Lookonchain APP

App Store

Databricks sweeps NVIDIA’s Kernel ranking with AI, securing first place in four categories across 235 tests, with token costs totaling roughly $70,000.

1 hours ago

Beating AI Briefing: The Databricks team announced that they integrated GPT-6 Astra and Opus 5 into an automated optimization loop, enabling the models to continuously rewrite, test, and refine GPU kernels, ultimately claiming first place across all four tracks of the NVIDIA SOL-ExecBench benchmark. The benchmark includes 235 GPU kernel tasks covering basic operators, complex fused operators, low-precision computing, and real-world large language model inference. The team stated that the total token cost for the entire process was approximately $70,000. Built on KDA and Humanize, the system allows AI to autonomously write kernels, run correctness and performance tests, and iteratively modify them based on results. The KDA team recently applied a similar method to Kimi Delta Attention, achieving 2.96x the speed of Moonshot AI’s official FlashKDA on the NVIDIA B300. Both results are different applications of the same agent kernel optimization system. The Kimi project proved that a highly complex attention kernel can be rewritten by AI, while Databricks’ work extended this approach to hundreds of distinct kernels.

Relevant content

Claude Code optimizes AI applications by adjusting prompts, switching models, and rolling back changes when performance degrades.

Beating AI Insight News Brief: Anthropic has added an automatic tuning workflow to Claude Code’s official claude-api Skill. The build-eval module first generates test sets and scoring metrics from real conversations, bugs, or manual cases; hillclimb then enables Claude Code to iteratively modify system prompts, tool descriptions, models, and inference intensity. After each adjustment, tests are rerun: improvements are retained, while regressions are rolled back. To prevent Claude from overfitting exclusively to test cases, tuning is not performed on all samples at once. A subset of cases is used to identify issues and adjust configurations, while a separate batch is reserved to validate if modifications work on unseen problems. If the primary score improves but results on the reserved test set do not, the modification may be deemed overfitting and withdrawn. Anthropic demonstrated this workflow using 44 customer support tickets. Claude Code sequentially cleaned prompts, added business rules, tested different models, and reduced inference intensity, ultimately settling on a low-effort configuration for Sonnet 5. On 14 tickets not included in tuning, accuracy rose from 78.6% with the initial configuration to 90.5%, while costs fell to roughly one-fifth of their original level.

9 minutes ago

Major South Korean internet securities brokerage Kakao Pay partners with Ondo Finance and Dinari to explore tokenization of Korean stocks.

South Korea's Kakao Pay Securities has partnered with Ondo Finance and Dinari respectively to explore tokenizing South Korean-listed stocks and providing overseas investors with on-chain investment channels. Kakao Pay Securities is one of South Korea's leading online securities firms, backed by the Kakao financial ecosystem. As of the first half of 2026, its client assets under custody exceeded 21 trillion won (approximately $150 billion), with around 1.61 million monthly stock trading users. The collaboration aims to research integrating South Korean stock assets onto the blockchain, boosting the accessibility of traditional securities assets via tokenization and creating a new channel for overseas investors to access South Korea's stock market.

9 minutes ago

Binance will delist the USDC spot trading pairs of AUCTION and VANA.

According to official announcements, Binance will delist the AUCTION/USDC and VANA/USDC spot trading pairs at 11:00 AM Beijing time on October 2.

9 minutes ago

X is testing an AI all-in-one subscription package, bundling Cursor, Grok, and X into a single plan.

Beating AI Express News: X is testing a new unified subscription plan. An early Android test page leaked by the community features an "X Subscription" upgrade card, which states that a single plan will include access to Cursor, Grok, and X’s core benefits, with a Grok Bot icon displayed at the top. Maxime Prades, head of SpaceXAI’s product team, previously confirmed the company is developing a "more unified subscription" that will roll out in "weeks, not months." Currently, several subscription tiers have partial integration but remain charged separately: Cursor Pro, Pro+, and Ultra all include Grok Bot access; SuperGrok, SuperGrok Plus, Heavy, and X Premium+ also grant Grok Bot benefits to Cursor accounts. However, Cursor’s official statement clarifies that this linking does not convert Grok or X subscriptions into Cursor plans, nor does it cancel existing Cursor subscriptions.

9 minutes ago

Japan's Financial Services Agency approves stablecoin trade settlement proof-of-concept experiment

According to an official announcement on the website of Japan’s Financial Services Agency (FSA), the regulator announced on September 29, 2026 that it has approved the fourth support project under the FinTech Proof-of-Concept Center’s Payment Advancement Project (PIP). The project was jointly applied for by TradeWaltz, NTT Data, Mizuho Bank, MUFG Bank, Sumitomo Mitsui Banking Corporation, and MUFG Trust Bank. The proof-of-concept experiment will verify the practical application of stablecoins in trade transaction fund settlement and sort out relevant legal arguments, running from September 2026. Upon conclusion, the FSA will publish the experiment results and relevant compliance and regulatory conclusions on its official website.

9 minutes ago

The Ethereum Foundation will activate the Glamsterdam upgrade on the Sepolia testnet on October 6.

The Ethereum Foundation announced that the Glamsterdam network upgrade will activate on the Sepolia testnet at 21:53:36 Beijing Time on October 6, 2026 (epoch 353,024, slot 11,296,768). Activation timelines for the Hoodi testnet and Ethereum mainnet have not yet been finalized. The upgrade integrates the Amsterdam execution layer and Gloas consensus layer, with core improvements including: built-in proposer-builder separation (ePBS, EIP-7732), which formally incorporates block proposal and construction into the protocol; block-level access lists (BALs, EIP-7928), enabling parallel state reads and transaction verification; and adjusted gas pricing that more accurately reflects execution and state growth costs (EIP-8037, EIP-8038, etc.). Node operators must update their execution layer and consensus layer clients prior to activation. Corresponding client versions have been rolled out sequentially; detailed compatibility lists are available in the official announcement. Mainnet users and ETH holders do not need to perform any actions at this time.

9 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano