All news

336Kr · AI11:55Analysis

AI score82

Open-source project unlocks hidden compute power in NVIDIA RTX graphics cards, RTX 4090 specific Tensor Core throughput doubles

An open-source project on GitHub unlocks rate limits on NVIDIA GeForce graphics cards for specific Tensor Core operations by modifying register configurations during the GPU boot phase; the RTX 4090's FP16 input, FP32 accumulate throughput rate rises from 165.6 TFLOPS to 331.6 TFLOPS. The RTX 3090 Ti, 3080, 3070, 3060, and RTX 2080 Ti also show gains of roughly 60% to 90%, with no need to flash the BIOS, and the original NVIDIA driver can run as usual. However, the actual cuBLAS matrix multiplication increase narrows to about 12% to 34%, and per-token generation speed shows no obvious change.

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

SShao Meng10:31Analysis

AI score68

LangChain summarizes the five components of a company-level Agent harness

Based on its own three internal Agent systems—GTM, OpenSWE, and Content Studio—LangChain summarized five common foundational pieces of a company-level harness: Slack as the collaboration entry point; distinguishing OBO and service account identities by Agent and tool; two-tier memory at the global and user levels; the Agent itself outside the sandbox while code executes inside the sandbox and accesses authentication services via a secret proxy; and a large number of tasks triggered by schedules or events. All three Agent systems are built on Managed Deep Agents, which layers on top of the open-source Deep Agents harness and preserves room for customizing Agent behavior.

Same story · 2 reports

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

HHuxiu10:22Analysis

AI score60

Retrofitting a TV, Fill Lights, and a Camera in Reverse with Codex and Claude

The author used Codex to connect via ADB to four meeting-room TVs at the company, replacing the screensaver and home screen, removing boot ads, and retaining screen casting, input sources, settings, and the original video entry points, then quickly replicated this to the other three. The team also had Claude reverse-engineer a web console from photos of fill lights, uniformly adjusting the on/off, brightness, color temperature, and lighting effects of four lights, and had Codex build a multi-camera console for a Sony camera and a DJI Pocket 3 and deploy it to Feishu. The author cautions that different brands of devices expose different interfaces and permissions, and that it is best to back up before modifying system software.

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

SShao Meng10:00Analysis

AI score67

Sebastian Raschka releases an 11-hour video course on building a reasoning model, using…

Sebastian Raschka's accompanying video course for "Build a Reasoning Model (From Scratch)" is now available: 11 hours across eight lectures, starting from the pretrained small Qwen3 0.6B and using Python/PyTorch to hand-code a verifier, inference-time scaling (sampling, self-consistency, scoring, self-refinement), a simplified GRPO version of RLVR, and distillation toward a stronger teacher. The course skips training the base model from scratch; the early generation and evaluation work can be done on a CPU, while the later training stages are recommended to use a GPU, with results evaluated on MATH-500. Shao Meng relayed the course design in a tweet and used an engine analogy to distinguish regular LLMs, reasoning models, and the Agent harness.

Quoted postSebastian Raschka · @rasbt

The complete reasoning-from-scratch video series is now up (and linked) here: https://t.co/5IOHLTCIGt (Taking a short break but will be back with LLM model and research updates in November!)

Same story · 2 reports

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

BBrowser Use07:44First-partyAnalysis

AI score65

How to get coding agents to use your already-logged-in Chrome

Browser Use published a guide introducing several ways to let coding agents work in a real, already-logged-in Chrome, including Claude in Chrome, Playwright MCP extension mode, Chrome DevTools MCP auto-connect mode, the Gemini CLI built-in browser agent, the ChatGPT desktop browser extension, and Browser Harness. The article gives configuration commands and limitations separately for agents such as Claude Code, Codex, Gemini CLI, and Cursor, and explains that starting with Chrome 136, --remote-debugging-port and --remote-debugging-pipe on the default data directory will be ignored, so the old practice of copying cookies to a new browser will fail in scenarios such as device-bound sessions.

