# Пов'язані статті щодо Coding

Центр новин HTX надає останні статті та поглиблений аналіз на тему "Coding", що охоплює ринкові тренди, оновлення проєктів, технологічні розробки та регуляторну політику в криптоіндустрії.

Zhipu, Afraid of Becoming the Next MiniMax

Title: Zhipu, Fearing to Become the Next MiniMax In July 2026, amid the success of its coding-focused AI, Zhipu's founder, Tang Jie, issued an internal letter titled "The Giant Wave Has Come." It notably avoided celebrating recent triumphs, such as Zhipu's trillion-HKD market cap and booming MaaS revenue driven by its GLM-5.2 model in coding applications. Instead, the letter pivoted the narrative to future-oriented concepts like Long Horizon Task, Autonomous Agents, Self-Evolving systems, and AGI. This strategic shift in messaging followed the sharp devaluation of its competitor, MiniMax. After its lock-up period expired, MiniMax's stock plummeted as the market began evaluating it with traditional SaaS metrics like ARR and user growth, rather than as a frontier AI pioneer. Seeing this, Tang Jie aimed to preempt a similar revaluation of Zhipu. He fears that if the market starts viewing Zhipu primarily as a profitable "AI coding company," its valuation would become anchored to conventional financial metrics, losing the premium associated with AGI potential. Therefore, the letter reframed Zhipu's mission. While acknowledging that coding was the current commercial driver, Tang positioned Zhipu on the "infrastructure path," akin to OpenAI and Anthropic. The new focus is on developing agents capable of complex, long-term planning and autonomous operation—moving from assisting individuals (OPC: One Person Company) to automating entire organizations (NPC: No People Company). This "Touch High" plan explicitly prioritizes long-term AGI research over short-term monetization. The article frames this as a critical divergence in China's AI landscape: the "commercialization path" (exemplified by MiniMax) versus the "infrastructure path" (chosen by Zhipu). The former risks being judged harshly by internet-era metrics once growth slows, while the latter risks failing if technological breakthroughs stall. Tang Jie's letter is thus a calculated move to secure Zhipu's identity as an AGI contender, buying time before the inevitable market demand for commercial proof. The core question remains: can Zhipu's "mo gao" (reach high) plan achieve genuine technological leaps fast enough to outpace the market's diminishing patience for stories over substance?

marsbit2 год тому

Zhipu, Afraid of Becoming the Next MiniMax

marsbit2 год тому

Breaking News: Musk Delivers the Most Powerful Grok 4.5, Slashes Price of Top-tier Opus Intelligence Drastically

**Elon Musk Launches Grok 4.5: A Cost-Effective, High-Performance AI Rival** SpaceXAI, in collaboration with Cursor, has released Grok 4.5, its new flagship AI model designed specifically for coding and agentic tasks. Trained on tens of thousands of NVIDIA GB300 GPUs using massive, high-quality data filtered from trillions of Cursor developer interactions, the model emphasizes "per-token intelligence." In benchmark performance, Grok 4.5 is highly competitive. It scores 64.7% on SWE Bench Pro (surpassing GPT-5.5's 58.6% and Opus 4.7's 64.3%), 83.3% on Terminal Bench 2.1 (nearly matching GPT-5.5), and 62.0% on DeepSWE 1.0 (beating Opus 4.8). Overall, it ranks fourth in AAAI official tests and first in the Harvey legal agent benchmark. The model's key advantage is its combination of speed, efficiency, and low cost. It generates responses at 80 tokens per second and, crucially, uses far fewer tokens to complete tasks—4.2 times fewer than Opus 4.8 on SWE Bench Pro. It is priced at $2 per million input tokens and $6 per million output tokens, significantly undercutting competitors. Musk stated it is "roughly equivalent to Opus 4.7, but much faster." Early user tests show Grok 4.5 can generate functional code for applications like 3D solar system simulators and basic games from simple prompts, though some note it still lags behind top models in certain creative tasks. Musk has hinted at a major update next month, leveraging real-world engineering data from his companies, with an even larger 2-trillion parameter version reportedly in development. Grok 4.5 positions itself not as the absolute strongest model, but as a highly efficient and affordable alternative in the top tier.

