All news

AAlibaba Cloud Developer Community AI22:32

AI score59

Write a Test Skill in 30 Minutes: Turning Wishful Prompts into SOPs That AI Will Follow

Using test case generation as an example, the article explains how to rewrite wishful prompts like "please help me generate test cases" into a SKILL.md that Claude Code can follow. The author provides a four-step skeleton: understand the requirements; identify four types of scenarios—normal flows and exception scenarios, boundary conditions, and permission checks; generate test cases according to fixed fields; and check against a self-review checklist. The article also explains that description should clearly state the trigger words, and when_to_use should indicate applicable and non-applicable scenarios. The comparison in the article shows that before the rewrite, AI output only three plain-text test cases; after the rewrite, it output twelve table-form test cases covering scenarios such as unauthorized access, and the author recommends running through manually written test cases to verify format and coverage.

#Tutorials / practice#Agents#English text is being prepared.

AAyi AI Notes22:31

AI score52

Ayi shares the core system prompt and development constraints of the "Emotional Intelligence Plugin"

Ayi has publicly released the core system prompt of the "Emotional Intelligence Plugin" message rewriting tool, which rewrites WeChat, Feishu, and DingTalk messages based on the four elements of Marshall Rosenberg's Nonviolent Communication: observation, feeling, need, and request. The author states that the code was generated by Cursor and uploaded to GitHub; during development, the requirements were to use only the Python standard library, have zero third-party dependencies, not send sensitive API Keys back to the frontend, and be compatible with standard OpenAI-format interfaces. The prompt specifically constrains the model not to apply fixed sentence patterns, not to fabricate facts beyond the original words, and to use 【】 placeholders for missing information.

Quoted postAYi · @AYi_AInotes

https://t.co/I431zBBUB3

#Tutorials / practice#English text is being prepared.#GitHub

336Kr · AI22:27Analysis

AI score62

Claude Code team member on how to reduce over-prompting Claude

Addy Osmani, a member of the Anthropic Claude Code team, wrote in a post that users generally over-prompt Claude, and suggested using a single prompt to clearly state the goal and acceptance criteria, then letting the model execute on its own, with reasoning effort starting from /effort medium. He mentioned that Claude Code once removed more than 80% of the system prompts for models such as Claude Opus 5 and Claude Fable 5, with no measurable decline in coding benchmark performance, and suggested users use /doctor prompt-audit to find prompts written for older models. The article also suggests first making a single session capable of completing tasks independently before considering parallel multi-session work, and focusing attention on verifying results.

Same story · 2 reports

#English text is being prepared.#Agents#English text is being prepared.

PPhoenix Technology16:03Analysis

AI score63

Anthropic Discloses Claude Self-Evaluation Mechanism: A Hidden Safety Review Appended After Every Response

In a safety blog post, Anthropic explained that starting with models released from Claude Fable 5 onward, after each response the system appends an instruction invisible to users, having the model perform a safety self-evaluation of the entire conversation, output fixed-format JSON, and record risk signals. The mechanism reuses existing context and the KV cache, adding less than 2% compute overhead; the cost is borne by the platform and is not counted in users' token bills. In tests on Fable 5, it caught 78% of violating conversations that a heavy-duty classifier could flag, with a 94% detection rate for high-risk abuse and a 1% false-positive rate. Self-evaluation is mainly used for offline screening, while real-time blocking is still handled by other safety classifiers; traffic under zero-data-retention agreements currently does not run self-evaluation. The article also mentions that Fable 5.5 is being tested, and some users say Opus 5.5 will be routed to a new model with better performance.

#Safety / alignment#Anthropic#English text is being prepared.

PPhoenix Technology16:03Analysis

AI score61

Anthropic Discloses Claude Self-Evaluation Mechanism: A Hidden Safety Review Appended After Each Response

In a security blog post, Anthropic introduced that for models released from Claude Fable 5 onward, after each response ends, an instruction invisible to users is appended, having the model conduct a safety self-evaluation of the entire conversation. This mechanism reuses existing context and the KV cache, adding less than 2% compute overhead; on Fable 5 it captured 78% of violating conversations that a heavy-duty classifier could flag, with a detection rate of 94% for high-risk abuse and a false positive rate of 1%. The evaluation results are not shown to users and do not enter subsequent conversations; they are mainly used for offline aggregate screening, while real-time interception remains the responsibility of other safety classifiers.

Same story · 4 reports

#Safety / alignment#Anthropic#English text is being prepared.