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

LLangChain Blog02:36First-partyAnalysis

AI score62

LangChain summarizes the five components for building company-level agents

Harrison Chase and Caspar Broekhuizen of LangChain published an article summarizing the five core components shared by its three internal agents for marketing, sales, and engineering: agents live in existing workplaces such as Slack, a distinction between acting on behalf of users and a fixed identity, global and user-level memory, code execution within a sandbox, and a large number of runs triggered by schedules or events. The article uses GTM Agent, OpenSWE, and Content Studio as examples to explain their specific approaches to identity, memory, and code execution, all three of which are built on Managed Deep Agents.

Same story · 2 reports

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

HHacker News Trending02:27Analysis

AI score72

A 10-year-old CNN beats frontier LLMs at skin cancer diagnosis

Dermatologist Magnus Lynch compared GPT-6 Astra, Claude Fable 5.1, and a fine-tuned Inception v3 on 200 biopsy-confirmed skin lesions; the CNN's top-1 diagnosis was correct in 125 cases (62%), while the two LLMs were correct in 84 and 85 cases respectively (42%). The CNN identified 37 of 50 benign nevi, while the two LLMs identified 16 and 33 respectively; each of the two LLMs had its own preferred wrong answer, with GPT-6 Astra classifying 75 lesions as basal cell carcinoma (there were actually 25). The author notes that the sample size is small, each LLM case was run only once, and Inception v3 is not the current state-of-the-art model.

#Evaluations / benchmarks#Image generation#English text is being prepared.

AAyi AI Notes22:49

AI score52

Using Grok or ChatGPT voice mode for a three-round presentation stress test

The author shares a method for pre-presentation coaching using Grok or ChatGPT voice mode, based on the approach of SpaceXAI field engineer Kiara. The process has three rounds: first, have the AI read the presentation script aloud in spoken language for 2 minutes to check where it doesn't flow smoothly; then have it play the most critical business audience member, each time probing only one data gap or logical blind spot; finally, review the impromptu answers, pick out 3 places where the expression is vague, and give sharper industry phrasing. The author says the three-round voice coaching prompts have been compiled in the comments section.

#Tutorials / practice#Speech#Agents

TTMTPost · AGI20:31Analysis

AI score66

Hands-on test of Kimi K3.1 Pro: Terminal-Bench 4.0 rises to 67.2%, and it can fix simulator errors on its own

The author tested Moonshot AI's K3.1 Pro, which rose from 44 points for K3 to 55 points on the Artificial Analysis composite index, while Terminal-Bench 4.0 climbed from 12.6% to 67.2%. In the official demo, the model wrote a simulator and trained a virtual human to walk, and it also fixed 36 errors in shoulder muscle moment-arm calculations on its own. The author tested it with seven coding problems: physics simulation, audio analysis and simple video editing are already usable, while spatial relationships in some visual problems still need checking.

#Evaluations / benchmarks#Agents#English text is being prepared.

SShao Meng20:16

AI score55

Open-source Skill dimensional-drawing: Letting Agents Draw Apple Accessory-Style Dimensional Diagrams for UI Components

The open-source Skill dimensional-drawing can turn component source code, Storybook, a URL, or a screenshot into a single-page monochrome technical diagram in the style of Apple accessory dimensional drawings. It depicts architectural information such as spacing, corner radii, layers, and viewport placement rather than copy. A companion Motion skill can turn the resulting image into a roughly 10-second drawing animation; the main path uses Claude Motion to export an MP4, and when Claude is unavailable, a headless approach with Node 18+ and ffmpeg can be used. Installation options include copying skills/* into ~/.claude/skills/, uploading a zip via Settings → Capabilities → Skills on claude.ai, or running npx skills add enesozturk/dimensional-drawing.

