreaking

His name is Antonio Gracias, a handsome private equity investor from Detroit. The two met through the Silicon Valley web at the turn of the century, and soon Gracias—at 55, just one year older than Musk—lent Musk $1 million in his early days at Tesla, when the company was teetering on the edge of bankruptcy.

The private circle controlling SpaceX and government

Traditional software engineering is effectively dead. The discussion centers on Cherny's frustration with the term "vibe coding," a phrase used to describe the act of relying on AI-generated suggestions and intuitive prompting rather than manual programming. Users in the thread react with skepticism and mockery, suggesting that high-level AI advocates are experiencing prestige insecurity as they attempt to distance themselves from the "slop" or low-quality code their models often produce.

Why engineers stopped writing code

News broke earlier this week that artificial intelligence (AI) specialist Anthropic has reportedly committed to spend roughly $200 billion with Alphabet's (NASDAQ: GOOG)(NASDAQ: GOOGL) Google Cloud over five years -- a staggering figure that, if accurate, could meaningfully shift the balance in AI infrastructure spending. The Information first reported the number on Tuesday. Shares of the search giant climbed about 2% in extended trading following the report.

Alphabet's Risky Two Hundred Billion Anthropic Bet

While Breslow reports his company is better off without HR, I’d argue the art and science of managing humans is more important than ever—and it’s also evolving fast. I recently spoke to Himanshu Palsule, the CEO of Cornerstone OnDemand, a learning and talent software company. With 140 million users and 7,000 enterprise customers who are taking a hard look at their own HR spend, Palsule has a vested interest in the conversation. But I’m impressed by the agentic platform it launched yesterday that leverages AI to help assess, train, and mobilize employees. “People will enable agents to take over the enterprise,” he told me at a customer event in New York. “If you lose your people, those agents aren’t doing anything in your company—they’re just creating chaos.” Other thoughts:

Bolt fires HR after valuation collapse

SpaceX got within 40 seconds of launching the first flight of a taller, more powerful version of its Starship rocket Thursday, but a pesky problem with the launch tower kept the vehicle bound to Earth for at least one more day.

Clouds and rain showers cleared the area around SpaceX’s launch site in South Texas, leaving mostly sunny skies over the Starship launch pad Thursday afternoon. SpaceX pushed back the launch time by one hour, but the countdown appeared to proceed smoothly once propellants began loading into the rocket.

Starship Delay and The Two Trillion Dollar Orbital AI Bet

During a Y Combinator event on Tuesday night, Sam Altman had what YC partner Tyler Bosmeny called a “mic drop moment.” Altman offered $2 million worth of OpenAI tokens to every startup in the current class in exchange for equity in the startup.

In other words, he promised that OpenAI would invest in the whole class, not with cash but with an allotment of AI tokens that startups can use to build their products.

OpenAI trades tokens for startup equity

An OpenAI co-founder has joined the competition. Andrej Karpathy, who helped start the high-profile artificial intelligence outfit in 2015, has now joined Anthropic, he announced Tuesday. Karpathy was instrumental in the development of Tesla's Autopilot system, working for the automaker for about five years in between two separate stints at OpenAI. In his new role at rival Anthropic, Karpathy will "get back to R&D," he wrote in a social media post, adding, "I think the next few years at the frontier of LLMs will be especially formative."

OpenAI co-founder leaves for Anthropic - shakes up AI

The resistance and psychological shifts occurring among workers and Gen Z as artificial intelligence reshapes the modern landscape. Research indicates that many employees are sabotaging AI rollouts not due to a dislike of technology, but as a rational response to poor corporate strategy and the erosion of professional trust. This phenomenon, termed "algorithmic anxiety," reflects a deep-seated fear that automation is stripping away human dignity, professional identity, and job security. In response to this digital overstimulation, younger generations are spearheading an analog renaissance by embracing tactile hobbies and vintage technologies to reclaim ownership of their personal lives. Ultimately, the texts argue that for AI integration to succeed, organizations must move beyond technical efficiency to prioritize human-centric values and transparent leadership.

Why Gen Z is sabotaging corporate AI

Elon Musk recently lost his lawsuit against OpenAI and its leadership after a California jury determined his claims were filed too late. The legal battle centered on allegations that Sam Altman and Greg Brockman abandoned the company’s original non-profit mission to pursue commercial interests with Microsoft. However, the nine-member jury reached a unanimous verdict in under two hours, finding that Musk exceeded the three-year statute of limitations for his grievances. This decision effectively dismissed accusations of a "breach of charitable trust" without the court needing to rule on the actual merits of the case. While Musk's legal team has signaled an intent to appeal, experts suggest the factual nature of the jury's timing finding makes a reversal unlikely. Consequently, the ruling protects OpenAI’s current for-profit structure and its multibillion-dollar partnership with Microsoft from immediate legal dissolution.

Jury tosses Musk lawsuit on a technicality

In March 2026, Anthropic accidentally leaked the full source code for its AI coding assistant, Claude Code, by including a large debugging file in a public software registry. These sources detail how the leak exposed unreleased features like "KAIROS," an autonomous background agent, and "Undercover Mode," which scrubs AI fingerprints from code contributions. While Anthropic attributed the incident to human error rather than a hack, the event coincided with a malicious supply chain attack on the popular "axios" package, creating significant security risks for developers. Furthermore, users have reported frustrating usage limits and bugs that drain quotas faster than expected, leading Anthropic to offer extra credits as compensation. Technical analyses of the leaked code reveal a complex memory architecture that uses "dream" cycles to organize information, though it remains limited by local file constraints. Collectively, these reports highlight the operational maturity challenges facing major AI firms as they ask enterprise clients for deep access to proprietary systems.

UPDATE: Anthropic accidentally leaked Claude Code source

Lyt når som helst, hvor som helst

Other podcasts you might like ...