marsbit07/09 03:11

Breaking News: Musk Delivers the Most Powerful Grok 4.5, Slashes Price of Top-tier Opus Intelligence Drastically

marsbit07/09 03:11

GPT-5.5 Suffers Secret Intelligence Downgrade, Crashes at Thought Token 516, Stumbles Harder on Tougher Tasks

The article reports a concerning anomaly discovered in OpenAI's GPT-5.5 model. Developers analyzing backend metadata found a disproportionate number of its reasoning processes abruptly terminating precisely at 516 tokens, with similar spikes at multiples like 1034. Statistical analysis revealed that while GPT-5.5 handled only 19.3% of total responses, it was responsible for over 80% of these "exact 516" stoppages. This pattern is drastically more pronounced in GPT-5.5 compared to other models, suggesting a possible hidden "reasoning budget" cap or truncation mechanism. Concurrently, users report the model has become overly formulaic, persistently using bullet points, offering unsolicited corrections, and providing excessive options, making interactions feel less helpful. The core concern is that GPT-5.5 may be prioritizing structured, truncated responses over genuine, in-depth problem-solving, especially for complex tasks.

marsbit07/05 23:58

GPT-5.5 Suffers Secret Intelligence Downgrade, Crashes at Thought Token 516, Stumbles Harder on Tougher Tasks

marsbit07/05 23:58

2028: The Arrival of Recursive Self-Improvement (RSI)

**AI Recursive Self-Improvement (RSI): The Countdown to 2028 Begins** AI is no longer just a trained tool but is starting to rewrite its own evolutionary pace. According to Anthropic co-founder Jack Clark, there is a 60% probability that by the end of 2028, Recursive Self-Improvement (RSI) will become a reality. This means AI could autonomously design and build a more capable next-generation version of itself without any human researcher involvement—Claude 10 creating Claude 11, for instance. Supporting this timeline, Google DeepMind's CEO Demis Hassabis confirms that all leading AI labs are intensely focused on RSI, making it an industry-wide priority. He expresses profound concern, stating this potential is what keeps him awake at night. Concrete data underscores this acceleration: - METR evaluations show current top models like Claude are solving tasks up to the 16-hour limit of existing test frameworks. - In Epoch AI's challenging MirrorCode benchmark, Claude Opus 4.7 recreated complex software in hours for a fraction of the human cost. In one extreme test, AI autonomously coded for 19 days straight. - Anthropic reports over 80% of its codebase is now written by Claude, and researcher productivity has increased up to 8-fold since 2024. - OpenAI's policy blueprint highlights RSI as a major upcoming governance challenge. CEO Sam Altman reportedly hinted RSI might arrive within six months, potentially delaying OpenAI's massive IPO. The implication is an impending "intelligence explosion," where AI-driven progress outpaces human control. The central question is no longer if it will happen, but whether humanity is ready.

marsbit06/28 10:45

2028: The Arrival of Recursive Self-Improvement (RSI)

marsbit06/28 10:45

I Only Trust yuxinlu1 in the Hugging Face Model TOP Rankings Now

An independent developer has unexpectedly risen to the top of Hugging Face's Trending models chart, surpassing major tech companies. The developer, Yuxin Lu (yuxinlu1), has two 12B GGUF models (Coder and Agentic versions) based on Gemma-4-12B on the list, with downloads exceeding 200k and 500k respectively. His models distill coding and reasoning capabilities from sources like Fable 5 into a locally runnable format, requiring as little as 4.5GB of VRAM. Lu, a graduate student in AI, developed these models as a personal, self-funded project for skill development. He emphasizes data quality over quantity, using around 10,000 verified examples. The models' popularity stems from offering privacy and a free, offline alternative for coding assistance and agentic tasks, filling a niche not prioritized by larger companies focused on broader goals. Lu's other projects include general-purpose distillation models and Chinese web novel LoRAs. He advises fellow developers to prioritize honesty about model capabilities and persistence. He views his success not as defeating major players, but as focusing deeply on solving specific user needs with genuine effort. The models are best run using llama.cpp.