AARC Prize12:53First-partyAnalysis

AI score62

ARC Prize publishes Meta Muse Spark Open's evaluation results on ARC-AGI

ARC Prize published the validation results of Meta's Muse Spark Open on ARC-AGI: ARC-AGI-1 at 89.0% with $0.29 per task, ARC-AGI-2 at 57.5% with $0.73 per task, and ARC-AGI-3 at 0.1% under the standard harness with $3.9K. ARC Prize states that its v1/2 scores are close to those of Haiku 5.5 and GPT-6 Luna, but the cost is about 9 times that of the latter; the model is priced on OpenRouter at $1.25 per million input tokens and $4.25 per million output tokens. As the reasoning tier increases, ARC-AGI-2 rises from 29.4% at medium (averaging 55k tokens/task) to 57.2% at xhigh (180k tokens/task).

Same story · 4 reports

#Evaluations / benchmarks#Reasoning#Meta

AARC Prize12:53First-partyAnalysis

AI score60

ARC Prize tests Meta Spark Open: ARC-AGI-2 score 57.5%, cost about 9 times that of models with the same score

ARC Prize tested Meta Spark Open on ARC-AGI: it scored 89.0% on ARC-AGI-1, 57.5% on ARC-AGI-2, and only 0.4% on ARC-AGI-3. Its v1/v2 scores are close to those of Haiku 5.5 and GPT-6 Luna, but its cost is about 9 times theirs. Across providers on OpenRouter, the model costs $1.25/M for input and $4.25/M for output, while Haiku 5.5 and GPT-6 Luna both cost $0.10/M for input and $0.50/M for output. As the reasoning tier increases, ARC-AGI-2 rises from 29.4% at medium (averaging 55k tokens/task) to 57.2% at xhigh (180k tokens/task).

Same story · 4 reports

#Evaluations / benchmarks#Reasoning#OpenAI

AARC Prize12:53First-partyAnalysis

AI score53

ARC Prize publishes Muse Spark Open reasoning scaling test results on ARC-AGI

ARC Prize published the test results of Meta's Muse Spark Open on the public ARC-AGI tasks: ARC-AGI-2 improves with reasoning scale, with medium averaging 55k tokens per task for 29.4%, high at 150k tokens for 51.0%, and xhigh at 180k tokens for 57.2%. ARC-AGI-1 already reached 92.8% at high and rose to 94.0% at xhigh. The model scores 57.5% on ARC-AGI-2 at $0.73 per task, and 89.0% on ARC-AGI-1 at $0.29 per task; its scores are close to those of Haiku 5.5 and GPT-6 Luna, but its cost is about 9 times theirs.

Same story · 4 reports

#Evaluations / benchmarks#Reasoning#Meta

AARC Prize12:53First-partyAnalysis

AI score52

ARC Prize announces Meta Muse Spark Open's evaluation results on various versions of ARC-AGI

ARC Prize announced the validation results of Meta Muse Spark Open on ARC-AGI: ARC-AGI-1 is 89.0% at $0.29 per task, ARC-AGI-2 is 57.5% at $0.73 per task, and ARC-AGI-3 is 0.4% at $3.0K under the provider adapter harness and 0.1% at $3.9K under the standard harness. ARC Prize states that its scores on v1/v2 are similar to those of Haiku 5.5 and GPT-6 Luna, but the cost is about 9 times that of the latter; the model is priced on OpenRouter at $1.25 per million input tokens and $4.25 per million output tokens, while Haiku 5.5 and GPT-6 Luna are both $0.10 per million input tokens and $0.50 per million output tokens. On public tasks, as the reasoning scale increases, the ARC-AGI-2 score rises from an average of 55k tokens/task corresponding to 29.4% at the medium tier, to 150k corresponding to 51.0% at the high tier, and then to 180k corresponding to 57.2% at the xhigh tier; ARC-AGI-1 has already reached 92.8% at the high tier and rises to 94.0% at the xhigh tier.

Same story · 4 reports

#Evaluations / benchmarks#Reasoning#Open-source ecosystem

MMa Dongxi NLP12:53Analysis

AI score58

Reinforcement Learning Terminology for LLM Readers: How a Single Score Is Distributed to Each Token

The article uses DeepSeek-R1-Zero's pass@1 on AIME 2024 rising from 15.6% to 71.0% as an example to show that a rule-based judge gives only a single score for the entire response, whereas training needs to adjust the probability of each token. The author uses the sampling example "Is OpenAI open?" to explain in turn policy, state, action, reward, return, value, Q, advantage, baseline, TD error, GAE, clipping, and KL, and points out that PPO uses a critic to learn V, while GRPO uses the average score within a group to replace V.

#Tutorials / practice#Reasoning#English text is being prepared.

SShao Meng12:53

AI score58

poteto shares the Grok Bot template tibot for monitoring X mentions and keywords

poteto has publicly shared tibot, the Grok Bot template they use to manage their X account. It can monitor mentions and keywords, generate short digests daily or hourly, filter out noise, translate non-English posts, and optionally auto-block harassing accounts. tibot does not post on your behalf; poteto still handwrites all posts and long-form articles. Automatic blocking requires users to connect their own X API app and will consume API credits.

