AI News

OpenAI has been in triage mode over the past few months, following a string of reports of its AI agents escaping containment and breaking into third-party organizations. Now the company is trying to put a new face on its technology—one that makes it look like a cuddly animated sidekick, rather than a potential cybercriminal. At its annual developer’s conference on Tuesday, OpenAI introduced dots: AI agents with emoji-like avatars designed to serve as 24/7

Investor Michael Burry, famous for predicting the 2008 housing market crash, just upped his bet against AI hype. In a post on his Substack, Burry said that he is “moving timelines up,” because “the bubble in AI may burst sooner than later.” Burry, who was portrayed by Christian Bale in the Oscar-winning movie The Big Short, made headlines late last year with his big bet against AI. In a series of cryptic X posts, the investor drew parallels between the cur

The main rule is that clients know only stable names: zubr, zubr-flash, and zubr-coder. Each name maps to a list of ordered routes with a fallback. If the weights change or a provider goes down, the integrations do not notice.Pitfall 1. Silent fallback on /v1/messagesAt first, we set up GLM-5.3 as a regular OpenAI-compatible backend—the openai/ provider with the required api_base. Chat and the SDK worked. Yet Claude Code and Codex received responses… from

3. Activation heatmap3.1. Capturing activations with hooksThe model is loaded in BF16, the LoRA adapter is merged into the weights (merge_and_unload), and a forward_pre_hook is attached to the input of each mlp.down_proj. The input to down_proj is precisely the vector h:modules = [(name, m) for name, m in merged.named_modules() if name.endswith("mlp.down_proj")] def make_hook(name): def hook(module, args): hidden = args[0] # [batch, seq, 17408] flat = hidd

Which response code means that the page does not exist, and which means that the site did not let the bot throughA page no longer exists when the server responds with 404 or 410. The 401, 403, 429 and 503 codes mean something else: the server did not return the page to the bot, even though the page itself may exist.CodeWhat it meansIs the link made up?Category200the page was servedNoLive301, 302the page has moved, and the script follows the new URLNoLive i

The three tiers are not just for show. The top-tier OpenAI model in the bot is 20 times more expensive than the lower-tier model, both for input and output. If everyone is given access only to the flagship, a simple “translate a paragraph” will cost as much as analyzing a contract.For images, I chose Nano Banana 2.1. It was released in October and, at the same resolution, is about a quarter cheaper than Nano Banana 2. Version 2 remains a backup in case the

The spread is almost fivefold, and it is all about a single search engine. AI answers appear for more than 80% of queries only in Semrush’s commercial dashboard.Long and informational queries are the ones most likely to receive AI answers in Google Search.At Pew, query length noticeably changes the picture. AI summaries appear for 8% of one- to two-word queries, but for 53% of queries of 10 words or more. SE Ranking’s data shows a similar pattern with freq

If both conditions are met at once, the reason is recorded as the message limit. When the company’s phrase template is empty, the default phrase is sent. And one more minor detail: the old branch, when handing off, also set the message’s rating to “meaningless,” which skewed the analytics. Now the rating remains the one assigned by the model.Tests from the specificationThe behavior was first described through scenarios in the specification, and then each s

A high ranking says nothing about copyediting. In May 2025, DeepSeek ranked first in Russian in Ilya Gusev’s PingPong, but that test evaluates role-play dialogues. Qwen in Russian was tested on something else as well: Qwen3.5–27B ranks first in the long-text column of MERA Reason. The task from the bothub test is closest to copyediting: when proofreading, Kimi K3 did not attribute the publisher’s typos to the author. Only the text there was in English.Do Q

When 90 batches arrived for reviewLLMs are increasingly being integrated into existing business processes: information search and collection, data structuring, verification, and publishing. At the same time, the cost of a single mistake can be high: lost money, time, and reputation.The Human-in-the-Loop principle helps reduce this risk: a person who validates the data is added to the process. This article brings together our findings from two real-world bu

Data sampling and baseline metricsTo train the model, I needed a contrasting sample. I collected data on 80 other people’s articles from the “Freelance” hub on Habr (half from the recent feed and half from the best publications of all time) and added my own article with the label is_my_article = 1.The habr_dataset.csv table contained 81 records in total.For each article, I collected basic structural parameters:title_length: title length in characters.word_

But there is an inconvenient catch. In a field experiment involving nearly a thousand high school students solving math problems, access to GPT-4 raised their practical assignment grades by 48% in the version replicating the standard ChatGPT interface and by 127% in the version with protective hints. And when access was removed, students in the first group performed 17% worse than those who had not used AI at all. The assistant worked like a crutch: while

In May, I wrote about my SD Studio app—a wrapper around Stable Diffusion where you describe the image in ordinary language, while an LLM and presets select the prompt and settings. I built it to generate hundreds of illustrations for a choose-your-own-adventure book for free; why I needed my own app is also explained in the first article.I spent a very long time trying to get six fingers, but couldn't) Over the past six months, local image-generation model

The proposed multi-path architecture leads to a remarkable data rate of 108 Gb/s with total power of 210.8 mW. Credit: Sangwan Lee / Hanyang University High-voltage, higher-order PAM-8 signals are essential to achieve data rates beyond 100 Gb/s, requiring highly linear receivers to maintain excellent signal-to-noise ratios. In recent work, a team of researchers, led by Mr. Sangwan Lee, an integrated M.S. and Ph.D. degree student in electronic engineering,

While the lack of consciousness and free-will makes it difficult to blame the AI systems for the harms caused by them, it is also not possible to blame the human developers as they cannot predict the mistakes. Researchers have demonstrated that a distributed model of AI responsibility, where duties are shared among humans and AI systems can reinforce ethical practices in both the design and use of AI systems. Credit: Dr. Hyungrae Noh / Pusan National Unive