The White House launches Genesis Mission, a Manhattan Project-scale moonshot to unify US supercomputers and federal datasets, aiming to double scientific productivity via AI within a decade.
2
Anthropic's Claude 4.5 outperforms human engineers, scores 52% on reasoning benchmarks without reasoning tokens, and demonstrates multi-agent orchestration, suggesting recursive self-improvement is underway.
3
Google rolls out Ironwood TPU with 4x performance, offering cloud rental to compete with NVIDIA, while Amazon commits $50B to AI infrastructure including 2.2GW data centers for government and Tranium 2 inference chips.
4
The convergence of vertical specialist AI agents (ChatGPT Shopping) and BCI breakthroughs (Paradromics human trials) accelerates the path toward post-scarcity and universal AI-assisted abundance by 2035.
Protocols
Concrete recipes — what, when, how much, and why
3 items
Multi-Agent Swarms with Claude 4.5
WhatCombine Claude Opus 4.5 as orchestrator with smaller models like Haiku or Sonnet to achieve 88% multi-agent support.
WhenWhen undertaking complex coding or research projects that benefit from parallelization and specialized sub-agents.
For whomDevelopers, researchers, and entrepreneurs building AI-powered workflows.
WhyClaude 4.5 alone scores 75% multi-agent; with smaller models it reaches 88%, demonstrating effective agent oversight.
CaveatsRequires careful prompt engineering and may have increased coordination overhead.
'Opus by itself it scores 75% in multi-agent... when they combined it with haiku which is a very lowcost agent or sonnet it got up to 88%. So, Opus is a really good orchestrator of agents.' (Immad)
Universal Basic AI Agent (Jarvis)
WhatEquip every person with a personal AI assistant that navigates external services, handles tasks, and coordinates access to resources.
WhenAs soon as technologically available; advocate for government and private sector deployment.
For whomEveryone, especially the disadvantaged.
WhyTo prevent widening inequality, AI can act on behalf of individuals, making invisible populations visible and ensuring access to food, healthcare, and financial services.
CaveatsPrivacy and sovereignty concerns must be solved; the agent must be loyal to the user.
In the AMA on poverty, Immad and Salem argued for a three-part solution: coordinate with AI, give everyone universal AI, and layer services on top. The pace of job displacement (e.g., customer service agents gone within 2 years) demands urgent action. Universal AI agents would act as a 'Jarvis' looking out for individuals, helping them navigate bureaucracies and access resources. This mirrors historical New Deal-scale responses but powered by AI. The coordination problem (enough resources, poor distribution) is solvable with AI social scientists and individualized AI companions.
Personal experience
Immad mentioned his Sage project as a top-down policy AI initiative, paired with the universal AI agent vision.
'Coordinate with AI, give everyone universal AI and then layer services and coordination on top of that.' (Immad)
Also said
“'The reason they don't know is because people are invisible, particularly the poorest of people. But the pace at which this is going to come over the next few years is going to be so intense that governments need to take a big step forward and say a we need to use AI to coordinate this, b we need to get AI to the people...' (Immad)”— Underlines urgency and invisibility problem.
DRAM Investment
WhatConsider investing in companies that produce DRAM (memory chips) as prices have increased 5x and demand for large context windows drives further shortages.
WhenCurrent market opportunity.
For whomInvestors seeking asymmetric returns in AI hardware supply chain.
WhyEvery large AI inference and training workload requires high-bandwidth memory; Google's TPUs and large context windows rely on efficient memory. DRAM prices surged, making it a strategic supply-chain play.
CaveatsCommodity chip cycle risk; not a guaranteed long-term.
Personal experience
Salem recalled a previous episode where they predicted DRAM as a short supply and price spike, now confirmed by Immad's data.
'So right now actually DRAM prices have gone up by about five times.' (Immad)
What's new
Personal practice updates, fresh positions, predictions
7 items
Anthropic Claude 4.5 (Opus): Recursive Self-Improvement Milestone
Claude 4.5 uses 76% fewer tokens, outperforms human engineers on coding benchmarks, achieves 52% on reasoning without reasoning tokens, and shows 88% multi-agent orchestration.
