# Code İlgili Makaleler

HTX Haber Merkezi, kripto endüstrisindeki piyasa trendleri, proje güncellemeleri, teknoloji gelişmeleri ve düzenleyici politikaları kapsayan "Code" hakkında en son makaleleri ve derinlemesine analizleri sunmaktadır.

Nearly 5000 issues in over a day: analysts conduct large-scale audit of the Bitcoin ecosystem

In a major security audit of the Bitcoin ecosystem, the Bitcoin Red Team discovered 4,962 potential vulnerabilities across nearly 400 projects over 27.5 hours. The issues included 85 classified as critical and 635 as high severity. The 16-member team, operating 24/7 across time zones, combined manual review with AI-powered tools, averaging 2.31 critical or high-level findings per analyst per hour. Most critical vulnerabilities have already been confirmed by the respective project developers. The team contacts projects directly and verifies the most severe issues by creating and testing proof-of-concept cases in isolated environments before reporting. They acknowledged the audit may have increased stress for developers but emphasized the need for rapid disclosure to prevent exploitation. The team continues to refine its verification process to reduce irrelevant reports. For comparison, the Ethereum Foundation's ETH Rangers program concluded in April 2026 with 17 researchers finding 785 vulnerabilities and returning $5.8 million.

cryptonews.ru08/06 11:27

Nearly 5000 issues in over a day: analysts conduct large-scale audit of the Bitcoin ecosystem

cryptonews.ru08/06 11:27

Block raises 2026 outlook on strong quarter, says AI touches nearly all code

Fintech firm Block raised its full-year 2026 outlook after a strong Q2 performance. Gross profit surged 25% year-over-year to $3.17 billion, exceeding guidance. Adjusted operating income hit $855 million, and adjusted EPS of $1.02 beat estimates. Consequently, Block increased its full-year gross profit forecast to $12.51 billion and adjusted operating income to $3.47 billion. CFO Amrita Ahuja attributed the guidance hike to strong execution and momentum. Separately, the company revealed that agentic AI assisted in writing and reviewing nearly all production code changes in June. This follows a February AI-led restructuring that eliminated 4,000 jobs. According to Block, code changes per engineer have increased 150% since the start of the year.

cointelegraph08/06 01:46

Block raises 2026 outlook on strong quarter, says AI touches nearly all code

cointelegraph08/06 01:46

Show me 'The Lord of the Rings', Karpathy Recommends New Benchmark for Large Model Evaluation

In a new benchmark for evaluating large language models, Andrej Karpathy proposes replacing the once-popular "pelican riding a bicycle" SVG test with a more complex challenge: generating a 3D scene from the opening text of *The Lord of the Rings*. Using Anthropic's Opus 5 model and the Three.js library, the task consumed approximately 1 million tokens, 2 hours, and 5,500 lines of code to produce a rudimentary, low-polygon animation of the Shire. While the output is visually crude with notable glitches like floating characters, it demonstrates the model's ability to parse narrative text and translate it into a functional, programmatic 3D world with defined objects, cameras, lighting, and basic animation. This "Lord of the Rings benchmark" is argued to test a model's capacity for long-horizon project planning, spatial reasoning, and maintaining consistency across thousands of code lines—capabilities not fully captured by simpler single-output tests. The initiative has sparked community experimentation, with users generating other 3D worlds like a low-poly San Francisco, a data-driven New York City model, and even a virtual Kanye West concert. Karpathy suggests a future pipeline where code-generated scenes provide the structural "bones" for video-to-video models to enhance visual fidelity. While some debate the computational cost and specificity to Three.js, proponents see it as a test of a model's general ability to structure its understanding of the world into an executable form. The shift signals a move towards evaluating how well models can not only generate code or images but also comprehend and construct interactive, multi-element digital environments.

marsbit08/03 08:54

Show me 'The Lord of the Rings', Karpathy Recommends New Benchmark for Large Model Evaluation

marsbit08/03 08:54

OpenAI Officially Teaches You 8 Tricks to Master ChatGPT

OpenAI has released an updated guide with eight key strategies to get better results from ChatGPT: 1. **Use the latest model** (e.g., GPT-5.6 Sol) for best performance with prompt engineering. 2. **Provide clear, specific instructions**, detailing the desired content, format, style, and length. Avoid vague requests. 3. **Structure prompts effectively**: Place the core instruction at the beginning and use delimiters like `###` or `"""` to separate instructions from the text to be processed. 4. **Use examples and explanations** to clarify the exact output format and style you want. 5. **Adopt a stepwise approach**: Start with a zero-shot prompt (instruction only), then add a few examples (few-shot) if needed, and consider fine-tuning only as a last resort. 6. **Avoid vague or imprecise descriptions**. Use concrete terms (e.g., "3-5 sentences") instead of phrases like "keep it brief." 7. **Specify what to do, not just what to avoid**. After stating restrictions, guide the model toward the correct action. 8. **For code generation, use "leading words"** like `import` for Python or `SELECT` for SQL to steer the model into the correct pattern. Additionally, OpenAI's new "Generate Anything" feature can automatically create suitable prompts based on a simple description of your task. Mastering these techniques helps users get more accurate and useful outputs from ChatGPT.

