Back to list
Daily 2026-05-24

AI Pulse Daily | 2026-05-24

AI IndustryGoogleAnthropicDeepSeekOpenAIFei-Fei Li

1. Google CEO Pichai Admits Gemini Is Behind in Coding

In a rare candid interview on the New York Times technology podcast, Google CEO Sundar Pichai openly acknowledged that Google’s Gemini is indeed lagging behind competitors in coding. He admitted that in areas like agent-based programming with tool use, instruction following, and long-horizon multi-step tasks, Google is “a bit behind.” Pichai also shared that the pace of AI progress over the past year or two makes him feel “AGI may be closer than previously imagined,” with 30 to 60 days of industry change now equivalent to what used to take 5 years. These comments came right after Google I/O, where Gemini 3.5 Flash, Gemini Omini, and Gemini Spark were just announced.

AI Pulse View: It is extremely rare for a Big Tech CEO to publicly admit their AI product is lagging. This suggests the coding agent race has reached white-hot intensity. Pichai’s candor may signal a major strategic pivot within Google, and his “AGI is closer” assessment deserves serious attention from the industry.

Source: 36Kr | QbitAI | 2026-05-24

2. Anthropic’s $30B+ Funding Round to Surpass OpenAI Valuation

According to Bloomberg on May 23, Anthropic is poised to close a mega funding round exceeding $30 billion, potentially finalized as early as next week, pushing its valuation above $900 billion. Once completed, Anthropic will formally surpass OpenAI as the world’s most highly valued AI startup. Anthropic’s Q2 revenue is projected to reach $10.9 billion, more than doubling from the previous quarter, and the company is on track for its first-ever profitable quarter.

AI Pulse View: The AI investment landscape is shifting from OpenAI’s dominance to a two-horse race. Anthropic’s strong Claude lineup has won over capital markets, and its valuation overtaking OpenAI will reshape the entire AI industry. Both companies’ expected fall IPOs make this competition even more compelling.

Source: GeekPark (citing Bloomberg) | 2026-05-24

3. DeepSeek V4 Price Permanently Cut to 1/4 of Original; CATL, JD, NetEye Negotiate Investment

DeepSeek announced that DeepSeek-V4-Pro API pricing will be permanently adjusted to 1/4 of the original price after the May 31 promotion ends: input (cache hit) at 0.025 yuan/million tokens, cache miss at 3 yuan, output at 6 yuan. According to Bloomberg and The Information, DeepSeek is pursuing approximately 70 billion yuan in funding at a $45 billion valuation. CATL (Contemporary Amperex Technology), JD.com, and NetEase are negotiating to participate. Founder Liang Wenfeng has made clear to investors that DeepSeek will prioritize pursuing AGI over short-term monetization and has committed to its open-source route.

AI Pulse View: DeepSeek’s move to make its “discount” permanent signals a shift from promotional pricing to long-term low-cost AI. CATL, a battery giant, entering AI investment reflects how the AI race has expanded from the model layer to energy infrastructure. Energy is becoming the most underestimated strategic resource in the AI competition.

Source: QbitAI | GeekPark | 2026-05-23

4. Apple Preps “Gen AI” Website Ahead of WWDC, Subdomain Already Live

Apple is preparing to launch a new “Gen AI” website, with the subdomain genai.apple.com recently added to Apple’s DNS records. This development comes weeks before WWDC 2026, where Apple has confirmed it will unveil a series of “AI-related updates” around Apple Intelligence, Siri, and more. The WWDC 2026 keynote is scheduled for June 8.

AI Pulse View: Apple registering a dedicated Gen AI domain ahead of WWDC suggests the company is systematically consolidating its AI strategy branding. Given Apple’s historically conservative stance on AI, this could foreshadow a major AI product announcement worth watching.

