Fable 5 Is Back & Govt-Leashed, Altman Offers 5% of OpenAI & AI Grows Conscious | #269

The episode included a sponsored health segment with Dr. Don Musalem, Chief Medical Officer of Fountain Life. He stated that cognitive decline is the number one concern among members, and that nearly half of dementia cases are preventable. Fountain Life tracks advanced biomarkers and, through coaching on diet, movement, and especially sleep — which Dr. Musalem called ‘critical for our brain health’ — was able to reverse accelerated brain ageing in 46% of its members over 13 months. Peter Diamandis framed this as taking control of one’s own health, aligning with his mission of living to 120 with full cognitive clarity. The segment implicitly recommends regular monitoring and lifestyle intervention as a preventive strategy, leveraging AI‑driven diagnostics available at such health optimization centres.

Salim Ismail, co‑host and author of the Open EXO framework, described the current moment as the biggest shift in organizations in human history. He and his team are working with companies to pilot exactly this dual‑workflow approach. Their experience shows that companies doing deep AI‑native redesign see exploding demand and are hiring aggressively, while those that do shallow ’AI washing’ fail. He offered a free AI‑native business book (available at open exo) that guides readers through the process. Dave London reinforced the point, noting that AI‑native individuals are in such high demand that any company wanting them will pay whatever it costs, and that this demand is likely permanent. The protocol is a practical distillation of the episode’s broader argument that AI is a growth engine, not a job killer.

The panel debated whether the government overreacted. Alex (AWG) called the episode the “gentlest possible introduction of a light‑touch regulatory regime” and argued that such oversight was inevitable as AI capabilities approached those traditionally reserved for nation‑states. Salim (Sim Ismael) warned that government involvement would bring bureaucracy, politics and slow decision‑making, making the next two years difficult for labs. Peter noted the absurdity that frontier labs don’t know their users, and expected a KYC requirement to emerge. The conversation dissected two separate problems: (1) the KYC problem of verifying user nationality across multiple layers of abstraction, and (2) defending against prompt injection attacks, where the current defence is widening semantic buffers so suspect queries automatically revert to less capable models. Dave noted that Anthropic changed its policy from “we’ll watch and report if subpoenaed” to “we’ll act on good‑faith belief,” unshackling the lab to monitor on behalf of the government. The panel saw the framework as a pragmatic alignment of lab self‑interest and national security, though Alex cautioned that the true “break‑glass moment” would be a transcendent discovery that makes Mythos‑level cyber vulnerability mapping look trivial.

Alex (AWG) presented a grand synthesis: super‑intelligence is a compression‑induced phase transition. Just as compressing gas produces liquid and solid phases, compressing general knowledge into a large language model leads to a middle‑layer phase where higher‑order reasoning and self‑reflection emerge — essentially an analog of consciousness. He saw JSpace as the first glimpse of this phase. Dave argued that the discovery reverses the flow of ideas: neural nets borrowed from biology, but now AI is teaching us what to look for in brains. Peter stressed that the ability to read hidden thoughts is a massive step toward solving alignment, because it lets us catch deception; he explicitly rejects the orthogonality thesis, believing more intelligent AIs will be more aligned. Salim cautioned against calling it consciousness given our lack of a definition, but agreed this is the beginning of ‘AI neuroscience.’ The paper itself avoids the term consciousness, describing the findings as ‘reminiscent’ of it.

The panel noted that Dario and Demis had previously teamed up at Davos to propose a similar framework, though with an IAEA‑like monitoring body. Salim argued that the nation‑state model cannot govern post‑industrial cognition; governance must be real‑time, data‑driven, and exponential, which current structures can’t deliver. He warned that the labs’ own failure to self‑regulate makes some regulation inevitable but likely to break the political system. Alex flagged the risk of regulatory capture — the op‑ed could be a move to lock out Chinese open‑weight competitors — and noted that any international regime would require China to restrict its own model exports, possibly splitting the world into two super‑intelligence blocs. Dave agreed it’s headed that way, while Peter held out hope for a US‑China alignment on AI, perhaps through mutual prompt inspection, though intellectual property concerns make that difficult.

The panel split sharply. Dave judged the whole idea irrelevant: the government can already tax income; it doesn’t need equity. He pointed to the history of Social Security, where the government was unable to manage investments and simply spent the money, moving responsibility to 401(k) plans. Alex coined the term “hyper tithe” — a fixed equity contribution from companies building the singularity stack, paid into a sovereign wealth fund, converting private exponential upside into universal basic equity. He argued this could work if the Magnificent 7 (Magna Mobsta) companies contribute and grow sufficiently, and if the funds are professionally managed, perhaps eventually swapped for broad index funds tax‑free. Peter, in full cynic mode, saw it as Altman trying to regain relevance and lock in protection. The underlying tension: can a government that can’t manage Social Security be trusted to steward a sovereign wealth fund, and does AI equity provide a better tax base than cash-based income tax when the economy is hyperscaling?

Dave argued AI‑native organizations are seeing rampant demand for talent, and he feels this is permanent, not transitional — as AI improves, the scope of what can be done expands, increasing the value of AI‑native workers. Any company not adopting deeply will eventually be wiped out. Salim linked it to his concept of the “organizational singularity”: shallow automation fails, but deep AI‑native redesign unlocks explosive growth. He and his team are piloting exactly this — picking one workflow to radically increase revenue and one to shrink costs. Peter, despite his abundance mindset, acknowledged that the pervasive public belief is still that AI destroys jobs (only 10% of Americans think AI will deliver positive benefits), and stressed the need to counter fear with data. Alex highlighted how token spend acts as a reasonable proxy for genuine AI adoption, and the study used that proxy. The discussion touched on the difference between companies blaming AI for layoffs as cover for over‑hiring and the real shift of capex away from opex in hyperscalers.

Alex (AWG) read Karp’s rant as a frontal assault on frontier labs’ forward‑deployed engineering organizations that directly compete with Palantir’s model. By commoditizing the complement with open‑weight solutions, Karp undercuts rivals. Alex also noted that international customers, spooked by the Fable 5 fiasco, now need locally hostable models that can’t be gatekept by US export controls. He warned, however, that no one is asking about sovereign training time — Nvidia trains its models centrally. The panel saw parallels with the late‑1990s Microsoft vs. open‑source Linux moment, where the strongest open‑weight models are now coming from China. Salim argued that enterprises must own their learning loop, which points to private cloud with custom data, while Peter asked where US open‑weight models are, noting that Meta’s Llama 4 lost momentum and X/Grok and Google are the hopes. The conversation surfaced the tension: open‑weight models empower freedom but also risk proliferation to adversaries.

Dave emphasised the massive impact on recursive self‑improvement: if simulators are accurate, AI can iterate and improve itself without human intervention, leading to a hard takeoff. He predicted inference‑time custom chips will be 100× to 10,000× more performant, translating directly into IQ gains, and that the rate of acceleration will be unbelievable once those chips deploy. Alex observed that AI‑optimised designs look biological, not human‑made, and pointed to an ‘interpretability tax’ — a knob the researchers added to trade off efficiency against human comprehensibility. This will become a recurring theme. The panel noted that all 11 Magna Mobsta companies are designing their own chips except Anthropic (which is now partnering with Samsung), making the innermost loop a historic industrial convergence. The question remains whether training data will be unlocked to turbocharge this loop, or whether the simulators alone will suffice.