Bernie Demands the Labs Stop, Wall Street Turns GPUs Into Bonds, Grok 4.7 Takes #1 with Emad Mostaque | EP #279

Grok 4.6 was described as newly launched and scoring 61 on the Artificial Analysis Intelligence Index, tied with GPT-5.6 Sol. Its focus is long-running agents that research, code, analyze, build products and verify their work; quoted pricing was $2 per million input tokens and $6 per million output tokens. Grok 4.7’s arrival in two weeks was a rumor, but Emad Mostaque predicted it could surpass Opus and take the top position. He said the model would grow from Grok 4.5’s 1.5 trillion parameters to 2 trillion, with Grok 5 planned at 6 trillion and later 10 trillion. Alex said his view that Grok 4.6 substantially reflects Cursor-style post-training was an interpretation, not insider information. Emad argued that training on SpaceX physics and engineering knowledge could help Grok deliver more than near-frontier performance.

Nvidia’s partnerships with Apollo, BlackRock, Blackstone, Brookfield and KKR are intended to mobilize more than $500 billion in third-party capital for AI infrastructure. That is not Nvidia borrowing: the proposed structures connect AI customers with institutional and private investors financing GPU purchases and data centers, potentially against individual compute clusters. Speakers said GPUs generally depreciate over three to five years but can continue generating revenue, leaving investors exposed if a technical breakthrough makes existing hardware less valuable. Alex said compute-backed financing at the scale of roughly $7 trillion in AI infrastructure spending would need sophisticated hedging, including futures and derivatives. He disclosed a financial interest in Orn, which he said is enabling this market. Others cautioned that exponential technological change makes depreciation hard to forecast and that compromised ratings could still precipitate a collapse.

Bernie Sanders formally urged the CEOs of Anthropic, Meta and OpenAI to pause AI development, arguing that capabilities had crossed a critical threshold. He said AI had been used to create a new virus and warned of the possibility of a biological weapon causing tens of millions of deaths, asserting that companies’ own safety commitments meant “that moment is here.” The panel disagreed over whether a pause could work: Peter and Emad argued that other countries and open models would continue advancing, making defensive capabilities necessary. Alex, who said he signed a pause letter two years earlier, now considers one too late and potentially harmful to growth and prosperity. He argued for controls preventing AI-assisted biological weapons rather than limits on intelligence, warning that pauses could let less-cooperative labs catch up and intensify competition when restrictions end.

The biological-risk debate included Stanford researchers’ use of the open-source, 40-billion-parameter Evo II to design about 300 bacteriophage DNA sequences. Sixteen synthesized designs produced viable phages capable of infecting E. coli strains that had not developed natural resistance. Speakers stressed that Evo II can design sequences but cannot make viruses, and that synthetic biology and bacteriophage research long predate generative AI. They described the work as dual use, with research and medical potential as well as risk. Proposed safeguards included oversight of DNA and RNA synthesizers and inexpensive, widespread sequencing to detect novel sequences and monitor their spread. The panel also warned that restricting AI tools could slow disease research, and debated global monitoring of prompts and tokens without resolving who would control access.

Mostaque said Higgsfield reached a $700 million revenue run rate in about a year and a half. He put CDANCE 2.5’s cost at about $3 for a 30-second clip, while models 10 to 20 times cheaper deliver roughly 90% of its quality. The host described Higgsfield’s 110-minute AI-generated feature Cully Hill Boys as a four-week production by a 28-person team, made with licensed celebrity likenesses. Its reported $2 million budget included $1 million in compute. Mostaque estimated that a comparable movie might require around $100,000 in compute today, potentially dropping to about $10,000 by the new year. He also said LTX can generate a 10-second high-definition clip in seven seconds on a laptop, and that licensed likenesses and other material could help avoid reshoots.

Peter described the $101 million healthspan XPRIZE as an effort to reverse age-related losses in cognition, muscle and immune function, rather than merely improve blood-test results. It drew 800 teams; $20 million had been awarded, including $1 million to each of 10 teams, with an $80 million grand prize due March 2. Finalists must test their approaches in human trials with control groups, probably involving around 150 people, and Peter expected a winner by 2030. He tied longevity escape velocity to demonstrated functional-age reversal, not a particular date. Speakers argued that advances in sequencing, molecule-making and AI measurement may make aging research tractable, while noting the field may lack enough qualified people. They cited $20 trillion in annual age-related disease costs against a $120 trillion global economy as a reason prevention could have major economic and productivity benefits.