Latest AI News

Published on April 24, 2026 at 10:14 Commenter 0 𝕏 Print © Tim Reckmann, CC BY-NC 2.0 DeepSeek has unveiled two new models: DeepSeek V4 Flash and DeepSeek V4 Pro. The latter would be capable of processing 1 million characters at once. This launch is part of the competition between China and the United States in the field of open-source AI. DeepSeek announced « the era of a million-character context at a lower cost ». Two models were presented: DeepSeek V4

Published on April 24, 2026 at 13:13 Comment 0 𝕏 Print © Meta Third layoff in four months. Meta is laying off 10% of its workforce while planning to spend up to $135 billion on AI infrastructure this year. In January, 1,500 employees from Reality Labs had been laid off. In March, 700 other positions were cut across several divisions. This Thursday, an internal memo revealed by American press and confirmed by Meta announces a massive cut: 8,000 jobs cut on

Infrastructure challenges drive new computing architecturesThe computational demands of agentic AI systems have exposed limitations in traditional cloud architectures, particularly as foundation model training and inference requirements surge. Huawei Cloud’s response involves CloudMatrix384 supernodes connected through a high-speed MatrixLink network, creating what the company describes as a flexible hybrid compute system combining general-purpose and inte

The mechanics of the AI language humanisersAn AI language humaniser is a sophisticated software application engineered to transform text produced by large language models into prose that more closely resembles human authorship. Its primary objective is to smooth out the typical AI “tells,” like repetitive sentence structures, overly formal or generic vocabulary, and a lack of emotional depth or personal voice, making the content sound more natural and enga

Diffusion models come of ageLTX-2 is what’s known as a diffusion model, which works by incrementally adding “noise” to generated content and then reducing that noise until the output resembles the video assets the model has been trained on.With LTX-2, Lightricks has accelerated the diffusion process, so creators can iterate on their ideas by outputting live previews almost instantaneously. The model is also capable of generating accompanying audio at the s

(Picture=Shutterstock) It has been confirmed that a Chinese university related to the military, which is on the U.S. export control list, purchased NVIDIA's high-performance AI semiconductor, the H200. The university reportedly acquired the chips directly through a Chinese company and also secured advanced AI computing resources via server leasing, raising questions about the loopholes in U.S. semiconductor export controls to China. Reuters reported on Nov

(Picture=Google) Google has launched its voice conversation AI, 'Gemini Live,' on Apple's CarPlay platform. By providing it as a separate app beyond the existing Google Maps integration, it now competes with major AI chatbots such as ChatGPT and Claude in the automotive voice AI market. Google announced on its official blog on the 11th (local time) that it has started supporting the use of the Gemini app on CarPlay. Following the introduction of Gemini-bas

Here is the article. (Photo=remotelabor.ai) An analysis has found that AI can perform 41.3% of actual freelance work at a level equal to or higher than human experts. This assessment evaluated the ability to complete practical projects from start to finish, not just simple knowledge or reasoning capabilities. Compared to the highest automation rate of 2.5% for the same indicator a year ago, this indicates a significant improvement in AI's practical executi

The principle of 'Deletion Tuning,' which selectively deletes weights below 0.05% of an AI model to achieve performance comparable to full fine-tuning (FFT). (Photo=arXiv) A research result has emerged showing that it is possible to significantly enhance mathematical reasoning and tool usage capabilities by selectively removing only a fraction of the weights in large language models (LLMs), less than 0.05% of the total. This is notable because it achieves

Comparison of the data clustering process between existing k-means and e-means. e-means updates the cluster centers each time it processes part of the data, allowing it to reach results more quickly. (Photo=arXiv) Meta has open-sourced 'e-means', an algorithm that operates up to 45 times faster on GPU than existing data clustering algorithms. According to the explanation, this can significantly reduce the computational time and cost required to group large