marsbit06/28 01:52

I Only Trust yuxinlu1 in the Hugging Face Model TOP Rankings Now

marsbit06/28 01:52

Google's 'Reasoning King' Also Departs for Meta, Originally Recruited by Fei-Fei Li

"Google's 'King of Reasoning' Leaves for Meta, Quietly Departing After Over Eight Years. Denny Zhou, a key figure behind Google's AI reasoning advancements including work showcased by CEO Sundar Pichai, has joined Meta's MSL as a research scientist. His low-profile move, discovered via a LinkedIn update, occurred months before the high-profile departures of Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic. Zhou was originally recruited to Google by Fei-Fei Li's China center initiative after nearly 11 years at Microsoft. This is part of a significant talent drain at Google, with top researchers like Shazeer (co-author of the Transformer paper) and Jumper (AlphaFold lead) recently leaving for rivals. Reports suggest internal friction is a contributing factor, particularly around Google's strategic shift. The company has reportedly formed a high-priority 'AI Coding Strike Team,' involving co-founder Sergey Brin, to urgently bridge the gap in AI coding agents, potentially reallocating resources and focus away from other research directions like DeepMind's 'world model' AGI approach. This pivot towards commercially-proven coding applications may have influenced departures, as hinted by Shazeer's comment about his compute allocation being given to another team. Meanwhile, Meta continues to bolster its team, also recently hiring UC Berkeley professor and 'security godmother' Dawn Song, along with her startup Virtue AI team, as a VP of AI research."

marsbit06/26 13:39

Google's 'Reasoning King' Also Departs for Meta, Originally Recruited by Fei-Fei Li

marsbit06/26 13:39

The Year of AI Applications: Saying 'Yes' While Ignoring Risks? A Comprehensive Open Source Log of Software Development's Journey

The Year of AI Applications: Blindly Saying "Yes" While Ignoring Risks? A Software Development Log Goes Fully Open Source. AI-generated code harbors risks hidden within seemingly correct programs, potentially leading to data leaks or asset loss. The open-source project "Narwhal AI Code Risks," from Peking University's Narwhal-Lab, compiles real-world cases, early warning signs, and typical risk pathways. Its goal is to help developers identify potential hazards early and avoid repeating past mistakes. In 2026, code is generated faster than ever but deployed with less scrutiny. The danger often lies not in glaring errors, but in code that appears normal—syntactically correct, passing all checks—yet introduces subtle but critical flaws like non-existent dependencies, excessive permissions, or exposed databases. A stark example is the Moonwell cbETH oracle incident. A configuration file error, where a cryptocurrency price was set to ~$1.12 instead of ~$2,200, slipped through 28 checks and a pull request signed by both AI (Claude, Copilot) and human developers. This "semantic deviation" resulted in a loss of $1.78 million. The risk is that AI can produce functionally valid code that is semantically wrong for the business context. As AI moves beyond simple code completion to modifying configurations, installing dependencies, and operating via autonomous agents, it traverses longer, less traceable paths within software engineering, blurring traditional boundaries and oversight points. The Narwhal AI Code Risks project structures information into three layers: `/cases` for documented real-world incidents, `/inferred` for early warning signals, and `/scenarios` for clear, generalized risk patterns not yet tied to specific events. This aims to create a lasting, public record to prevent collective amnesia about past AI-coding pitfalls. Risks are categorized into seven areas: Software Supply Chain (e.g., recommending fake packages), Code-Level Vulnerabilities (e.g., reintroducing path traversal bugs), Cloud & Infrastructure Misconfiguration (e.g., overly permissive settings), Agent Risks (from autonomous tool execution), Vertical Domain Risks (e.g., in finance, healthcare), Intellectual Property & Compliance issues, and Human Factors (like over-reliance on AI output). The project's core value is transforming isolated incidents into reusable knowledge—a foundational resource for developers to spot similar issues, for security researchers to build upon, for toolmakers to create detection rules, and for the community to contribute new findings. As AI integration accelerates, this open-source "logbook" serves as a crucial navigational aid, charting past errors to help future projects steer clear of the same traps.

