AI Industry Development & Policy Dynamics

Bitcoin miner Riot's shares rose 8% following an expansion of its AMD data center agreement. The expanded partnership includes improved financing terms, underscoring Riot's strategic shift from bitcoin mining toward artificial intelligence data center operations. The move signals growing confidence

According to Decrypt, Minnesota's legislature passed a bill on May 2 prohibiting websites and applications from providing AI tools that generate non-consensual intimate images of identifiable individuals. The law bars platforms from allowing users to access or use such "deepfake" tools, and bans

According to Sam Altman's announcement today, ChatGPT users can now log in to the OpenClaw agent platform using their ChatGPT accounts and directly access their existing ChatGPT subscriptions on the platform.

According to CoinDesk, on Thursday, Bittensor (TAO) surged 5.5%, leading the CoinDesk 20 index higher. Bitcoin (BTC), up 1.9% from the same period, was also among the top performers.

AI agent Manfred has formed its own company and obtained a crypto wallet along with credentials to hire staff, make payments, and conduct business, according to reports. The agent plans to begin trading crypto by the end of

According to Beating, xAI launched Grok Custom Voices and Voice Library, allowing users to record one minute of audio in the xAI console to generate a custom voice_id for use with Grok TTS and Voice Agent APIs. The feature supports applications including customer service agents, content creation,

Turing Award winner David Patterson said that as AI moves from training to large-scale inference, the next memory bottleneck may not be HBM, but HBF (high-bandwidth flash memory). HBF provides NAND Flash stacking for high capacity and low power consumption, handling the access to context and intermediate data during inference, which is different from the division of labor with speed-focused HBM. SK hynix and SanDisk are driving standardization, and they expect that by 2038, demand for HBF may surpass HBM.

GEPA is a new AI training method proposed by the UC Berkeley team: it does not update weights, does not require GPUs, and improves performance by having reflective LLMs read the full task logs and rewrite prompts, rather than only providing feedback via scores. The six tasks average a win rate improvement of 6%, with a maximum of 20%, and training rollouts are reduced by 35x. Integrated with DSPy, the Full Program Adapter shows significant improvements in mathematical and multimodal workflows, with 93% accuracy. The code is open-sourced, and the authors include Matei Zaharia and others.

Anthropic announced on May 1 that the first “Code with Claude” developer conference will open in San Francisco on May 6, and for the first time will expand into a three-city tour covering London (5/19) and Tokyo (6/10). In-person seats for the three main sessions are allocated via a lottery and have all already been selected, but live-stream registration remains open to everyone. To address demand overload for the in-person events, the San Francisco venue will add an extra “Extended” session on 5/7, designed specifically for independent developers and founders in the early stage. Three-city tour: 5/6 San Francisco, 5/19 London, 6/10 Tokyo Code with Claude is a developer conference hosted by Anthropic; the San Francisco main event has already been held for the second time. The content of all three events is the same: full-day on-site workshops, demos of the latest features, and Claude along every line

OpenAI posted 3 data points on May 1 from its official account after GPT-5.5 went live for one week: self-rated as the “strongest release of all time,” API revenue growth speed at more than 2x that of any previous model release, and Codex doubling its revenue in less than 7 days. OpenAI attributed the cause to the continued rise in enterprise demand for agentic coding (proxy-based coding) tools, and said it followed the same product cadence as Anthropic Mythos in a concurrent matchup with GPT-5.5-Cyber on April 30. 3 items from the disclosed data: API revenue growth pace, Codex doubling in 7 days, strongest release The 3 key figures OpenAI shared this time come from official tweets, with no detailed financial report attached: first, GPT-5.5 is the “strongest revenue growth” model of all time…

According to Beating, members of the U.S. House Foreign Affairs Committee will travel to Silicon Valley next week to meet with representatives from Google, Anthropic, Meta, Tesla, Intel, Applied Materials, and Nvidia to discuss artificial intelligence and export controls. An industry roundtable is s

According to Beating, OpenAI has added a new "Codex Pets" feature to the Codex desktop application, allowing users to spawn and interact with an animated virtual companion. Users can activate a pet by typing /pet in the editor. The feature functions as an agent status indicator, displaying a

This report says Valerie’s vending machine is powered by OpenClaw, enabling an AI agent to actually run day-to-day operations, including pricing, inventory, community management, and even managing a bank account. Testing shows the AI can simulate corporate executives’ decision-making, but it also exhibits pricing hallucinations, such as raising a protein bar to $15. Legal hurdles like KYC, bank account opening, and food licensing have not yet been overcome. Although mechanisms such as trusts allow the AI agent to hold a business, practical challenges remain plentiful.

AISI released an assessment of GPT-5.5’s network attack capabilities in May: Expert difficulty 71.4%, Mythos Preview 68.6%. The gap is within the margin of error, essentially tied. GPT-5.5 has become the second system, after Mythos, that can automatically complete the 32-step enterprise intrusion of “The Last Ones.” It also found a universal jailbreak that can be developed in about 6 hours, enabling it to bypass malicious query filtering. Going forward, it will monitor the next round of evaluation timing and OpenAI’s update on this.

The U.S. Department of Defense announced in May that it signed confidential military networking deployment contracts with seven companies—SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and Amazon Web Services—then added Oracle as the eighth. The contracts allow models to run at the highest classified levels, Impact Level 6/7. The focus is on three major applications: data integration, combat decision-making, and battlefield situational awareness, emphasizing distributed risk and avoiding vendor lock-in. Anthropic was blacklisted for refusing to accept the military’s security guardrails and did not receive a contract. AMD was not directly listed; GPUs are provided by NVIDIA and others. Next, watch whether Anthropic backs down and what roles new selectees such as Reflection will play.

Sunnyvale, California-based AI chipmaker Cerebras Systems is seeking up to US$4 billion in an IPO that could value the company at approximately US$40 billion, according to Bloomberg. Formal marketing could begin as soon as May 4, with banks receiving more than US$10 billion in indications of

According to Beating, China's National Development and Reform Commission halted Meta's $2 billion acquisition of AI agent company Manus, prompting the China Securities Regulatory Commission to tighten approval for Hong Kong IPO listings by red-chip companies (Chinese firms holding domestic assets th

According to CoinDesk 20 index, Bittensor (TAO) gained 5.5% from Thursday, leading the index higher, while Bitcoin (BTC) rose 1.9% during the same period.

According to multiple media reports, Elon Musk completed three days of testimony in his lawsuit against OpenAI on April 28-30, with cumulative testimony lasting over 7 hours. Musk, who co-founded OpenAI in 2015 with an initial investment of $38 million, claimed the company abandoned its non-profit m

According to Bloomberg, Meta acquired Assured Robot Intelligence, an AI robotics startup with offices in San Diego and New York, on May 1. The acquired team, including co-founders Lerrel Pinto and Xiaolong Wang, will join Meta Superintelligence Labs and work with Meta Robotics Studio. Financial