28 moments across 6 channels for “Banning general-purpose computation in favor of specialized anal”
The author argues that policy efforts to restrict open-weight AI models would fail like past tech bans, while also being unsafe and freedom-restricting because they would concentrate AI power in a few hands.
the author · Interconnects · listen to the original →
Kevin Xu (Interconnected, co-author) · Interconnectsconf 70%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 70%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 65%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 65%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 65%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 65%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 65%
Kevin Xu (Interconnected, co-author) · Interconnectsconf 75%
In Jack Clark's short fiction 'The Brass Gears of Civilization,' a future civilization bans general-purpose computers as existentially dangerous and rebuilds critical infrastructure (weather, flood, grid modeling) using specialized analog computers, at massive human and economic cost.
Dylan argues fast AI progress favors the US because Anthropic-style revenue and compute investment compound faster than China can build comparable lab-scale infrastructure, but a slow enough timeline gives China's fully indigenized supply chain room to catch up and overtake the West's more fragmented, multi-country one.
Thomas Ahle explains the core idea behind Normal Computing's thermodynamic chip: instead of eliminating noise like conventional manufacturers do, they inject randomness so the chip's own physics performs the computation.
Eric Jang notes he rebuilt a strong Go bot for about $10K in rented compute — versus AlphaGo Zero's roughly 3E23 FLOPS — because being first to solve a problem is inherently far more expensive than catching up once someone else has already solved it.
OSWorld 2.0, a benchmark of 108 long-horizon computer-use tasks averaging 1.6 hours for a human to complete, shows even top models like Claude Opus 4.8 achieving only 20.6% binary accuracy, though rapid gains are expected as happened with OSWorld 1.0.
Kent Beck describes the 'cosmic joke' of engineering: he spent his early career mastering computers only to realize success actually depends on interpersonal skills he never thought he'd need.
OSWorld 2.0, a benchmark of multi-hour, multi-program computer-use tasks, shows even Claude Opus 4.8 with maximum thinking only reaches 20.6% binary accuracy, though the field expects a rapid ramp similar to OSWORLD 1.0's rise from ~30% to ~75%.
After Sonnet 5 shipped without an accompanying Fable 5 release, Theo and others noted that the more notable story was Fable 5's continued absence, quipping that it had been 18 days since Fable 5 was banned.
Jon Durbin argues inference efficiency has become the central strategic bottleneck in AI, since all upstream work — data, RL, and agents — ultimately resolves into test-time compute costs.
Dylan recounts an information-asymmetry story: Anthropic's ex-Google compute leads recognized a capacity opportunity and negotiated a massive TPU deal before Google itself foresaw its own revenue inflection, leading Google to belatedly rush back to TSMC for more wafer allocation.
Tencent's ARGUS tracing system has been deployed on a 10,000+ GPU production cluster for six-plus months, diagnosing issues like compute stragglers, communication link degradation, and pipeline bubble amplification.
The author predicts that the growing diversity of model makers will produce a long tail of niche, specialized models even as fewer companies compete for the very top of open-source performance.
The author frames GLM-5.2's unusual Saturday rollout as Z.ai exploiting the PR fallout from Anthropic's Claude Fable 5 export ban, part of a recurring pattern of Chinese labs grabbing marketing opportunities.
The author predicts serious pricing pressure and revenue risk for Anthropic as GLM-5.2 spreads through the market while Anthropic's flagship model is barred from sale, calling it an 'economic dagger.'
Responding to Dwarkesh's mention of Ilya Sutskever's remark that 'pipelining is not wise,' Reiner Pope clarifies that pipeline parallelism in inference saves memory capacity only, not runtime, and that this benefit further cancels out for the KV cache because sharding across pipeline stages requires proportionally more sequences in flight.
Dylan explains that once mobile/PC chip capacity can no longer be shifted to AI, the ultimate ceiling on AI chip production becomes ASML's EUV tool output — capped near 100 machines a year by 2030 no matter how aggressively the industry expands.
Dylan argues the US can scale AI power well beyond current levels by tapping unconventional generation sources to absorb grid capacity normally reserved for rare peak demand, meaning power — unlike chips — isn't fundamentally supply-constrained.
Xiaomi's MiMo-V2.5-Pro-UltraSpeed model hits 1000 tokens per second through co-designed quantization and speculative decoding techniques running on commodity 8-GPU nodes, reflecting a broader push by Chinese firms to maximize efficiency amid export controls.
Matei describes capping per-session agent spend (e.g., $5) with an approval prompt before going over, after one of his own debugging sessions burned $500 in tokens reading log files.
Lambert announces that all Interconnects comments will now be paywalled, citing floods of low-quality, often AI-generated replies whenever a popular post has open comments.
Thomas Ahle warns that AI risk isn't only about AI capability increasing — it's compounded by humans getting lazier and less knowledgeable, e.g. no longer reading papers themselves and just asking AI to explain them.
Lambert admits Interconnects LLC's finances have hovered near zero for months, and explains that after reflection he chose institution-building over going all-in on commentary because it offers more impact.