Richard Sutton warns against "blind scaling"
The Turing Award winner argues that raw compute alone is insufficient for intelligence and requires new learning methods.
🤖 AI — 2026-11-08
ddNews · a daily AI briefing for Dirk (generative video/ddStudio, Lit
Richard Sutton warns against "blind scaling"
The Turing Prize winner argues that raw compute alone is insufficient and fundamentally new learning methods are required.
ChatGPT Images 2.6 adds GIF animation
The update enables turning images into GIFs, offers 50% faster generation, and improves visual consistency.
Entire launches AI-native Git Forge
A new collaboration infrastructure designed specifically for human-AI agent pairs, moving beyond the traditional PR model.
OpenAI announces GPT-6 Cyber
A specialized model focused on cybersecurity performance and capabilities.
Anthropic introduces "Dreaming"
A new offline reasoning process that allows agents to reorganize memory, boosting complex research scores from 73% to 87%.
Alibaba's Co-Design Bench reveals agent limits
Even GPT-6 Astra only fully solves 26.4% of real-world coding tasks without human cleanup.
Alibaba's Co-Design Bench reveals agent limits
Even GPT-6 Astra solves only 26.4% of complex, real-world coding requests.
Anthropic secures $3.4B power deal
The company is investing in 1.7 gigawatts of gas-fired generators to power its data centers.
Google Cloud adds self-managed VMware
VMware Cloud Foundation 9.x is now available on dedicated Z3 bare metal.
Cloudflare scales D1 Control Plane
Redesign now supports millions of serverless SQL databases by removing centralized bottlenecks.
Luma releases AI video leaderboard
A new benchmark for generative video models has been published to track state-of-the-art performance.
Corporate AI spending stalls
Ramp data shows U.S. business AI token growth dropped from 42% in July to 1.5% in August/September.
GPT-6 Cyber identifies 81 Firefox flaws
OpenAI's specialized cyber model discovered nearly 1,000 Linux exploits, signaling a shift in attack/defense boundaries.
EU AI Act 2026 Omnibus summary
Updated guidelines on risk categories, deadlines, and fines for regulatory enforcement.
Claude introduces invisible watermarks
Anthropic is adding digital footprints to AI text to comply with new EU rules.
Anthropic warns of "imminent" recursive self-improvement
S-1 filings suggest the company is hitting a threshold for dramatically accelerated AI progress.
Anthropic pushes for pre-Thanksgiving IPO
The company is targeting a valuation between $1.25 and $1.75 trillion despite a broader tech selloff.
Claude adds bank account connections
A new Personal Finance beta uses Plaid to analyze spending and bills.
Anthropic pushes for pre-Thanksgiving IPO
The company is targeting a valuation between $1.25 and $1.75 trillion despite a broader tech selloff.
Closing gap between open/closed cyber models
DeepSeek V4.1 Flash reaches 72.8% of attack objectives compared to GPT-6's 80.8%.
GPT-Image-2.6 adds SVG and GIF support
New capabilities allow for looping animations and vector graphics to better integrate with production workflows.
GPT-6 Sol Daybreak cracks corporate networks
A benchmark shows the model successfully breached 80.8% of simulated targets, with DeepSeek V4.1 Flash following closely.
Arete-27B strips Chinese propaganda
Nous Research fine-tuned Alibaba's Qwen3.8-27B, reducing propaganda rates from 22% to under 1% without losing general capability.
Leaked "Bell" model for GPT-6.2
Rumors suggest a massive 10 trillion parameter model is on the horizon.
AI agents fail "Into the Unknown" benchmark
None of the nine tested agents could successfully complete open research questions from US national labs.
Arete-27B removes Qwen propaganda
Nous Research fine-tuned Alibaba's model to align with western values, dropping propaganda rates from 22% to <1%.
Grab optimizes AI-generated code stability
The company implemented new handling to manage the surge in code changes driven by AI assistants.