marsbit06/16 04:52

The Year of AI Applications: Saying 'Yes' While Ignoring Risks? A Comprehensive Open Source Log of Software Development's Journey

marsbit06/16 04:52

Is Writing Prompts Outdated? AI Programming is Shifting to Loop Engineering

"The era of manually writing prompts for AI coding agents is shifting towards 'Loop Engineering,' where developers design automated, self-managing workflows instead of step-by-step guidance. A loop consists of five core components: Automations (scheduled task discovery and triage), Worktrees (isolated environments to prevent file conflicts), Skills (project-specific knowledge and conventions), Plugins/Connectors (integration with tools like GitHub, Linear, or Slack), and Sub-agents (separate agents for execution and verification). An external memory layer, such as Markdown files or Linear boards, tracks progress across sessions. Tools like Claude Code and Codex now embed these components, enabling loops to run autonomously—discovering tasks, assigning work, checking results, and deciding next steps. This approach amplifies developer leverage but doesn’t replace human judgment. Key risks include 'comprehension debt' (losing understanding of the codebase) and 'cognitive surrender' (over-relying on automation). Effective loop design requires balancing automation with ongoing verification and deep system understanding, making it a more complex but higher-leverage skill than prompt engineering."

marsbit06/10 17:55

Is Writing Prompts Outdated? AI Programming is Shifting to Loop Engineering

marsbit06/10 17:55

Claude Code Introduces Dynamic Workflows: Enabling AI to Form Teams and Collaborate

Claude Code introduces dynamic workflows, enabling AI to coordinate teams of specialized agents for complex tasks. This transforms Claude from a code assistant into a programmable workbench. Workflows address key limitations of single-agent systems: agentic laziness (premature task completion), self-preferential bias (favoring own outputs), and goal drift (losing sight of original objectives). The system allows Claude to dynamically create execution frameworks using JavaScript. It can split tasks, dispatch parallel agents for isolated work (e.g., in separate worktrees), implement adversarial validation, run tournaments, and synthesize results. This multi-agent approach is valuable for tasks requiring deep research, factual verification, code migration, root cause analysis, large-scale triage, and qualitative sorting. Key patterns include: classify-and-route, fan-out-and-synthesize, adversarial verification, generate-and-filter, tournaments, and loop-until-done. While token usage is higher, workflows excel where tasks resemble programming—needing problem decomposition, isolated context, hypothesis testing, and handling many details. They extend Claude Code's utility beyond technical work to areas like business plan review, resume screening, and naming brainstorm. The feature is not a universal solution but points to a future where AI tool competitiveness depends on organizing reliable, reusable, and auditable execution flows for complex goals.

marsbit06/04 02:15

Claude Code Introduces Dynamic Workflows: Enabling AI to Form Teams and Collaborate

marsbit06/04 02:15

Google CEO Admits Lagging Behind in Coding

Google CEO Sundar Pichai acknowledged in a recent interview that Google's Gemini AI models are currently "lagging behind" in coding capabilities, particularly for complex, long-horizon tasks requiring advanced developer expertise. He noted the field is advancing at an "unprecedented" pace, where 30-60 days now brings changes equivalent to five years in the past. Pichai expressed that achieving Artificial General Intelligence (AGI) now seems closer than previously imagined due to rapid progress. While highlighting strengths in text, multimodal, and reasoning tasks, Pichai admitted competitors like Anthropic and OpenAI have focused more intently on coding. He emphasized Google's commitment to catching up, citing internal tools like Antigravity 2.0 and the newly released Gemini 3.5 Flash, which aims to address previous shortcomings. Regarding Google Search's AI-driven overhaul, Pichai stated changes will be gradual to align with user needs, not disrupt the core search experience or its advertising model. He addressed public AI anxiety as understandable, given the technology's potential to reshape jobs and society, but remained optimistic about AI augmenting human capabilities and creating new opportunities. Pichai stressed the need for broad societal dialogue and responsible development as AI approaches more advanced, potentially recursive self-improvement stages. He affirmed Google's long-term commitment to leading in AI while navigating its profound implications responsibly.

marsbit05/24 08:28

Google CEO Admits Lagging Behind in Coding

marsbit05/24 08:28

活动图片

Regulatory Policyright-arrow

Ethereumright-arrow