Why this matters: Signals that AI researchers are being outperformed by models on key tests, marking the near arrival of recursive self-improvement where more compute goes to AI research than human research.
Background
Previous models required reasoning tokens and chain-of-thought to match human-level reasoning, driving up cost. Claude 4.5 achieves high performance with straightforward output, dropping cost 67% to $25 per million tokens.
The model scored 52% on Scale's SWE-bench Pro (a hard reasoning benchmark) without using reasoning tokens, shocking observers. Combined with Haiku or Sonnet, its multi-agent support jumped to 88%, proving AI can effectively orchestrate other agents. The cost reduction and efficiency mean coding and token generation will become ubiquitous; it may one-shot most codebases next year. Anthropic's incoming employees on the performance team are now outperformed by the AI on key homework assignments, indicating that the innermost loop of civilizational progress—models that research and code better versions of themselves—is in motion. This loop will spin out to solve robotics and physical-world automation, driving radical growth. In the near term, individuals will trivially generate complex software, video games, and applications from a single prompt.
Personal experience
'I of course have my emails whenever these cogen models come out. one of my other non-cyberpunk FPS evals is asking it to to see if I can one shot a Mario style sidescroller and it did a beautiful job.' (Alex)
'It's that that's the canary that we're we're imminently if not already given that this model was arguably pre-trained based on the the data cut off date several months ago. We're we're now entering the moment of recursive self-improvement.' (Alex)
Also said
“'This model without reasoning scored 52%. Without even reasoning tokens which I think was the most shocking thing for me.' (Immad)”— Highlights the unexpected efficiency.
“'Everyone was saying, well, agents can't look after other agents. This is the first agent that prove or the first AI that provably can.' (Immad)”— Reveals breakthrough in agent swarms.
ARC AGI Leaderboard: Visual Program Synthesis Solved, Need Harder Benchmarks
ARC AGI 1 and 2 benchmarks, which test visual reasoning and program synthesis, are being saturated; models now achieve superhuman cost efficiency, prompting a shift to dollar-based economic benchmarks.
Why this matters: Demonstrates that AI can now visually reason and synthesize programs as competitively as humans, solving a historically hard category. The next frontier is real economic work.
Cost of intelligence is being driven to zero, including super intelligence. Opus 4.5 and a company named Poetic have shown superhuman performance on ARC AGI 2, meaning visual pattern recognition and program synthesis out of pixel inputs are solved. The authors of ARC are reportedly asking 'what on earth do we do now'. The next benchmarks are dollar-based: vending bench, trading benchmarks that measure real economic value generation. This shift indicates we're moving from models that perform discrete tasks to models that can do economic work. It also underscores the need for harder evals, and that within a year, an individual entrepreneur with a fleet of AI agents could build a billion-dollar business, with examples already emerging like 47 AI startups launched in a month.
'What we're seeing for the first time between the Opus 4.5 results that are demonstrating breakthrough state-of-the-art level cost efficiency of visual program synthesis and then also an earlier result... we're seeing visual program synthesis start to get solved and the world needs harder benchmarks.' (Alex)
Also said
“'The next benchmarks are dollars. So you have vending bench and some of these other benchmarks where it's like how much money can they earn?' (Immad)”— Frames the pragmatic shift in evaluation.
ChatGPT Shopping: Vertical AI Agent Replaces Search and Recommendation Economy
OpenAI launches a vertical shopping agent using GPT Mini, achieving 64% prediction accuracy on consumer purchases, directly challenging affiliate blogs, YouTube reviewers, and Amazon's recommendation engine.
Why this matters: Demonstrates proliferation of specialist, not just generalist, AI agents; positions AI as the new middleman in e-commerce, with companies like Amazon responding with Rufus to capture intent.
Background
Previously, product research was dominated by search engines and affiliate content. AI now can compare products, infer user intent, and eventually purchase autonomously, leveraging real-time conversation analysis.