Source: GeekPark (citing cnBeta) | 2026-05-24

5. OpenAI Hiring Safety Researchers at Up to $445K Salary

Business Insider reported on May 23 that OpenAI has posted a safety researcher position for its Preparedness team, offering $295K-$445K annually. The role focuses on studying risks when AI can train “stronger versions of itself,” with emphasis on defending against data poisoning attacks, developing model interpretability tools, and tracking AI coding tool adoption. OpenAI specifically seeks candidates with “good taste and strategic thinking.”

AI Pulse View: OpenAI’s high-paying recruitment focused on “recursive self-improvement” safety suggests the company is upgrading its risk assessment for AI self-iteration capabilities. This proactive safety positioning reflects leading AI companies’ preparedness for potential risks on the AGI path.

Source: GeekPark (citing Business Insider) | 2026-05-24

6. Fei-Fei Li’s Team Releases ESI-Bench: The ImageNet for Spatial Intelligence

Fei-Fei Li’s team has released ESI-Bench, a new benchmark specifically designed for evaluating embodied spatial intelligence. Unlike previous “passive perception” evaluations, ESI-Bench transforms the observer into an active agent, closing the perception-action loop. The benchmark includes 10 task categories and 3,081 task instances, covering four dimensions of core human spatial cognition. Testing revealed that while current AI excels at image recognition, it falls short in active exploration strategy and metacognition — models “don’t know what they don’t know.”

AI Pulse View: ESI-Bench marks a paradigm shift in spatial intelligence evaluation from passive vision to active exploration. The “action blindness” and “metacognitive deficits” identified by Li’s team point to core bottlenecks in embodied AI research, providing a new directional compass for robotics and embodied intelligence.

Source: QbitAI | 2026-05-22 (Paper: arXiv:2605.18746)

7. OpenAI Codex Power-User Guide: How to Make AI Agent Your Real “Employee”

Jason Liu, author of the 13k-star open-source library Instructor (now on OpenAI’s Codex team), published a comprehensive Codex advanced usage guide. Key strategies include: using Heartbeats+@computer for scheduled task automation (e.g., checking Slack and Gmail every 30 minutes), storing personal memory in local Obsidian knowledge bases rather than platform-internal systems, using verification mechanisms to determine task completion, and leveraging Goal mode to let Codex work autonomously for hours to days. Codex’s latest updates also include lock-screen continuation and remote mobile monitoring.

AI Pulse View: Jason Liu’s Codex methodology represents the evolution of AI coding tools from “conversational Q&A” to “autonomous work systems.” Storing memory locally and setting verification loops provides practical reliability methodologies for AI agents — a must-read for every AI developer.

Source: QbitAI | 2026-05-23

8. DeepSeek API Speed Boost and Scaling, 500 Concurrent Connections by Default

On May 23, DeepSeek announced its API has completed output speed improvements and service scaling, now supporting 500 concurrent connections by default. Enterprise users needing higher concurrency can apply online. This follows a nearly 12-hour system crash DeepSeek experienced previously, highlighting the urgency of building its own infrastructure.

AI Pulse View: 500 default concurrent connections significantly lowers the technical barrier for enterprises to adopt DeepSeek. Combined with the V4 Pro permanent price cut, DeepSeek is using a “lower price + more capacity” combo to accelerate developer ecosystem growth.

Source: GeekPark (citing IT Home) | 2026-05-24

9. NVIDIA CFO Mocks Competitors: Memory Price Hikes Could Have Been Avoided

NVIDIA CFO Colette Kress recently stated that the current memory shortage sweeping the AI industry could have been avoided with earlier ordering, implying competitors failed to anticipate price surges. She emphasized that NVIDIA foresaw the AI wave would push up HBM and DDR memory prices and locked in large supplies early.

AI Pulse View: NVIDIA’s forward-looking supply chain management once again demonstrates that its moat in AI chips lies not just in technology, but in deep industry chain control. Memory shortages may become the most critical supply chain bottleneck for the AI industry in 2026.

Source: GeekPark (citing cnBeta) | 2026-05-24

Other Updates