Lookonchain APP

App Store

Google employees kept a report critical of DeepMind on their desks for daily review, and three months later, research firm SemiAnalysis apologized, admitting it had misjudged DeepMind.

11 minutes ago

Insight Beating AI News Flash: Arthur Douillard, a member of Google Gemini’s reinforcement learning team, revealed that the team printed out SemiAnalysis’s bearish report on DeepMind and kept it on their desk to read daily. While the broader industry viewed Google as having fallen irreparably behind and unable to catch up, the team was determined to prove the analysts wrong. The report, published in August this year, was titled “Gemini is Cooked but GCP is Cooking”. At the time, SemiAnalysis argued that DeepMind had lost a large number of top talents, Google Cloud had seized massive computing power, and model development could no longer keep pace with OpenAI and Anthropic. It even directly asserted that DeepMind was no longer a top-tier AI lab and had basically no chance of catching up with the first echelon. Just over three months later, Google released four Gemini models in succession: Gemini 4 Argon, Carbon, Hydrogen, and Neon. Carbon once claimed the top spot on Artificial Analysis’s overall intelligence ranking, while the latest Neon ranks third globally overall, taking first place in programming agent evaluations—slightly outperforming Claude Fable 5.2, with average task costs less than half of the latter’s. Arthur specifically took to social media to thank SemiAnalysis: “Thank you for the motivation. The issues outlined in your report did exist at the time, but things have changed.” This time, SemiAnalysis backed down, posting a “Google Apology Form” listing several of its own mistakes, including “only looking at X”, “failing to review the model card”, and even blaming Mercury retrograde. It finally admitted that it had misjudged Gemini 4. Demis Hassabis, founder of DeepMind, also joined in the fun: “We accept your apology! But you missed one key point: underestimating this extremely formidable Google DeepMind team.”

Relevant content

Japanese and South Korean stock markets both closed down more than 5%.

According to Bitget market data, South Korea’s KOSPI index closed 267.93 points lower on Friday (Nov 20), down 5.21% to 4869.96 points, marking its lowest level since Jan 16. Samsung Electronics fell 7%, while SK Hynix dropped 5.91%. Japan’s Nikkei 225 index closed 3007.56 points lower on Friday (Nov 20), a 5.06% decline to 56436.65 points, its lowest since April 9.

1 seconds ago

Binance will launch TradFi US dollar-margined perpetual contracts for JNJ, AMDL, and NIO, with up to 20x leverage.

Binance announced it will launch JNJUSDT, AMDLUSDT, and NIOUSDT USDT-margined perpetual contracts in sequence on November 20, with trading opening at 09:30 UTC, 09:35 UTC, and 09:40 UTC respectively. The three contracts are pegged to Johnson & Johnson (NYSE: JNJ), the 2x Daily Long AMD ETF (Nasdaq: AMDL), and Nio Inc. (NYSE: NIO). All support 24/7 trading, with a maximum leverage of 20x and a minimum notional value of 5 USDT. The funding rate for these contracts settles every 8 hours, with a cap of +2% and a floor of -2%. Binance specifically clarified that even if the funding rate hits these limits, it will not automatically adjust the settlement frequency to hourly.

1 seconds ago

Binance to Remove MORPHO and HYPE Seed Tags, Adds BANK, INIT and Others to Watchlist

Binance announced that, based on recent review results, it will expand its Monitoring Tag list on November 20, adding six tokens: Lorenzo Protocol (BANK), Initia (INIT), Mitosis (MITO), Harmony (ONE), THENA (THE), and Zilliqa (ZIL). Meanwhile, Binance will remove the Seed Tags from Morpho (MORPHO) and Hyperliquid (HYPE). The exchange stated that tokens with a Monitoring Tag usually have high volatility and carry significant risk. It will conduct regular reviews of these assets, and projects that no longer meet its listing criteria may face delisting. Review indicators include team commitments, development activity, trading volume and liquidity, network security, token economic models, and project operational transparency.

1 seconds ago

Xie Jiayin: Bitget’s VIP round and public round rANT tokens have been credited to users’ accounts; unallocated funds are available for participation in high-yield current wealth management.

Xie Jiayin, head of Bitget's Greater China region, announced that Bitget's VIP round and public round rANT tokens have been credited to users' accounts. Unsubscribed USDT and USDGO can be used to participate in a 15-day current wealth management product with an annual percentage rate (APR) of up to 10%.

1 seconds ago

WEEX and Jingzhe Quant Reach Strategic Cooperation to Co-Build New Multi-Strategy Trading Scenarios

Crypto exchange WEEX has announced a strategic partnership with Jingzhe Intelligent Strategy Platform to further expand its WEEX Builder ecosystem. Under the collaboration, WEEX will leverage its WEEX API trading capabilities to integrate with Jingzhe Quant’s DIY strategy plaza, creating a seamless end-to-end workflow spanning strategy screening, historical backtesting, parameter adjustment, and live trading. Jingzhe Quant, which has undergone two years of iteration, has cumulatively served over 100,000 users, covering multiple strategy types including trend, range-bound, and short-term trading. It offers features such as unified APP management, AI assistant-aided research, take-profit/stop-loss, and risk control, aiming to deliver the most premium trading experience for users. Both parties noted that the partnership is designed to enrich WEEX’s quantitative trading use cases, provide users with a more convenient strategy management and trade execution experience, and support channel partners in expanding their customer service capabilities. Going forward, the two sides will deepen cooperation around strategy ecosystem, product experience, and user education.

1 seconds ago

Attacker steals 5.2M $XMR1, dumps on Hyperliquid and buys 7,467 $ETH with 13.22M $USDC

Has $XMR1 been hacked? Attacker 0x1464 received 5.2M $XMR1 and dumped on Hyperliquid, then used the proceeds to buy $ETH. So far, the attacker has spent 13.22M $USDC to buy 7,467 $ETH.

1 seconds ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano