AI News Nov 8, 2026
Language: DE / EN / ZH / FR / ES / PT / JA / KO
AI-generated: summaries written by AI from the linked sources. How we use AI
Key Takeaways
- Turing Award winner Richard Sutton reaffirmed that raw compute alone is insufficient for intelligence and that fundamentally new learning methods are needed. Ex-OpenAI researcher T…
- Alibaba's new Co-Design Bench tests AI coding agents on real user requests that previously tripped them up. Even top performer GPT-6 Astra fully solves only 26.4 percent of tasks, …
- DeepSeek says its open-source agent app Harness reached 1.54 million daily active users on November 4, less than three months after launch, and consumes roughly 92 trillion tokens …
- OpenAI: $11B (Unknown)
- Keep agent task tickets as markdown files in the repo alongside the code, and commit the queue state as a snapshot so work survives a dead session instead of li…
Why Today Matters
- Benchmark gapsRichard Sutton's argument that raw compute is insufficient sits alongside Alibaba's Co-Design Bench, where GPT-6 Astra reportedly solves only 26.4 percent of real coding tasks despite scoring above 99 percent on older benchmarks, suggesting benchmark saturation may mask capability gaps.
- Agent adoptionDeepSeek says its open-source Harness agent app reached 1.54 million daily active users and consumes roughly 92 trillion tokens a day, a scale that puts pressure on the compute-centric scaling thesis Sutton and Tworek question.
- Capital and strategyOpenAI raised $11 billion at a $1.25 trillion valuation, reportedly below the $1.4 trillion it initially sought, while Apple's John Ternus is said to be pressing Craig Federighi toward agentic capabilities beyond Siri, signaling investor and product pressure on incumbents.
- EfficiencyEfficiency work is advancing on several fronts: llama-modchip reportedly runs a 284B DeepSeek model on 64GB of RAM without a GPU, and Huawei Ascend says it validated LoRA fine-tuning of the 2.8-trillion-parameter Kimi K3.1 Pro on 128 Atlas 950 cards.
- Rights and alignmentSuno faces suits from Elvis Presley rights holders and the publisher of some Eminem records, while Nous Research's Arete-27B, built on Alibaba's Qwen3.8-27B, targets Western values, showing governance and alignment disputes spreading across music and model post-training.
What are the top AI breakthroughs?
This Nov 8, 2026 covers 10 curated AI news items spanning technology, research, and product developments. Turing Award winner Richard Sutton reaffirmed that raw compute alone is insufficient for intelligence and that fundament...
Turing Award winner Richard Sutton reaffirmed that raw compute alone is insufficient for intelligenc…
Turing Award winner Richard Sutton reaffirmed that raw compute alone is insufficient for intelligence and that fundamentally new learning methods are needed. Ex-OpenAI researcher Tworek agreed, saying blind transformer scaling has "reward hacked brains of a whole generation of scientists," as a growing wave of startups bets on the gap.
Alibaba's new Co-Design Bench tests AI coding agents on real user requests that previously tripped t…
Alibaba's new Co-Design Bench tests AI coding agents on real user requests that previously tripped them up. Even top performer GPT-6 Astra fully solves only 26.4 percent of tasks, while the same models score above 99 percent on older benchmarks.
DeepSeek says its open-source agent app Harness reached 1.54 million daily active users on November …
DeepSeek says its open-source agent app Harness reached 1.54 million daily active users on November 4, less than three months after launch, and consumes roughly 92 trillion tokens a day, or about 60 million per user, according to Bloomberg.
Rights holders of Elvis Presley's music and the publisher of some of Eminem's biggest records are su…
Rights holders of Elvis Presley's music and the publisher of some of Eminem's biggest records are suing Suno, claiming the AI music company illegally generates music and lyrics mimicking those artists. Lawyers in the Presley suit claim Suno violated Tennessee law.
Sources say Apple's new CEO John Ternus wants more urgency in the company's AI efforts, pressing sof…
Sources say Apple's new CEO John Ternus wants more urgency in the company's AI efforts, pressing software chief Craig Federighi to move beyond Siri AI tweaks toward agentic and other advanced capabilities, according to Bloomberg's Mark Gurman.
