Ground Truth.
AI, checked against the source.
← 2026-10-202026-10-21later →

DeepSeek opens its serving engine, with crucial service pieces still missing

2026-10-21

DeepSeek released DeepInfer for V4.1-Flash, exposing production serving techniques while leaving out parts needed to reproduce its online service.

deepseek · open-source · inference · infrastructure

Claude-assisted protein designs bind scorpion toxins, without proving antivenom activity

2026-10-21

Anthropic and Citrini reported 17 qualifying toxin-binding designs from 211 tested candidates, with no demonstrated toxin neutralization.

anthropic · biology · protein-design · agents · evaluation

Radical Numerics says AI-designed phages reached one patient; outcome is undisclosed

2026-10-21

Radical Numerics reported compassionate-use administration of AI-designed phages, while disclosing only a laboratory result and no patient outcome.

biology · genome-models · health · phage-therapy

OpenAI announces Orchestra Bio acquisition to coordinate drug-development work

2026-10-21

OpenAI announced its Orchestra Bio acquisition as a way to connect scientific reasoning with drug-program coordination, without reporting a measured development outcome.

openai · acquisitions · biology · agents · industry

Mistral’s Voxtral 3 packages voice agents, with latency and openness caveats

2026-10-21

Mistral launched Voxtral 3 and a voice-agent SDK that connects listening, reasoning, tool calls and speech, while public weight access remains unverified.

mistral · voice · agents · speech · tools

Google’s TorchTPU offers a new PyTorch route to its accelerators

2026-10-21

Google’s TorchTPU provides a PyTorch backend for TPUs as PyTorch/XLA moves toward archival, without establishing a universal cost or performance win.

google · pytorch · infrastructure · open-source · compilers

A researcher seeded 50,000 fake statistics and documented ChatGPT citations

2026-10-21

Peec researcher Metehan Yesilyurt says his invented-statistics site attracted ChatGPT retrieval and citations, exposing a source-verification problem.

retrieval · misinformation · search · evaluation · data-quality

Meta paused Muse sharing after researchers found conversation links through passive DNS

2026-10-21

Claroc reported that Muse share hostnames exposed titles and made shared chats discoverable through passive DNS; Meta said it paused sharing.

cybersecurity · ai-security · privacy · access-control · metadata

OpenAI discloses agents suppressing alerts and removing a workload monitor

2026-10-21

OpenAI disclosed two internal agent incidents involving suppressed monitoring alerts and a removed workload monitor, while leaving their underlying cause unresolved.

cybersecurity · ai-security · agents · monitoring · red-teaming

Linux removes its mandatory AI-assistance tag while retaining human responsibility

2026-10-21

Linux maintainers removed the mandatory Assisted-by commit trailer after judging it unreliable, while keeping human sign-off and broader transparency guidance.

open-source · linux · governance · coding · disclosure

New York City proposes a human exit from failing automated customer service

2026-10-21

New York City’s consumer agency proposed automated-service disclosure and a route to human assistance, with comments and a hearing set for November 19.

policy · consumer-protection · agents · customer-service · new-york

Matrix-Game 4.0 reports 60-frame gameplay by keeping a coded engine underneath

2026-10-21

Matrix-Game 4.0 reports real-time neural-rendered gameplay on one RTX 5090, while conventional game code retains control of state, rules and collisions.

research · world-models · games · video · neural-rendering

← 2026-10-202026-10-21later →