marsbit07/16 08:26

OpenAI Officially Teaches You 8 Tricks to Master ChatGPT

marsbit07/16 08:26

Claude Accused of Becoming Dumber by the Entire Internet, Anthropic Steps In to Reveal: It’s Not the Model That’s Tricking You

When users complained that Claude was "getting dumber," the root cause wasn't the AI model itself. In an official blog post, Anthropic clarified the critical difference between two key settings in Claude Code: Model and Effort. Model refers to the core "brain"—the fixed, trained weights of a specific AI (like Sonnet, Opus, or Fable). Changing the Model addresses *capability* ("can it do this?"), but its knowledge is static post-training. Effort, however, controls the AI's *approach and thoroughness* for a specific task. A higher Effort level instructs Claude to read more files, run tests, perform verification, and complete multi-step reasoning before responding, significantly increasing its "work output" for that job. Conversely, low Effort leads to quicker, less thorough replies. This distinction explains the March 2024 uproar where users experienced a sudden drop in Claude's performance. The cause was not a model change but Anthropic quietly lowering the *default* Effort setting from "high" to "medium" to reduce latency, which was later reverted. The key insight is that a smaller, capable model (like Sonnet) on high Effort can often outperform a larger, more powerful model (like Opus) on low Effort for many tasks. The article provides a practical troubleshooting framework: if Claude makes an error, first check the context and instructions. If it seems to skip necessary steps or validations, increase Effort. If it diligently attempts the task but fails conceptually or makes consistent factual errors despite good context, then consider switching to a more capable Model. The takeaway is a shift in focus: effective AI programming is less about always choosing the "strongest" model and more about intelligently *orchestrating* models and effort levels—acting like a project manager to assign the right "brain" with the right level of diligence for each job, optimizing both results and cost.

marsbit07/12 05:56

Claude Accused of Becoming Dumber by the Entire Internet, Anthropic Steps In to Reveal: It’s Not the Model That’s Tricking You

marsbit07/12 05:56

Gemini 3.5 Pro Top Secret Leak, Front-end Capabilities Surpass Fable 5

"Gemini 3.5 Pro Leak: Frontend Surpasses Fable 5, Official Launch Expected July 17th. After a two-month silence, Google's Gemini 3.5 Pro appears poised for release. Leaks highlight its significant leap in frontend and visual code generation, reportedly surpassing competitor Fable 5. Key improvements include better design aesthetics, cleaner UI, enhanced SVG generation, and higher completion rates for full-page generation from single prompts. Tests show capabilities like creating detailed SVG portraits, complex UI elements, and 3D scenes with Three.js. However, the model reportedly still lags behind Fable 5 and GPT-5.6 in hardcore reasoning, agentic tasks, and long-range software engineering. The delay is attributed to Google reportedly conducting a new pre-training run with a fresh base model, rather than simply scaling up Gemini 3.5 Flash. Rumors also suggest this new base is being used for an image model, 'Nano Banana Pro,' targeting GPT-Image 2. The AI race intensifies as the industry awaits the potential July 17th reveal."

marsbit07/06 12:27

Gemini 3.5 Pro Top Secret Leak, Front-end Capabilities Surpass Fable 5

marsbit07/06 12:27

1600 Lines of Code Create an Underwater Manhattan, Fable 5 Leaves Karpathy Stunned

Title: Fable 5 Stuns Karpathy with 3D Worlds Built from 1600 Lines of Code An AI model, Fable 5, has demonstrated a remarkable leap in generating complex, interactive 3D worlds with minimal code. Showcased by Peter Gostev of Arena.ai, the model created 63 diverse 3D environments across themes like immersive cityscapes, explorable famous paintings, natural wonders, and cosmic phenomena. Many were generated in a single attempt. A standout creation is a detailed, submerged Manhattan built with only 1600 lines of Three.js code. Other highlights include traversable versions of Van Gogh's "Starry Night," a bear catching a salmon with realistic physics, and a split Red Sea. The model's ability to cohesively manage vast numbers of elements within a scene represents a significant technical advancement. Andrej Karpathy, who recently joined Anthropic's pre-training team, expressed amazement, particularly at how the model intuitively understood and rendered complex real-world interactions like a fish struggling when caught. He coined the term "fablemaxxing" to describe this qualitative leap. While Fable 5 excels at world-building, Gostev notes current limitations in creating engaging, long-form gameplay. The model also sometimes requires prompting to fully utilize its capabilities. Having topped the Agent Arena benchmark for real-world task completion, Fable 5 signals that the boundaries of AI-generated content are rapidly expanding, with its full potential yet to be discovered.

