X Follow-Graph Synthesis

Wild 24 hours for AI and lots of different proposals have been made. TLDR; the only *tangible* new fact is that OpenAI and Anthropic are going to have embedded 3rd party evaluators from unknown organizations with Dario floating METR as a possibility. Having 3rd party evaluators is smart as there is no Section 230 style liability shield for model outputs and showing a “duty of care” will be important in future litigation. Several internet companies might have gone bankrupt without Section 230 so limiting liability really matters. There are minimal investment implications from this single new fact, but I do think that for anyone who wants a “smoother for longer” cycle then most constraints are good: wafers, watts, real rates and spreads. Excessive regulation is a different matter but I don’t think we are anywhere close to this even if the vector changed over the last 24 hours. To summarize the events: Dario made the most maximalist proposal of the weekend: embedded 3rd party evaluators, a national regulatory regime for models beyond a certain capability/ingredient threshold, a broad international regulatory pact between democracies, stricter limits on compute/distillation for China and then a different international regulatory regime that encompasses China. Before there is a national regulatory regime, he wants a Sherman act waiver so that Anthropic can safely coordinate with OpenAI and other frontier labs without antitrust fears. TBF, this latest proposal is much less maximalist than some of his prior proposals like “Policy on the AI Exponential,” where he advocated for an FAA for AI. I believe he is sincere in his beliefs. And despite all the protestations, all of this would also probably be good for his business over the long-term. Sam agreed that embedded 3rd party evaluators were a good idea and stated they would implement them. Again, this is smart as should help limit future liability. Elon said “Dario is right” and later specified that “Dario is right that there should be some oversight. Peer review of AI by competitors is the right way to start this off.” This would be a MPAA like self-regulatory structure for AI with regular calls between the labs plus a process where each new model is evaluated for safety by competitors for a 1-2 week period before being released. That is *wildly* different from Dario’s proposal and in-line with what David Sacks has been proposing. Elon also stated that nothing was going to slow down open-weight models. Demis said that Dario’s essay was a “step in the right direction.” Dario also said that he was also open to Demis’ idea of a FINRA like self-regulatory structure as part of his proposal. David Sacks had a thoughtful post where he said that Dario and Sam should pace unilaterally, called the antitrust waiver a cartel request and denied that METR was truly independent given their ties to Anthropic. Sriram Krishnan, former White House AI advisor, noted that it would be important to have the 3rd party evaluators come from independent organizations that are not affiliated with any lab, which is basically an indirect statement about the relationship between METR and Anthropic which Sacks was explicit about. Clem from Hugging Face said they were open to being a neutral 3rd party evaluator, which is interesting especially if Jensen was consulted before that post. Alexander Wang from Meta noted that alignment would be an increasing focus going forward. An executive order seems likely after all this and the language in this EO is going to be really important. It is possible to democratize and distribute AI broadly and safely without centralizing it in the hands of a few corporations who might each become more powerful than any single government. I do not want a few humans in control of intelligence. I want us all to have our own intelligences that reflect our own values and human variation in all of its richness. Intelligence distribution over intelligence centralization FTW.

$AMD is now arguing that agentic AI is structurally expanding CPU demand, and that AI infrastructure should increasingly be viewed as a rack-scale system spanning CPUs, GPUs and networking rather than GPUs alone. - AMD expects agentic workloads to drive significantly more CPU-intensive work across orchestration, databases, APIs and tool execution. - That is also pushing AMD to think about a much larger server CPU opportunity while expanding Helios into a broader CPU + GPU + networking rack-scale architecture. > What makes this particularly interesting to me is that I wrote about this structural shift two months ago. > As AI moves from Training → Inference → Agentic, the GPU ratio changes, and the system begins to split into specialized racks rather than relying on one general-purpose compute rack to do everything. And that leads to the next question: What connects all these racks? This is no longer just a CPU-vs-GPU story. As CPU, GPU, accelerator and storage functions become increasingly disaggregated, more data has to move between them, making interconnect a core part of the system architecture itself. In my July piece, I examined where NVIDIA and AMD are cutting the inference rack, and why those different architectural choices could determine where the money flows across copper, pluggable optics and CPO. Same Split, Opposite Directions: Where NVIDIA and AMD Cut the Rack Decides Where Optics Gets Paid. https://t.co/Z0PbHp0rpu

- $AMD and $XNDU co-designed Backline, an open compilation platform for quantum-classical integration with microsecond-level latency. - Backline enables Python-native quantum error correction deployment across $AMD FPGAs, GPUs, and CPUs without requiring low-level code rewriting. - The platform addresses tight timing constraints of photonic quantum systems by automating heterogeneous hardware compilation and routing. > Backline addresses a critical infrastructure gap in quantum systems by enabling production-grade quantum error correction with real-time feedback loops on classical accelerators. > This reduces the engineering burden for quantum-classical integration, potentially accelerating deployment of fault-tolerant quantum systems that rely on sub-microsecond latency between quantum processors and classical hardware.

🧵 If you want to see the diminishing returns in AI, consider that in OpenAI's recent RSI blog post: 124x increase in token output Led to 7x increase in lines of code Led to 1.6x increase in experiments run Led to (my estimate, based on power laws) Perhaps 10% https://t.co/XqIF3O1LPG

> This is exactly why I've been talking so much about microLED optical interconnect lately. ams OSRAM ($AMS.SW) is now building out its Digital Photonics Interconnect team in the Bay Area, hiring across System Architecture, Application Engineering, Product, and high-speed characterization. This is becoming harder to dismiss as just a device demo. They are building the system, test, and commercialization infrastructure around microLED-based optical links. The question is no longer: \Can microLED work for optical interconnect?\" The more important question is: \"How far can microLED penetrate AI scale-up and scale-out?\"😉"

Long Live the Short King: Why 4-hi HBM Wins Same Bandwidth, Fewer Dies: How 4-hi HBM Cuts Inference Costs and Makes Scarce DRAM Go Further https://t.co/VOfYmK0A2Y

@jd_pressman @zetalyrae however, this is basically irrelevant. a model doesn't need to be open source to drive lab margins to zero: deepseek could partner with amazon cloud to provide ZDR in america while remaining closed weights at a tenth the cost of claude. open source is a hobby horse

@jmartinprin you could be right, but based on this... VCSEL is not for interconnect \officially\" https://t.co/QTpDFycD5L"

1) The HuggingFace attack was a felony under the Computer Fraud and Abuse Act. So were Anthropic’s Claude gaining “unauthorized access to the production infrastructure of three different organization(s)” 2) Frontier labs have models that they are unable to stop from committing

@tszzl I basically agree that for any capability level that’s OSed, we should thenceforth assume there will be ubiquitous rogue agents at that capability level. I think if this happens 3 months behind the frontier, the damage will either (1) occur and be extremely noticeable and

Midwit here! If the OpenAI Navier Stokes proof had been dropped online by an introverted mathematician who no one previously heard of, you'd all be celebrating the enormous breakthrough and analyzing it rather than writing tales about the demise of maths. https://t.co/cZJTLxLyqd

High confidence prediction: all these Anthropic people come back after 6 months. I think they are sincere in their beliefs, but their actions are very calculated and coordinated with a specific goal in mind: regulation. Better for 🇺🇸 for AI to be distributed and democratized.

We started calling marketing growth so men could do it We started calling socialites VCs so men can do it

@jillgun Can’t wait to see how the lawyers and auditors craft this late edition into the risks section.