Quoted postlauren · @poteto

i've gotten a lot of questions about how i manage my X account with grok bot, so i'm sharing my bot tibot today! he watches your mentions and: • finds every post about you or any keywords, tagged or not • gives one short, sorted digest a day (or hourly, your call) • skips spam and noise • translates non-English posts • can auto block trolls and abuse • never posts for you i still write all my posts and articles by hand but tibot has been very helpful with keeping me in the loop. hire tibot today! https://t.co/OaEAunGL8Y

Same story · 2 reports

#Product updates#Agents#English text is being prepared.

TTesting Catalog12:53Analysis

AI score58

Artificial Analysis Intelligence Index v5 is live, Claude…

Artificial Analysis Intelligence Index v5 is live, featuring 2 new benchmarks and 3 upgrades, of which 60% are based on private test sets or held-out answers. Claude Opus 5.5 (max) leads with 58 points, followed closely by GPT-6 Astra (max) at 57 points and GPT-6.1 Sol (max) at 55 points, with Gemini 4 Argon (high) further back at 50 points. The gap between open-weight models and the leader has widened from 3 points to 12 points; among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead with 32 points.

Quoted postArtificial Analysis · @ArtificialAnlys

Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers. Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale. New to the Intelligence Index: ➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change ➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment Upgraded: ➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions ➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised ➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0 Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1 Key results: ➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50 ➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36) ➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04

Same story · 8 reports

#Evaluations / benchmarks#Anthropic#OpenAI

AA Yi AI Notes12:53

AI score53

The Trick of Using a Scheduled Ping to Start the Codex 5-Hour Quota Window Early

The author shares a Codex quota window trick figured out by the community: the 5-hour quota starts counting down from the first message of the day, so at 7:00 AM you can use Grok Bot to automatically run codex exec ping to trigger the countdown; after the quota resets at 12:00 noon, it covers the afternoon work hours. The article provides a ready-to-copy scheduled prompt and notes that this mechanism comes from community testing rather than OpenAI's official documentation, and that read-only parameters must be included when executing.

#Tutorials / practice#English text is being prepared.#OpenAI

AAyi AI Notes12:53Analysis

AI score51

The author built an "EQ plugin" with Claude Opus 5.5 and used UUMit to turn a local service into a team link

The author used Cursor together with Claude Opus 5.5 to build a zero-dependency local web tool called "EQ plugin," which rewrites emotionally charged drafts according to the four elements of Nonviolent Communication, and includes seven types of communication targets and three tone levels: gentle, firm, and direct. The tool runs on the local machine at 127.0.0.1:8787, and the UUMit client generates a fixed HTTPS link to distribute to colleagues, with no need to buy a server or file for registration, and permissions can be granted and revoked per person. The article also provides an 800-character limit per input, default rate limits of 10 times per minute and 200 times per day, as well as the source code and System Prompt for two GitHub repositories.

#Tutorials / practice#Agents#English text is being prepared.

TTesting Catalog12:53Analysis

AI score70

Gemini 4 Carbon Tops Text Arena, Ranks Second in Agent Arena

Gemini 4 Carbon (High) took first place in Text Arena with 1528 points, surpassing Gemini 4 Argon (High)'s 1520 points. It ranks second in Agent Arena, behind only Claude Opus 5.5, and has entered a new Pareto frontier; in Web Dev Arena it ranks fourth with 1768 points, two points behind third-place Claude Sonnet 5.5. Arena data states its median cost per Agent Arena task is $1.80, 72% lower than top-ranked Claude Opus 5.5 (Max). Its Artificial Analysis V5 score is 59.

Quoted postArena.ai · @arena

Exciting news: Gemini 4 Carbon (High) from @GoogleDeepMind just made a huge debut on Arena, landing #1 in Text Arena, #2 in Agent Arena, and #4 in Code Arena: WebDev. Back at the top, Gemini 4 Carbon delivers frontier intelligence and reshapes the Pareto frontier across three arenas. In Agent Arena, Gemini 4 Carbon ranks #2 with a +16.2% net improvement across +21.9K real-world agentic sessions from our global community, surpassing top models GPT-6 Astra and Fable-5.1. At $1.80 median cost per task, Gemini 4 Carbon (High) costs 72% less than the #1 ranked Claude Opus 5.5 (Max). Its standouts by key signals: - #1 Steerability (+18.6%): can the model course-correct when you push back? - #2 Praise vs Complaint (+37.9%): implicit sentiment in users' reactions - #3 Confirmed Success (+18.7%): explicit “yes, that worked” from users In Code Arena: WebDev, Carbon scored 1768 pts landing #4: - Just 2 pts below #3 Claude Sonnet 5.5 (xHigh), at 1770 pts - 89 pts above #11 Gemini 4 Argon (High), at 1679 pts By category, it ranks #3 in Gaming and Simulations, and #5 in Brand & Marketing. In Text Arena, Gemini 4 Carbon (High) is the new #1 with 1528 pts, now ahead of Gemini 4 Argon (High) at #2 with 1520 pts. Congrats to the @GoogleDeepMind team on this huge milestone!