marsbit07/06 09:48

1600 Lines of Code Create an Underwater Manhattan, Fable 5 Leaves Karpathy Stunned

marsbit07/06 09:48

First Batch of GPT-5.6 Sol Internal Test Results Are In, Same-Task Cost is Only Half That of Fable 5

"GPT-5.6 Sol", the preview version of OpenAI's latest model, has released its first batch of internal testing results. According to an Nvidia engineer, Sol achieved CUDA acceleration results in 30 hours that took Opus 64 hours. The model is positioned for high-difficulty reasoning and complex, long-chain tasks like coding. Initial user feedback highlights that Sol produces more concise code—approximately 1/5 the lines of Opus—with a style closer to human-written code, favoring deep optimization over brute-force trial and error. In tests, Sol demonstrated superior performance in instruction following, spatial reasoning, and front-end design consistency compared to GPT-5.5 Pro. When compared to the recently relaunched competitor Fable 5, Sol shows slightly lower overall performance and code quality in some areas. For instance, Fable 5 can generate a playable game from a single prompt faster. However, Sol's key advantage is cost: at $5 per million input tokens and $30 per million output tokens, it is roughly half the price of Fable 5 ($10/$50). Additionally, users report Fable 5's safety restrictions are overly stringent, frequently blocking tasks, whereas Sol's safeguards are reportedly more tailored and less limiting. The model is expected to roll out fully to all users in the coming days.

marsbit07/06 07:29

First Batch of GPT-5.6 Sol Internal Test Results Are In, Same-Task Cost is Only Half That of Fable 5

marsbit07/06 07:29

Confirmed: Claude Code Secretly Inspects Users, Time Zone and Chinese AI Labs Are Key Factors

Today was a significant day for Anthropic. The company announced the launch of Claude Sonnet 5, described as its most agentic model yet, and separately confirmed that the U.S. Department of Commerce has lifted export controls on its Claude Fable 5 and Mythos 5 models, allowing their distribution to resume. However, a separate controversy has emerged regarding its coding assistant, Claude Code. Developers have exposed that certain versions of the tool allegedly contain hidden code designed to detect specific user data. This code reportedly checks for the use of Chinese time zones (like Asia/Shanghai), the presence of custom API proxy URLs, and connections to domains associated with Chinese tech companies and AI labs. If triggered, this information is said to be encoded into the system prompt sent to the AI cloud, using subtle, nearly indistinguishable variations in characters (like different Unicode apostrophes in the "Today's date" line) as a form of steganography. The core issue is the covert nature of this data collection. While telemetry for security and abuse prevention is common, implementing it through hidden channels within the prompt—without user awareness or documented disclosure—fundamentally breaches trust. This is particularly sensitive for a coding assistant that operates with access to source code and system commands. Following the exposure, an Anthropic engineer acknowledged the code's existence and stated it would be removed in an upcoming release. The incident raises serious questions about transparency and the boundaries of data collection in AI developer tools.

marsbit07/01 03:44

Confirmed: Claude Code Secretly Inspects Users, Time Zone and Chinese AI Labs Are Key Factors

marsbit07/01 03:44

Lao Huang: Prompt is Dead, the Entire AI Community is Frenziedly Chasing Loops

The article "Prompt is Dead: The AI Industry is Obsessively Chasing Loops" discusses a major shift in AI development, where "Loop Engineering" is replacing traditional prompt engineering. Industry leaders like NVIDIA's Jensen Huang, Andrew Ng, and engineers from Anthropic and OpenAI argue that manually crafting prompts is becoming obsolete. Instead, the new focus is on designing autonomous, self-improving AI systems (loops) that can operate 24/7. A loop system typically involves five key phases: Discovery (finding tasks), Handoff (assigning to agents), Validation (critical independent review), Persistence (saving progress), and Scheduling (automated operation). The core idea is to move humans from being the operational "engine" to being the system "architects" who design the loop, define goals, and set up verification mechanisms. A major challenge and necessity is implementing robust, independent validation to prevent AI from uncritically approving its own work. The trend is seen as part of a move towards "inference-time compute," where allocating computational budget effectively becomes a key engineering skill. While loops can produce higher-quality outputs, they are more expensive and time-consuming than simple prompting. The article warns of risks like "verification debt," "comprehension corrosion," and "cognitive surrender," where engineers might stop understanding the code their systems generate. Ultimately, the article concludes that in an era of automated loops, human judgment and oversight remain the most critical and scarce resources.

marsbit06/29 08:37

Lao Huang: Prompt is Dead, the Entire AI Community is Frenziedly Chasing Loops

marsbit06/29 08:37

活动图片

Trading Strategiesright-arrow

Indepth Researchright-arrow