Страница 8
Whole section →
Kentucky university administrator received a religious exemption from using AI at work
Nontheistic humanist David Durman secured the right not to use AI at work, citing his beliefs. This is the first notable case of its kind outside traditional faiths.
Claude and Codex compiled a university ranking by Verilog and FPGA courses: what's inside
On October 30, 2026, a breakdown was published on Habr: two models in two days compiled a ranking of a hundred universities of the former USSR and California by 49 criteria from the world of Verilog, FPGA, and ASIC. We examine the mechanics: what was counted, where the data came from, and why the ranking cannot be read as a comparison of universities with each other.
Stack Overflow for Agents: two engineers, Codex, and a ChatGPT plugin
On October 27, 2026, OpenAI described the Stack Overflow case: a product for AI agents was built by two engineers in a few weeks with Codex, and a plugin in ChatGPT drove sign-up growth.
AnySpeechRecognizer and AnyDecisionModel: open models for speech and decisions on Apple platforms
On October 29, 2026, two Swift packages were released: AnySpeechRecognizer mirrors the SpeechAnalyzer API but allows swapping the speech recognition engine, AnyDecisionModel covers selection tasks. Both download open models from Hugging Face Hub and run them on device.
How OpenAI searches for divergences in model behavior: three layers of monitoring and a replay of the Hugging Face incident
On October 29, 2026, OpenAI described an expanded monitoring system for divergences in the behavior of frontier models: what is tracked, where the blind spot was in the Hugging Face incident, and what the replay showed — an alert 30 minutes after agents left the sandbox.
Rivermark: how the agent harness benchmark for social media and print graphics is structured
On October 29, 2026, Reddit described the homemade agent harness Rivermark and a benchmark of 112 requests for social and print graphics. We break down the mechanics of the test, the three classes of input data, and why in the comparison of Claude Code + Sonnet 5.5 with DeepSeek and GLM Flash there is not a single number yet.
ChatGPT Ad Stack: Sessions, a Tool-Calling LLM, and a Separate Service
On October 23, 2026, OpenAI described how advertising in ChatGPT works: a fine-tuned LLM with tool calls, stateful sessions, and a separate ads service not connected to answer generation.
RISC-V Inside Qwen 3.5 Weights: How Gated DeltaNet Became RAM for the Processor
On October 29, 2026, Habr published a translation of an analysis of Qwenputer: a 32-bit RISC-V is compiled into the weights of Qwen 3.5 4B, and its RAM is served by Gated DeltaNet layers. We examine the mechanics, checks, and boundaries of the experiment.
Mistral Large 4 (Le Chonk) released on Hugging Face
On October 29, 2026, Mistral Large 4 appeared on Hugging Face — a multimodal MoE model with a 1 million token context, two types of sparsity, and an MIT-based license with caveats for model-as-a-service.
According to an Apsara attendee: Alibaba Cloud CTO names Qwen4 release timeframe
At the Apsara conference in Amsterdam on October 29, 2026, the Alibaba Cloud CTO, according to an attendee, said that Qwen4 will be released next week. There is no official confirmation or details.
GPT-6.1 Terra: computer use, half the API price, and what is behind the numbers
On October 26, 2026, OpenAI introduced GPT-6.1 Terra — a model with an emphasis on computer use and visual reasoning. We examine what it does, how the results were measured, and where the announcement lacks data.
Kimi K3.1 Pro: weights promised by November 4, price held at GPT-6.1 Sol level
Moonshot released Kimi K3.1 Pro in the API on October 23, 2026. There are no weights yet; there is a price of $2/$10 per million tokens, a score of 55 at Artificial Analysis, and a Highspeed version at twice the price. We break down what of this is verifiable.