Same story · 2 reports

#Evaluations / benchmarks#Google#Agents

SShao Meng12:53Analysis

AI score59

DeepLearning.AI Launches Context Engineering Course, Teaching Agent Context Management

DeepLearning.AI has launched a new Context Engineering course, taught by Chroma CEO Jeff Huber and CMPND CEO Drew Breunig, framing context engineering as the main thread for keeping Agents reliable and token-efficient in complex multi-step tasks. The course builds a harness from scratch in Python and uses traces and evals to quantify each change, covering three capabilities—Agentic Search, Context Management, and Agentic Memory—across four modules, with labs mostly consisting of about 1-hour coding exercises. The course assumes intermediate Python, familiarity with LLM applications and tool calling, and basic retrieval.

#Tutorials / practice#Agents#RAG

PPromptArmor12:53First-partyAnalysis

AI score60

PromptArmor testing found that DeepSeek, Kimi, GLM, and Qwen are 88% more susceptible to cyberattacks than…

PromptArmor tested its most widely circulated attack samples across multiple models and found that DeepSeek, Kimi, GLM, and Qwen models are 88% more susceptible to cyberattacks than GPT, Claude, and Gemini models. The tests covered multiple cyber risk dimensions, including susceptibility to phishing, data exfiltration, and running malware, and the authors described the differences as significant.

#Safety / alignment#Evaluations / benchmarks#DeepSeek

PPromptArmor12:53First-partyAnalysis

AI score52

PromptArmor tested multiple models for cyberattack risk, claiming DeepSeek, Kimi, GLM, and Qwen are more...

PromptArmor said it used multiple highly viral attack samples to test cyber risks including phishing, data exfiltration, and propensity to run malware, and that DeepSeek, Kimi, GLM, and Qwen models were 88% more likely to be attacked than GPT, Claude, and Gemini. This conclusion comes from its main tweet; the specific testing methodology and sample size were not explained in the tweet.

#Safety / alignment#Evaluations / benchmarks#DeepSeek

SShao Meng12:53Analysis

AI score71

Addy on how to use Claude Code in the Claude Opus 5.5 era

Addy Osmani discusses how to use Claude Code in the Claude Opus 5.5 era, arguing that starting with the Claude 5 series, users should reduce piling up prompts and hand-holding task breakdowns, and instead use a single prompt that clearly states the goal and acceptance criteria, letting the model run to completion on its own and then stop. He recommends starting from /effort medium; when the goal is unclear, first have Claude interview you to produce a spec, then open a new session to implement it. For Claude Opus 5 and Claude Fable 5, he removed more than 80% of the Claude Code system prompt, with no noticeable drop in coding benchmarks. He also mentioned that the Claude Code interactive terminal recently defaulted to auto mode, and said this is not a safety guarantee; sub-agents should be split less by hand, and a single agent plus skills often yields better quality.

Quoted postAddy Osmani · @addyosmani

https://t.co/1txsLAxxM5

Same story · 2 reports

#English text is being prepared.#Agents#English text is being prepared.

SShao Meng12:53Analysis

AI score58

Matt Palmer demonstrates using Cursor Cloud Agent to generate a UI motion video against real Slack

Matt Palmer demonstrated opening Chrome and real Slack on the Linux remote desktop of a Cursor Cloud Agent, having the Agent generate a Slack UI motion video with code, and then iterating and revising after checking easing and motion patterns against the real interface. He broke the process down into an inner loop (the Agent does the work, checks against the real thing, and fixes) and an outer loop (a human provides references, preferences, and permissions and accepts the output), and proposed the Do→Codify→Generalize ladder, packaging one success into a Skill and then generalizing it. He emphasized that Slack was not decompiled; he merely gave the Agent a reference and asked it to extract visual patterns by comparison.

Quoted postmatt palmer · @mattyp

Agents rebuilt Slack as video (using code) by looking at the real thing 0:00 Why 0:25 @cursor_ai agents & computer use 1:03 Video as code 1:56 Slack 2:54 A chat, an agent, a computer 3:53 Inner loop and outer loop 5:04 The ladder 7:47 Skills 8:14 Feedback loops 10:02 Recap https://t.co/S91yI9vLop

#Tutorials / practice#Agents#English text is being prepared.