TypeSafe has upgraded Jev to version 1.14: The tool can locate objects from images, and returns coordinates even when it fails to identify the target.

Beating AI Express: 17-year-old AI researcher Nathan Chen (Chen Guangyu) announced he has recently left the Kimi team of Moonshot AI. He joined in November 2025 at just 16 years old, with his tenure lasting less than a year. Chen stated that the past year has been the best experience of his life to date, but he did not disclose the reason for his departure or his next career move. Chen gained prominence for Kimi’s Attention Residuals paper—research publicly praised by Elon Musk—for which he, Zhang Yu, and Su Jianlin served as co-first authors. Su Jianlin first proposed replacing traditional residual connections with attention mechanisms and validated the preliminary effects. Chen and Zhang later joined the study, co-developing Block AttnRes, a design that reduces memory usage and communication overhead, enabling the architecture to be applied to large-scale model training, and was subsequently adopted in Kimi K3. In his departure post, Chen also reviewed other work he contributed to: he and Zhang optimized the GPU compute kernel for Kimi Delta Attention, researched model training optimizers, and helped create the demo cases for K3’s launch. He thanked the team for offering him the opportunity when he was only 16, adding that he plans to continue exploring more directions in the future.

As of October 15, 2026, Bitcoin has dipped three times to around $81,000 on October 9, 12, and 15, repeatedly testing the same support zone. The latest downturn was compounded by macro headwinds including higher-than-expected U.S. inflation, growing market concerns over further monetary tightening by the Federal Reserve, and weakness in U.S. stocks, putting pressure on risk assets overall. Crypto market analyst Murphy noted that data from on-chain entity-adjusted realized losses shows a clear declining trend in loss volume across the three dips: the loss on October 12 was smaller than that on October 9, and while the loss on October 15 expanded slightly from the 12th, it still remained lower than the level on the 9th. This indicates that October 9 marked a peak where short-term price-sensitive chips were concentratedly sold off; during the second dip, the market was dominated by wait-and-see sentiment with weak selling willingness; the third decline, catalyzed by macro headwinds, forced some undecided chips to stop loss again. Therefore, the current market is closer to a gradual clearance of "swing chips" rather than a full-scale panic sell-off. If the loss ratio of short-term holders remains at only around 31%, below the historical cumulative average of 42.7%, it suggests that chip clearance may not be complete, and further volatility and selling pressure cannot be ruled out going forward. Attention should next focus on whether the scale of realized losses continues to converge: if prices stay near the current support zone while loss release gradually weakens, it means passive selling pressure is decreasing, and the phase correction may be nearing its end. Conversely, if the loss scale expands significantly again, it indicates the market still needs more sufficient chip turnover.

According to Bitget's market data, the Nikkei 225 index extended its decline, plunging 5%.

According to Bitget market data, South Korea’s KOSPI index plunged 8.00% intraday to 5527.29 points. SK Hynix fell over 11% while Samsung Electronics dropped more than 9%. Meanwhile, South Korea’s exchange activated the circuit breaker for the KOSDAQ (South Korea’s secondary growth index), suspending trading for 20 minutes.

Beating AI News Flash: AI infrastructure firm Modal announced it has closed a $750 million Series D funding round led by Accel, with a post-money valuation of $15.75 billion. In May this year, Modal raised $355 million at a $4.65 billion valuation, when its annual recurring revenue (ARR) had just crossed $300 million. In less than five months, its ARR has surged past $1 billion, usage has tripled, and the company reports its business has grown more than 15x year-over-year. Modal helps developers run AI models, training tasks, and agents without needing to purchase GPUs or build their own servers. It integrates computing power from over 20 cloud platforms and has developed its own file system, container runtime, and scheduling system from scratch. Its Sandboxes business is a key growth driver: AI-generated code requires a secure environment for execution and testing, and Modal has turned this isolated setup into an on-demand cloud service, with the segment now generating $300 million in annual recurring revenue alone. The platform has launched nearly 3 billion sandboxes in total, and achieved 1 million concurrent runs in internal testing. Its customers span multiple industries: Cognition and Ramp use it to run programming agents, Runway leverages it for video generation, and Physical Intelligence uses it for robot training. Modal also highlighted TypeSafe AI’s Jev model, which processed over 1 trillion tokens via its compute platform just three days after launch.

According to Bitget market data, South Korean stocks SK Hynix fell more than 11% and Samsung Electronics dropped over 9%. The Southern 2x Long SK Hynix ETF declined 22.04%, while the 2x Long Samsung Electronics ETF fell 19.72%.