On 3 September 2026, Google Research and HHMI Janelia collaborators announced in Cell a complete, human-proofread connectome of the male Drosophila melanogaster central nervous system, comprising over 166,000 neurons and about 125 million synaptic connections and including the ventral nerve cord. The map is released via Neuroglancer and complements existing female fruit-fly connectome resources.
Constraints shifted from producing synaptic-resolution reconstructions of individual fruit-fly brain halves (~25k neurons in 2020) to proofread complete adult central nervous systems (~166k neurons including optic lobes and ventral nerve cord). The source identifies proofreading/annotation, not image acquisition, as the remaining bottleneck and reports AI training with synthetic neurons to reduce that human correction load, which is the pre-condition for scaling toward vertebrate whole-brain projects.
Single-source assessment built on a Google Research blog that cites a Cell paper but gives no DOI, data portal details, error rates, or comparator counts. The magnitude claims (166k neurons, 125M synapses, largest-by-neuron-count) are stated in the source but not independently verifiable from this document. Treat as an official announcement with moderate-to-high prior plausibility, pending direct inspection of the published paper and released data.
NeoMME is a from-scratch family of 260M and 800M bidirectional encoders for text and images; its fine-tuned retriever reaches ViDoRe v3 nDCG@10 of 0.523 at 260M parameters and compresses late-interaction indexes from about 1.5 MB to 6 kB per page at over 95% retained quality.
Visual document retrieval previously paid a VLM tax: a pretrained vision tower, a connector, and a causal decoder. NeoMME removes both pretrained stages and trains one bidirectional Transformer on raw image patches and text; the retriever emits dense and late-interaction vectors in one pass and compresses large high-resolution index payloads by about 255x. This shifts the practical constraint from model architecture and index storage to training a competent multimodal encoder from scratch.
All headline performance, throughput, and compression numbers come from one Hugging Face blog/technical release; NeoMME scores are marked as the authors' own evaluation, while comparison rows from MTEB are daggered. Independent verification, external benchmark runs, and ablations are absent, and the Pareto-frontier wording conflicts with the included Vultron Flash table row when read strictly as model size versus nDCG@10.
OpenAI's GPT-6 Astra launch combines strong reported advances in agentic computer use, coding, and ARC-AGI-3 (99.9% via a custom harness) with independent indexes showing more uneven gains and a system card that reports decreased chain-of-thought monitorability alongside improved alignment. The constraint shift is from inspectable reasoning and default-harness evaluation to opaque reasoning state and harness-mediated evaluation.
Two evaluation and safety constraints have shifted. First, chain-of-thought monitoring is weakened: UK AISI measured Astra's no-CoT time horizon at 30.9 minutes versus 3.6 minutes for GPT-5.6 Sol, so dangerous reasoning may not surface in inspectable text. Second, state-of-the-art scores on ARC-AGI-3 require a custom harness that preserves opaque reasoning state between requests, meaning benchmark results now depend as much on harness engineering as on model capability.
Evidence is mixed and unevenly verified. Multiple early-access and vendor-adjacent sources report large gains, but the strongest benchmark figures are harness-sensitive: ARC-AGI-3 drops from 99.9% under the Provider Adapter harness to 62.7% under the default harness, and Artificial Analysis places Astra equal to GPT-5.6 Sol on its Intelligence Index. System card claims about alignment and monitorability come from OpenAI and UK AISI evaluations, not from peer-reviewed third-party replication.
Doctorow argues that removing Trump will not restore American stability because he is an effect of a far-right coalition, not its cause. The essay's distinctive claim is that US platform infrastructure has become an immediate coercive weapon and that the only resilient response is a post-American internet built from sovereign public digital goods.
The essay's central constraint shift is from surveillance to direct coercive shutdown: Big Tech is described as fully fused with Trump and willing to revoke services at presidential request, while the pre-Trump elite's 'iterated game' restraint is said to have collapsed. Because Trump's successor factions cannot afford to lose power, waiting for Trump's death or for a return to normal is insufficient; the binding constraint becomes how quickly non-American jurisdictions can create sovereign alternatives.
The document is a single-voice opinion essay. It asserts specific episodes, such as the ICC prosecutor's account loss and Microsoft action against a Brazilian judge, without intra-text citations or alternative sources. The core capability claims about 'america-only' tech shutdowns are analogical and not operationalized. The source is useful as an argumentative scenario but requires external verification before its factual assertions are treated as established.