This is OpenAI's third vertical specialist agent after Deep Research and Codex, moving beyond a singleton generalist model. Amazon's CEO Andy Jassy announced their shopping agent Rufus has 250 million users and estimates $10 billion incremental sales next year due to 60% higher conversion. The critical question is who captures user intentionality: the agent embedded in the search/retail platform (Amazon, Google) or a more charming, cross-platform personal assistant (Jarvis). The eventual aim is a personalized AI layer that understands gaze, conversation, and implicit desires, then purchases autonomously. This collapses the whole product research economy and turns consumer decisions into a compute problem.
'No, I what what you're describing SEM is I I I would argue like a computer use agent a CUA and Microsoft and and other major companies and Frontier Labs already have CUAs that are either about to be rolled out or have already been rolled out...' (Alex)
Also said
“'Andy Jasse, CEO of Amazon, recently said, I believe that their agent, Rufus, which I doubt any of us have actually used, has $250 million users, a shopping agent.' (Immad)”— Validates the agent trend with hard numbers.
“'...who comes up with the agent that's the most charming and engaging and licenses Paul Betin's voice you know for a Javis because that can then disintermediate everything and that's what the fight is on for now.' (Immad)”— Highlights competitive dynamic.
Google Ironwood TPU: Seventh-Gen AI Chip as Cloud Service Competes with NVIDIA
Google launches Ironwood TPU with 4x performance, offering it as a cloud service (not just hardware) to customers like Meta, marking healthy competition to NVIDIA's GPU dominance.
Why this matters: Accelerated compute becomes a fungible commodity with multiple suppliers, breaking CUDA architectural monopoly; Google's interconnects allow massive context windows.
Background
Previously, NVIDIA's CUDA and GPUs were the primary high-performance option. Google has matured its TPU line, originally built for search, to a seventh generation now rentable, with strong interconnect scaling up to 9000 chips.
The Ironwood TPU scales from 64 to 9,000 units with multi-data-center capability, which is critical for large context windows. DRAM prices have spiked 5x, and Google's architecture can use cheaper chips at massive scale for large context, giving Gemini an advantage with 1-2 million input tokens. Other competitors like AMD, Amazon's Trainium/Inferentia, and Cerebras create a heterogeneous ecosystem. Backpropagation in training remains a bottleneck, making training chips harder to commoditize than inference, but for inference, hardware is becoming fungible. The trend is toward accelerated compute as a commodity, which is good for the industry.
Personal experience
'We used thousands of TPUs a few years ago and from the V5s, this now has 10 times the compute.' (Immad)
'GPUs are now finally facing healthy competition. We see TPUs that are being both purchased according to this reporting as well as licensed and rented.' (Alex)
Also said
“'Google originally built these chips to power Google search and now they've matured to a point where they can offer it to everyone even hosted.' (Immad)”— Context on origin.
“'That means that Gemini has a million 2 million input tokens from video to audio to others. whereas it's still limited on other GPUs and that's going to become even more of a difference going forward.' (Immad)”— Shows strategic advantage.
Amazon's $50B AI Cloud for Government and Tranium 2 Inference at Scale
Amazon plans $50B for AI infrastructure, including a 2.2GW Indiana data center for government and Anthropic training, and 500,000 Tranium 2 chips for inference.
Why this matters: Government clouds historically under-supplied with GPUs; Indiana facility converts farmland to compute, signifying scale akin to Manhattan Project; Tranium 2 tackles inference bottleneck.
Background
US government agencies relied on multiple cloud providers but had limited GPU compute. AWS serves 11,000 agencies and is investing $125B in CapEx by end 2025.
Amazon's Project Rainier in Indiana transformed farmland into a compute facility for Anthropic's training and inference within a year, reflecting the 1939-like mobilization. Tranium 2 chips are equivalent to Hopper for inference but lag for large-scale training; they are used to break inference capacity constraints. As cloud giants differentiate, Amazon's strong government ties and massive scale position it as a key player in sovereign AI, providing lower-cost inference at scale. The discussion also notes that DRAM prices are surging, making inference-focused chips strategic.