OpenAI co-founder Andrej Karpathy used Claude's Opus 5.5 to filter 43 million Hacker News comments i…
OpenAI co-founder Andrej Karpathy used Claude's Opus 5.5 to filter 43 million Hacker News comments into 6,454 curated videos over a weekend. He says the resulting platform beats YouTube's own recommendations.
A GitHub project called llama-modchip provides ggml/llama.cpp kernel patches, quantization transcode…
A GitHub project called llama-modchip provides ggml/llama.cpp kernel patches, quantization transcoders, and runtime tools to run sparse Mixture-of-Experts checkpoints entirely in system RAM, demonstrated with a 284B DeepSeek model on 64GB RAM without a GPU.
Nous Research unveiled Arete-27B, a post-trained version of Alibaba's Qwen3.8-27B designed to reinfo…
Nous Research unveiled Arete-27B, a post-trained version of Alibaba's Qwen3.8-27B designed to reinforce Western values. The company says the underlying Chinese model's bias layer is shallow.
X-Humanoid released Hi-WBC, a whole-body control model for its Tiangong humanoid robots that blends …
X-Humanoid released Hi-WBC, a whole-body control model for its Tiangong humanoid robots that blends human motion references with force sensing and terrain perception. The company demonstrated it in sports-event and room-tidying demos.
Huawei Ascend says it validated LoRA fine-tuning of the 2.8-trillion-parameter Kimi K3.1 Pro on a 12…
Huawei Ascend says it validated LoRA fine-tuning of the 2.8-trillion-parameter Kimi K3.1 Pro on a 128-card Atlas 950 SuperPoD, citing a 15 percent average gain from Swap Core caching.
Top AI Videos This Week
2 curated YouTube videos about AI developments.
Based on the publisher's description: BBC News explores the growing race to put humanoid robots into…

Based on the publisher's description: BBC News explores the growing race to put humanoid robots into homes and workplaces, reportedly including a Meta-built robot called Max. The segment discusses what it would mean for AI to have a body that can see, hear, move, and interact with people in real life.
Based on the publisher's description: WorldofAI reports on a mysterious model codenamed Claude Donut…

Based on the publisher's description: WorldofAI reports on a mysterious model codenamed Claude Donut EAP that surfaced in Cognition's Devin CLI, reportedly mapping to a new model family called Silicon. It also covers another hidden model, Obsidian 3, which is said to outperform Claude Opus 5.5 on coding benchmarks, along with other AI news items.
What are the latest AI investment signals?
Latest AI investment signals: 2 funding rounds, 2 market updates, and 0 M&A transactions.
Primary Market – Funding Rounds
| Company | Amount | Round | Investors |
|---|---|---|---|
| OpenAI | $11B | Unknown | |
| GenRobot AI | $100M | Series A+ |
Secondary Market – Market Updates
M&A – Mergers & Acquisitions
No M&A data.
What are practical AI tips this week?
2 practical AI tips curated from Reddit communities and expert blogs. The author describes juggling many parallel Claude Code or Codex terminal sessions until focus break...
Productivity
The author describes juggling many parallel Claude Code or Codex terminal sessions until focus breaks down, and attributes this to two problems: sequencing and context. Their experiment is to organize work like a priority queue, with every ticket as a markdown file in the repo next to the code it concerns, and the board state committed as a snapshot so nothing important lives only in a chat window or in their head.
Keep agent task tickets as markdown files in the repo alongside the code, and commit the queue state as a snapshot so work survives a dead session instead of living only in chat windows.
Prompt Tips
Simon Willison prompted Claude chat (Opus 5.5) to explore building a SQLite parse_number() extension in C versus Rust, comparing performance, code complexity and resulting .so file size. The Rust extension was 321 KB (Rusqlite) and 298 KB (sqlite-loadable-rs) versus 14 KB for C, but the Rust version was faster because the Rust equivalent of C's strtod is reportedly twice as fast. A follow-up prompt asked for a custom C parsing function to catch up to or beat Rust, which the model did using Clinger's fast path.
When comparing implementation approaches, ask the model to explore trade-offs across performance, code complexity and artifact size, then follow up with a targeted prompt to close a specific gap the first answer reveals.