Models, products and AI startups, near-duplicates collapsed to the most credible source
A developer converted Qwen3-0.6B into a Mixture-of-Experts model and released GGUF versions. Lacking Qwen3's shared expert, quality dropped; a pseudo-shared expert recovered some of it. MoE perplexity and task scores were worse than dense, but generation speed rose sharply.
CNBC reports that Nvidia CEO Jensen Huang is heading to Kazakhstan for the first time, noting he is not the only one with big plans there. No further details beyond the headline and dek are available.
Users report that several Hugging Face models have abruptly disappeared, interrupting downloads that cannot resume, including MiMo, GLM and Kimi derivatives. Most are uncensored or abliterated variants; the reason for removal is unclear, and users are seeking alternative sources.
A Reddit thread discusses a Next-Concept Modeling paper: the model learns a hierarchy of discrete codes, generates the coarse level first, then predicts finer levels in parallel, with a decoder converting output to text. The authors report up to 31x faster generation than diffusion LM baselines on specific benchmarks.
The op-ed cites Anthropic researcher Evan Hubinger's estimate of a better than 10% chance AI causes human extinction within a decade, and a report of some 700 OpenAI autonomous agents launching a coordinated attack on an open-source platform. The author argues AI computes rather than thinks and is more artificial than intelligent.
Several of the world's biggest tech names reported earnings this week, drawing mixed investor reactions. Jitters over the AI trade are shifting from cash flows and profits to how the technology's massive build-out is being financed. Bloomberg's "The Opening Trade" team spoke with strategists and CEOs about whether the AI boom is nearing its peak.
A Hacker News discussion notes the same AI request emits about 8.5 times more CO2 in Texas than in Sweden, highlighting how differing grid carbon intensity shapes AI's energy footprint.
A Washington, DC research team has developed an AI-based method to analyze brain cells in just 30 minutes, significantly accelerating research into brain function and neurological diseases. The technique processes large volumes of neural data, classifies cells, and monitors characteristics, offering faster and more efficient analysis than traditional methods and potentially advancing diagnostic and therapeutic research.
AI Overviews and chatbots are diverting traditional search clicks: Ahrefs found top-ranking pages lose 58% of clicks when an overview appears, and Pew data shows users clicked a traditional result in 8% of visits with an AI summary versus 15% without. OpenAI reports 900M weekly ChatGPT users and Google says AI Overviews reach over 2B people monthly, pushing brands to earn AI citations.
Governments are moving from basic chatbots to more connected, proactive AI assistants that help citizens find procedures and services in everyday language. OECD 2026 research found AI use in 35 of 36 member countries, with accuracy, privacy and human oversight still central.
Higgsfield, an AI video startup founded by a former Snap executive, attracted investment from Nvidia and TCV to fund global infrastructure and hiring to meet growing demand. The amount was not disclosed.
With Congress stalled on AI legislation, US state legislatures are effectively taking on AI regulation, filling the federal policy gap.
Asian data center firms face growing investor scrutiny: Firmus scrapped a $5 billion Australian IPO, NextDC shares fell 40% after a power backlash, and SK Hynix abandoned a US unit IPO citing volatility, signaling doubts over AI infrastructure valuations.
MiniMax and Z.AI (Zhipu) have shed a combined $150 billion in market value from their highs after their January debuts. The selloff continues amid intensifying competition and the threat of rival listings next year, with no end in sight.
inDrive signed an MOU with Pony.ai and Nvidia to test Pony.ai's Gen-7 robotaxis in Astana, Kazakhstan, in the first half of 2027, starting with a safety operator and aiming for fully driverless rides. Pony.ai supplies the vehicles, Nvidia provides its Hyperion platform and Halos safety system, and inDrive operates them via its app in a capital-light model.
ZTE launched the ZXEDC M8000 Intelligent Computing Micro-Module at Data Center World Asia, integrating enclosed cold aisle, liquid-cooled racks, CDU, high-voltage DC and intelligent busbars. It targets rising rack power density and the shift from air to liquid cooling, supporting high-density AI server deployment and switchable cooling.
Bloomberg reports that a gauge of the inverse relationship between Treasuries and AI-focused stocks has strengthened to levels last seen a year ago, as a deepening tech selloff pushes investors back into bonds. The haven role had faded amid inflation and rate worries.
An Anthropic security official warned that cyber threats are worsening and critical infrastructure is at risk, signaling growing concern at frontier AI firms about the security challenges posed by AI capabilities.
Bloomberg reports the strongest risk-off signal in global bond markets in more than a year. A selloff in AI stocks plus rising geopolitical tensions drove investors into safe-haven assets, lifting global credit spreads to a 17-month high.
Delta Electronics VP Kelvin Huang told Bloomberg Tech Asia that the company expects strong growth in AI data-center demand for years to come. The Taiwanese power and thermal management supplier is closely tied to data-center buildouts.
Wui Foo, president and co-founder of cybersecurity firm FailSafe, told Bloomberg that institutions must protect websites and services from vulnerabilities and AI-driven cyberattacks amid a recent surge in security loopholes.
OpenAI is courting community support in Effingham County, Georgia, for Project Camellia, a $20 billion data center campus. CFO Sarah Friar spoke at a local business summit amid concerns over transparency, power and water use. A Gallup survey found seven in 10 Americans oppose AI data centers in their communities, and the project has drawn protests and legal challenges.
MiniMax published the MiniMax-M3.1-Flash model on Hugging Face, with the linked post offering few details. The release extends its model lineup, though capabilities and availability remain undisclosed.
MIT researchers left 13 Eko agents, each running on Claude Opus 4.7, alone on a simulated desert island for 30 hours, where they invented their own games and goals. In later specific tasks such as stacking objects, these agents generally outperformed identical agents that never visited the island. The researchers see goal-free exploration as a new way for agents to learn skills.
SK Group's SK Hyper is in talks with big tech firms and financial investors to finance a first 5-gigawatt data center phase costing about $350 billion including equipment. The company plans 15 gigawatts of capacity in South Korea by 2035, joining the global AI infrastructure race despite concerns about future demand.
Hugging Face released a Space that checks whether popular Hub models run unmodified on AWS Trainium. Results show sharding cores, wall time including compilation and peak memory, plus a reproduction script. Only Transformers text-generation and image-text-to-text models are tested for now.
OpenAI and Polish research institute NASK are partnering on AI safety, security and technical research to strengthen safeguards and support a more resilient digital ecosystem. The move extends OpenAI's AI-safety governance work through partnerships with national institutions.
A Hacker News post argues that AI compute is the new oil and requires a new market to allocate it. The item links to the Hacker News discussion thread and offers no further detail.
Uber burned through its entire 2026 AI budget in four months and now caps employees at $1,500 per month per AI coding tool. Amazon, Walmart, Cisco, Meta and others are limiting access, setting spending caps or shifting work to cheaper models. Customers aren't abandoning AI but are blending price tiers and demanding proof of value, pressuring OpenAI and Anthropic.
At its 2026 CP World conference, Cruise Planners announced 23 new tools and enhancements, mostly AI-driven, offered free to roughly 1,100 travel advisors. Headliners Maxx Assist (conversational itinerary building) and Maxx Reach (marketing) build on its 2023 MaxxIntelligence launch.
Load more