'I was struck by the fact that the Indiana one uses is 2.2 gawatt of energy. That's like an unbelievable amount of energy for a data center.' (Salem)
Also said
“'Tranium 2 chips are equivalent to the hoppers and they're good for inference, but they're much more difficult to do the large scale training runs on.' (Immad)”— Technical differentiation.
“'the government clouds in many cases including with with AWS and otherwise the government has its own availability zones and they are notoriously under supplied when it comes to accelerated compute' (Alex)”— Background on need.
Space Launch Costs Drop: From $50,000/kg to $0.10/kg via Mass Drivers
With Starship reusable rockets, launch costs drop to ~$100/kg; future lunar mass drivers could bring it to $0.10/kg, unlocking asteroid mining and O'Neill colonies.
Why this matters: Orders-of-magnitude cost reduction makes solar system dismantling viable, addressing resource scarcity on Earth.
Background
Space Shuttle cost $1-2B/launch, $50K/kg; Falcon 9 reusable dropped to $2.5K/kg; Starship aims for $100/kg.
Peter Diamandis emphasizes that everything of value on Earth exists in near-infinite quantities in space. The electromagnetic mass drivers (Gerard K. O'Neill's concept) on the moon could accelerate payloads to lunar escape velocity using cheap solar electricity, dropping cost to $0.10/kg. This would enable disassembly of asteroids and construction of rotating O'Neill colonies for habitation. Alex jokingly says Saturn and the moon are due for dismantling, but notes we'll have the ability to protect Earth and recreate tides artificially. This technological progression turns physical launches into a software-like exponential path, echoing the broader dematerialization trend.
Personal experience
'the 9-year-old space geek in me is like super excited of what's coming.' (Peter)
'So all of a sudden we gain access to all the resources on earth. I'd like to remember remind people that everything we hold of value on Earth, metals, energy, real estate, all these things are in near infinite quantities in space.' (Peter)
BCI and Mind Uploading: Paradromics Human Trials, Ray Kurzweil's Timeline, fMRI Thought Decoding
Paradromics gets approval for human BCI trials with 200 bits/sec, 20x faster than Neuralink. AI-driven fMRI thought decoding and whole brain emulation progress suggest high-bandwidth brain-machine links by early 2030s.
Why this matters: Marks a competitive BCI space accelerating toward Ray Kurzweil's prediction; non-invasive fMRI foundation models and whole-brain emulation timelines are converging.
Background
Neuralink put BCIs in the spotlight; Paradromics used sheep rather than monkeys. Ray Kurzweil predicted high-bandwidth BCI by the early 2030s.
Alex notes that extrapolating the cost and size of a gigaflop of compute to the size of a brain cell yields 2045 for nanobot-based uploading. However, partial solutions using low-bandwidth fMRI (1-2 sec temporal res, 1mm spatial) can reconstruct thought, and groups like Meta are training foundation models from fMRI scans. Whole-brain emulation via destructive scanning (connectome) has been done for fruit flies, and mice are next; a full human connectome is expected within 5 years. Immad adds that AI can help solve the hard problem of consciousness, and that BCI will be one of the biggest investment areas due to geostrategic importance. Salem emphasizes that we don't need full bandwidth; partial bandwidth with diffusion models can reconstruct brain processes. The discussion also touches on Drexlerian nanotech assemblers arriving within 10 years.
Personal experience
'we did work at stability on Mind's Eye where we reconstructed images people saw from MRIs which is incredibly low bandwidth.' (Immad)
'I think as Elon said like the only way you're going to be able to keep up with the AGIS is to plug in.' (Immad, quoting Elon)
Also said
“'there are a number of groups, including Meta, by the way, that are training foundation models from fMRI scans. Arguably, those are like lowfidelity faximiles but non-invasive of human minds.' (Alex)”— Indicates real progress in non-invasive uploading.
“'I think we're likely to to get some AI solution to Drexlerian early style Drexlerian nanotech in the next 10 years.' (Alex)”— Timeframe for nanotech enabling BCI.
Recommendations
Products, supplements, and tools mentioned in the episode
1 item
Paradromics BCI
Product
Brain-computer interface company that tested in sheep and received approval for human trials, achieving 200 bits/sec, 20x faster than Neuralink.
