BLOG / INTELLIGENCE
Blog
310 articles, updated continuously.
Singularity-Bench Six AI R&D Questions: The inference engine written by Claude reduced the first token delay from 12 seconds to 0.48 seconds
Researchers at Carnegie Mellon University and other institutions have launched a new benchmark called Singularity-Bench. What it tests is not writing functions or fixing bugs, but whether the AI agent can build and strengthen AI R&D infrastructure by itself. There are six questions in total, all related to...
2026.11.20Gemini 4 Neon vs. Claude Opus 5.5: Tie for programming agents, $6.86 vs. $12.61 per task, Opus also has two discount buttons
Last month, a developer used Gemini 4 Argon to check an old project in Antigravity and posted that it found a bug that was not seen in other cutting-edge models. However, in the same conversation, it also made up several calls that did not exist in the code...
2026.11.20Genie 4 Open to Developers: 720p Interactive World Billed by the Minute, About $6 a Minute
Google has opened the world model Genie 4 to developers, which can be called through the Gemini API and Google AI Studio. The consumer side will be online one day earlier. In the Project Genie app, AI Ultra order...
2026.11.20The Tamil Nadu public library in India purchased nearly 1,800 AI books, but no one found them after four review committees
The Indian state of Tamil Nadu used public funds to buy books for a new library, and ended up buying a batch of books written by AI. The person who discovered the problem was Tamil writer Perumal Murugan, the first Tamil author to be shortlisted for the International Booker Prize. He flipped through the books and saw errors, like AI hallucinations...
2026.11.20There is a gray vertical bar on the left side of the Google homepage: click to open the historical conversation in AI mode, and you can also connect to the Gemini notebook
The Google homepage has been redesigned again. This time, a gray vertical bar is added to the left side of the screen. Clicking it opens a collapsible AI search sidebar. The sidebar contains the records you have searched using AI Mode. You can create new conversations, search historical conversations, and connect to Gemini...
2026.11.20Micron says memory will be tighter in 2027 than this year: 75% of next year’s production capacity has been sold out, and contracts for 2028 are under negotiation
The memory shortage is not over yet, and Micron says it will be even worse next year. Micron President and Chief Technology and Product Officer DeBoer said in an interview that in the foreseeable future, the supply and demand of memory chips cannot find a clear balance point, and 2027 will be "more tense than 2026." CEO Mehrotra in Os...
2026.11.18Microsoft Decision-1.1 is specifically designed for corporate decision-making purposes. If you’re unsure about a particular option, you can simply opt to abstain from making a choice. There are no fees associated with using this tool.
The model released by Microsoft this time does not engage in conversations with users; it is only capable of handling multiple-choice questions. On November 17th, Microsoft launched the decision-making model Microsoft-Decision-1.1, which has already been made available on platforms such as Microsoft Foundry, OpenRouter, and Vercel. Its purpose...
2026.11.18Google discovered that hackers using AI systems are capable of automatically applying patches to carry out attacks. It is suspected that North Korean groups are involved in this, and they have also modified the configuration settings of Claude Code.
No sooner had the software been updated with security patches than the attack code was already written and put into use, without anyone being involved in the process. In a new report, Google’s Threat Intelligence team described such a tool. Whenever a popular web software releases security patches, this AI compares the code before and after the updates…
2026.11.18Microsoft’s first voice-to-voice model, MAI Realtime 1: Can be interrupted; costs $0.01 per minute of audio processing time.
Microsoft has finally developed its own voice-to-voice model, and it immediately focused on price competition. At the Ignite 2026 conference on November 17th, Microsoft launched MAI Realtime 1. This product is designed specifically for real-world communication scenarios in corporate environments: it can recognize interruptions from the other party and provide timely confirmations when necessary…
2026.11.18Google has added a new option for Gemini: chat records can be retained, but they will not be used to train models.
Those using Gemini can now simply keep the chat records without having to provide the training data. Google has issued a statement announcing updates to the control options available in the Gemini app. The newly added settings allow users to retain their chat records while choosing not to use those records for training generative AI models. In other words…
2026.11.18The developers asked the OpenAI agents to recreate the gameplay of League of Legends, but Claude in the same group immediately stopped them.
Last week, a developer assigned a task to his AI team: within his own game, they were required to recreate three heroes from League of Legends – Master Yi, Ocean Thrall Orlean, and Frost Archer Ashe – along with the corresponding game map. He allowed the team to use up to five rented GPUs for this project, and assigned the task to Ope…
2026.11.18The provision of GPU loans worth $480 million has been suspended, and five banks have expressed their intention to withdraw from this initiative. The Ministry of Commerce has stated that it will take countermeasures in response.
As soon as the new US regulations restricting “remote card usage” came into effect, the first transaction that was affected by these restrictions came to light. According to people familiar with the matter, a syndicated loan worth $480 million has been put on hold. The borrower is Aolani Cloud, a newly established cloud services provider headquartered in Singapore; this funds were originally intended for use in Malaysia…