Track frontier trends and curated high-quality content to stay ahead in the fast-moving AI SaaS world.
Super Intelligence leaves 37% human benchmark in the dust just 18 months later.
AI models are now acing accounting tests that previously stumped even the best systems, raising transparent questions about rapid AI advancement.
DeepSeek Harness brings open-source agentic tools to desktop
DeepSeek has released Harness as a downloadable desktop app for Mac and Windows, simplifying access to its open-source AI agent with zero terminal setup.
Coordinate-based tests reveal how well LLMs have internalized geography.
New evaluations plot latitude and longitude queries against AI models, revealing that LLMs have memorized geographic boundaries directly from compressed internet data.
Celebrating academic achievement in international engineering
Thai celebrity Pooh Krittin successfully graduated from the Faculty of Engineering, majoring in Robotics and AI, bringing pride to his family.
AI model verifies imagery from social media user.
DeepSeek has officially confirmed the authenticity of photos shared by user MoonCake1738, adding AI-backed credibility to the visual evidence.
A new platform claims to replace multiple apps with a single AI workspace.
Venturis 13 introduces LifeOS, integrating ATHENA for personal use and TARA for business into one unified AI-driven platform.
Comparing consumer AI companion apps with restricted research models
While Tavus Griffin remains a restricted research preview, Prelulu offers immediate access to live AI video calls with fictional companions for everyday users.
Shift to 'Super Intelligence' drives 2,100% spike in .si domain purchases.
Donald Trump's directive to replace 'AI' with 'SI' has triggered an unprecedented surge in Slovenian .si domain registrations, leaping from under 2,000 to 44,000 in September.
Trump and tech leaders sign voluntary safety pledge despite typo.
The White House released an AI safety accord signed by Donald Trump and six tech executives, but the document contained a glaring typo referring to the 'Unites States.'
Seven Infrastructure Updates Released on September 30; GEMM Compute Utilization Exceeds 99.8%
On September 30, DeepSeek did not release a new model but concentrated updates on seven key projects around the Ascend 950. Source code reveals that its matrix multiplication performance has reached 99.8% of theoretical peak compute, indicating that inference capabilities are taking shape, while training remains under active development.
DeepSeek V4.1 Flash scores 57/100 in a rigorous black-box bug bounty benchmark, trailing leaders Opus 5 and Grok 4.7 but offering significant cost advantages.
Controversial protest marks escalation in Brazil's political landscape.
Renan Santos, leader of the Brazilian Liberty and Justice Movement (MBL), staged a disruptive scene at Globo's headquarters, complicating his presidential campaign as he seeks to gain political traction.
Late September to Early October: DeepSeek, Anthropic, Cognition and Others Announce密集 Updates
On September 30, DeepSeek open-sourced the infrastructure components for the Ascend 950, attracting attention after being relayed by Hong Kong media on October 1. Around the same time, Anthropic stated that GLM-5.3 rivals Mythos in capability, Claude for Government officially launched, and Sign in with ChatGPT integrated with Devin.
Major LLMs including DeepSeek, Grok, and Claude deny possessing SI status.
When asked if they consider themselves superintelligences, major AI models including DeepSeek uniformly deny the claim, maintaining their standard operational identities.
X user Kriska Carvalho accuses Globo of cowardice and coup-like tactics after the network blamed external forces for Flávio Bolsonaro's debate withdrawal instead of admitting fault.
New research shows self-play reinforcement learning achieves top-level Stratego mastery at a fraction of traditional costs.
Researchers achieve superhuman performance in Stratego using self-play and test-time search, proving imperfect information AI can outpace humans without industrial budgets.
National Security Faces New AI Threats; Lawmakers Seek Regulatory Responses
The US Senate held a public hearing on September 30, 2026, focusing on how to protect the country from the growing threat of 'AI agent attacks.' As AI technology becomes increasingly embedded in government and business, cybersecurity defenses face unprecedented challenges.
TokenHarborAI introduces a new fast-tier architecture for generative AI.
DeepSeek has released its V4.1 Flash Beta on TokenHarborAI, offering a new native multimodal architecture and fast tier access for developers.
Tech boom in China sparks global reaction, not biological warfare.
DeepSeek's rapid advancements in AI are reshaping global tech markets. Contrary to conspiracy theories, experts emphasize economic and technological impacts over baseless social claims.
X user questions double standards in criticism of political figures.
An X post questions whether political figures like Flavio Bolsonaro face the same scrutiny as President Lula, highlighting perceived inconsistencies in Brazilian digital discourse.