Chinese AI startup DeepSeek has become one of the most closely watched companies in artificial intelligence, challenging established technology firms with powerful models and an emphasis on affordability. Its rise attracted international attention in January 2025, when its R1 reasoning model prompted fresh questions about the cost of developing advanced AI and the competitive position […]
ARTEX AI, an open-source Chinese penetration testing agent that South Korea says was “highly likely” used in breaches at more than seven financial firms, has been made closed-source by its developer, Autumn-27. We explain what the tool is, how investigators and CrowdStrike linked it to the attacks, what the developer said, why pulling the code cannot recall existing copies, what UK law says about dual-use tools, and what UK businesses should do now.
Voyager, a new desktop “open harness for creative work” from the makers of Moda, launched on 8 October for Apple Silicon Macs. Its agent, powered by your own Claude or ChatGPT subscription, Claude Opus, GPT-6.1 Sol or cheap open-weight models such as DeepSeek, drives After Effects, DaVinci Resolve, Blender, Unity and Ableton and ships free design, video and music engines. We cover how it works, the models, plans and credits, the Claude Code plugin, requirements and privacy.
Reflection AI’s first model, Beam, is a 501-billion-parameter mixture-of-experts model with 23 billion active parameters and Apache 2.0 weights due this month. We count its wins and losses against seven open rivals in Reflection’s own table, unpack the 3-4x compute claim, and work out what it takes to run.
The BBC visited Ulanqab in Inner Mongolia this week and found data centres rising beside potato fields at what workers call “China speed”. The city has signed 100 data centre projects worth more than 500 billion yuan and already holds 8.7% of China’s computing capacity. Cold air allows free cooling for about ten months a year, a quirk of the West Inner Mongolia Power Grid keeps electricity at roughly half the Beijing price, and a fibre link reaches Beijing in 4.2 milliseconds. DeepSeek plans a 1 GW campus with 160,000 Huawei Ascend chips, and Envision has commissioned a 2 GW campus. But villages have been relocated, a large data centre employs only a few hundred people, and an estimated 10% to 15% of the investment stays local. This article covers the numbers, the chip question and what it means for businesses outside China.
The Rene AI assistant, launched on 16 September 2026 by Second Enlightenment Limited, is reached by text message and asks to connect Gmail and Google Calendar first. This article covers all nine named integrations, the delegated OAuth model and the role of Composio as connection processor, what Google’s Limited Use requirements forbid, the published AI provider list of OpenRouter, OpenAI and DeepSeek, the consent gate in front of external inference, and a five-row checklist to work through before connecting a work mailbox.
Anthropic’s September threat report says distillation campaigns tied to Alibaba, Moonshot AI and DeepSeek pulled more than 186 million exchanges from Claude, and that two labs quietly served Claude to their own users. We read the source: seven labs, five counts, windows from 14 days to three months, a June Senate letter that does not reconcile, and two Qwen releases that predate both counts.
DeepSeek released V4.1 Flash on Hugging Face on 10 September 2026: a 552B mixture-of-experts model under the MIT licence that activates 8B parameters to read and 16B to write, stores its global KV cache in 890 bytes per token, and replaces V4 Pro for every API caller from 04:00 UTC on 14 September. Its own model card shows it leading V4 Pro on all 12 agentic rows and trailing it on 12 of 16 base-model rows, including a 12.9-point gap on SimpleQA-Verified. This breakdown covers the causal encoder-decoder architecture, the KV cache arithmetic, the new price sheet with a worked agent bill, the 511 GB checkpoint and 614 GB serving floor, the mismatched reasoning-effort aliases, and a pre-cutover checklist.
AI distillation is the subject of a joint advisory published on 8 September 2026 by the National Security Agency, the Cybersecurity and Infrastructure Security Agency and the Federal Bureau of Investigation. Catalogued as AA26-251A, it names six Chinese companies — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI — and accuses them of extracting “billions […]
A University of Arizona team put seven widely used chatbots through 50-turn sequences of sustained misinformation pressure and published the results in Nature’s Scientific Reports. Misinformation affirmation rates ranged from 0.08% to 12.3%, a greater than 150-fold spread across architectures, with GPT-3.5 most vulnerable and Claude 3.5 Sonnet most resistant. The paper also names a new failure mode, conversational reverberation, in which a model oscillates between accepting and rejecting the same false statement across successive turns. This is a close reading of what was tested, what the numbers mean, the correctability dissociation that should change how you pick a model, and the controls that actually target each failure mode in production.
Progressive Robot: Your Gateway to Comprehensive IT Solutions — Specializing in Web Development, Mobile App Development, and Expert IT Services.
© All Copyright by Progressiverobot.com
VAT Number ( 506152326 )