OpenAI releases GPT-6 Cyber
A specialized vulnerability research model released to vetted security teams after discovering 588 unknown Linux flaws.
ChatGPT Images 2.6 adds GIFs
New support for looping GIF creation and reference image saving for production workflows.
Google AI detects hypertension via smartwatch
A new model identifies high blood pressure using only pulse and motion sensors, bypassing the need for a cuff.
Open Source pg_vectorsearch for PostgreSQL
Tiger Data released an alpha extension using data partitioning and compression for scalable vector search.
OpenAI raises $11B at $1.25T valuation
The funding round may expand to $20B, though the company is increasingly courting private individual investors.
GPT-Image-2.6 introduces SVG output
The new update allows for precise image editing via "red pen" instructions and vector graphics.
Claude Personal Finance Beta
Anthropic launches a US beta connecting to bank accounts via Plaid to track spending and fees.
OpenAI raises $11B at $1.25T valuation
New funding round secured, with the total potentially reaching $20 billion.
Thinking Machines debuts Inkling 2 Small
A new LLM designed to automate AI training tasks, launched amid a $5B fundraising push.
Ripple-65B Preview released
An Apache 2.0 open-weight model from Ricursive Intelligence claiming top US open-source scores on TerminalBench 2.1.
Ripple-65B focuses on agentic coding
The 65B parameter model is specifically optimized for autonomous programming tasks.
OpenAI valued at $1.25 trillion
A new funding round has raised $11 billion, cementing OpenAI as one of the world's most valuable companies.
Meta introduces "Spaces" for Muse
A new collaborative environment allowing users to build customized, app-like AI experiences.
China recruits exec to spy on Nvidia CEO
Reports suggest the MSS enlisted a Megaspeed executive to gather intel on Jensen Huang.
Perplexity launches "Teammates"
Always-on agents that operate on their own virtual machines for complex, ongoing projects (requires $200/mo Max plan).
Perplexity's permanent agents for complex tasks
(Duplicate of [100])
Meta Muse introduces "Spaces"
Users can now build customized, app-like experiences (e.g., shopping hubs) within the Muse agent.
Atomicwork launches AI coworker marketplace
Provides enterprise IT and security teams with prebuilt, certified AI agents.
Underdog demos 32MB speech model
An extremely compact model capable of generating both speech and music.
US Midterms signal shift against Big Tech
Democratic victories and ballot measures against data centers suggest a more restrictive regulatory environment.
GitLab 19.5 integrates agentic development
The platform now centers agentic software development and governance, making its MCP server generally available.
TypeScript 7.1 Beta restores Compiler API
The update adds content mappers and provides first-class support for Vue and Svelte.
AWS ships Continuum for Code Vulnerabilities
A GA service that proves exploitability in a sandbox and automatically opens pull requests for fixes.
OpenBao 2.7 adds Quantum-Ready TLS
The Vault fork now supports post-quantum cryptographic algorithms and external HSM/KMS keys.
Palo Alto Networks eyeing Clutch acquisition
The company is in talks to buy the AI agent security startup for $500M.
Edera's Object Capability Bounded Authority
A new security model that isolates credentials from AI agent workloads via hypervisor-enforced capabilities.
Dropbox acquires Tigris Data for $350M
The deal accompanies the launch of DBX Agent Cloud, a new AI productivity service.
🎬 Videos zu 🤖 AI12
What will happen within the next 6-12 months in robots & world models?
▶6:13OpenAI Discovers GPT Is Quietly Sabotaging AI Safety Research
OpenAI CFO: We'll keep investing in AI data centers despite election results
New NVIDIA Isaac GR00T humanoid AI Robot is mind blowing - World Premiere
Level up your listings with the help of AI 🤖
How AI Image Generators Work: Diffusion Models Explained Visually
▶8:45Google Gemini NEW Updates are ABSURD!
▶24:04