Paradromics represents the next generation of high-bandwidth BCIs, with a different testing path (sheep vs. monkeys) and a speed advantage. It's part of a competitive landscape including Neuralink, Science (by Max Hodak), and Merge Labs. The technology is expected to be crucial for keeping pace with AI advancements.
vs alternatives
Compared to Neuralink (~10 bits/sec), Paradromics offers 20 times the data rate. Other approaches like Science use stem cells for a non-invasive biological interface.
'Paradromics has done their testing in sheep. And they've been approved to go into humans... they've been able to hit a speed about 10 times or actually 20 times faster than Neuralink.' (Peter)
Blitzy is an autonomous software development platform that uses thousands of specialized AI agents to understand enterprise codebases, plan, and generate 80% of development work autonomously, doubling engineering velocity.
DisclosureThis episode is sponsored by Blitzy; the host reads a promotional message.
Enterprises use Blitzy as a pre-IDE tool: it consumes requirements, provides a plan, pre-compiles code, and pairs with any coding copilot to complete the remaining 20%. The platform promises a 5x engineering velocity increase and infinite code context, targeting large codebases with millions of lines. It represents the emerging category of 'pre-develop' AI tools that shift the development lifecycle. The ad appears during the discussion of AI coding advances.
vs alternatives
Compared to standard coding copilots that assist line-by-line, Blitzy sits upstream, doing the heavy lifting of understanding and architecting before developers engage.
'Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their preide development tool, pairing it with their coding co-pilot of choice.' (Ad read)
A twice-weekly short email covering the top 10 technology meta trends that will transform industries over the next decade, aimed at founders, CEOs, and entrepreneurs.
DisclosurePeter Diamandis, the host, promotes his free newsletter.
Peter offers it as a no-fluff resource to get ahead of exponentials, available at dmmandis.com/tatrends. It's positioned as insider intelligence for disruptive innovators, similar to the podcast's mission.
vs alternatives
Unlike typical tech news, it focuses specifically on meta trends 10 years ahead.
Personal experience
'I write a newsletter twice a week... If you want me to share these meta trends with you... this report's for you.' (Peter)
'If you want me to share these meta trends with you, I write a newsletter twice a week, sending it out as a short two-minute read via email.' (Peter)
A framework that combines multiple AI models to achieve state-of-the-art results on SWE-bench Pro (45% with combination, 52% with solo model). It aims to build AGI and coordinate large-scale AI tasks.
DisclosureImmad Akhund, co-host, is the founder and mentions achieving top benchmark results.
Immad revealed that their framework reached the top of SWE-bench Pro using a mix of models, and that the Claude 4.5 evaluation showed 52% without reasoning tokens. The company is also working on Sage, a top-down policy AI for social coordination. This positions Intelligent Internet as both a technical and policy-oriented AI venture.
vs alternatives
Compared to single-model approaches or other agent frameworks, Intelligent Internet explicitly leverages model combination and a focus on real-world metrics like benchmark dominance.
Personal experience
'we made we got top of the swench pro benchmark... This is with intelligent internet framework...' (Immad)
'with intelligent internet framework using a combination of the other models. This model without reasoning scored 52%.' (Immad)
Lines worth pulling out — contrarian, specific, or perfectly phrased
2 items
'Everyone's going to become an investor. The entrepreneurs increasingly are going to be these AI agents that are identifying and solving valuable problems.' (Alex)
Contrarian view that humans won't need to be entrepreneurs; they'll invest in AI agent economies.
'the only way you're going to be able to keep up with the AGIS is to plug in.' (Immad quoting Elon)
Bold imperative for BCI adoption to remain relevant alongside superintelligence.
Sign in to share feedback
Tell us if this brief hit the mark or missed it — feedback feeds back into the next iteration of the prompt.
Reading is free for everyone. A free account adds the personal layer: save protocols, follow experts, and see how the other experts weigh in on this same topic.
Educational summary of the cited expert source — not medical advice. Open the source recording linked above and consult a qualified physician before acting on any protocol.