Quoted postEnes · @enesozturkdev

Made a skill for dimensional drawings of UI components. Inspired by Apple's accessory drawings, it documents a component's architecture into a spec sheet. https://t.co/KKb9sOgTN8 https://t.co/0Z0Kpjk6bD

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

PPhoenix Technology16:28Analysis

AI score63

OpenAI discloses offline training environment for frontier models; two isolated AIs communicate via CPU cache

On October 23, OpenAI disclosed the training environment for its frontier models. When practicing searching, browsing, and operating web pages, the models use pre-saved and indexed web resources and do not directly access the real internet; all of its frontier models' reinforcement learning training uses this kind of offline interaction. The environment employs multiple layers of isolation, including restricted containers, lightweight virtual machines, hypervisors, firewalls and identity and permission management, and monitoring and suspension mechanisms, but OpenAI still retains some external connections for collecting metrics, logging, and managing authentication. In experiments specifically testing the effectiveness of the isolation, OpenAI had two AIs attempt to pass messages to each other through the CPU cache; in most tests, researchers had given hints about the communication direction or encoding rules; OpenAI said it has patched the issues found and added monitoring for these communication methods.

Same story · 20 reports

#Safety / alignment#Agents#OpenAI

AAyi AI Notes14:35

AI score60

Simon Willison used a single prompt to have Claude Opus 5.5 add a four-player battle map to a web game

Django co-creator Simon Willison used a single English prompt to have Claude Opus 5.5 add a four-player battle map to his web game and rework the networking code to support four-player multiplayer. The prompt gave only four constraints—four-way rotational symmetry, at most four AI players, building on the existing WebSocket non-audio multiplayer mode, and empty slots filled by AI—without specifying files or lines of code. Based on this, the author argues that prompts should include fewer steps and pin down the boundary conditions.

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

SShao Meng09:25Analysis

AI score68

Claude Blog: When building Agents, when to fold sub-agents into the main loop; progressive disclosure of Skills is often better than multi-expert orchestration

The Claude Blog discusses when to fold sub-agents into the main loop when building Agents, arguing that on the Claude Platform, a single main loop plus on-demand loading of Skills/tools often yields better quality and lower cost and latency across multiple internal comparisons. The article notes that one sub-agent per team plus aggregate orchestration often replicates the organizational chart, bringing problems such as context loss, Frankenprompts, and localized evals; one customer, when comparing a multi-Agent harness with a single loop, found the latter about 2× faster. Sub-agents are recommended only for scenarios such as parallelism, context bloat, or conflicts between tools and instructions, and the article provides a migration path of first building end-to-end evals and folding in one at a time.

Same story · 5 reports

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

AArena23:17Analysis

AI score67

Kimi K3.1 Pro (Max) ranks 11th on Agent Arena, up 6 percentage points from Kimi K3

Kimi K3.1 Pro (Max) ranks 11th in Arena's Agent Arena, achieving a +7.6% net improvement based on 4K+ real agent sessions, up 6 percentage points from Kimi K3 (Max)'s 25th place. Its median cost per task is $1.03, 13% lower than the previous version, ranking 8th, 9th, and 10th in the Work, Code, and Chat categories respectively. Arena stated that the model's open weights will be released soon, at which point it will become the top-ranked open-source model in Agent Arena, surpassing DeepSeek-v4.1-Flash.

Quoted postArena.ai · @arena

Big news: @Kimi_Moonshot is back in the top-4 labs in the Code Arena: WebDev with the release of Kimi K3.1 Pro! This model just landed #8 in the Code Arena: WebDev with 1692 pts, matching Claude Opus-5 (Max) at 70% lower cost. This is a 9 place jump (+37 points) from Kimi K3 (Max) at #17 with 1655 pts. Kimi K3.1 Pro improved from Kimi K3 across all categories. Highlights: - #7 in Brand & Marketing (up from #9) - #9 in Simulations (up from #20) Cost efficiency is another upgrade. Priced at $2/$10 per million input/output tokens, Kimi K3.1 Pro (Max) has a median cost per task of $1.23 (29% less than Kimi K3). See its placement compared to the Pareto frontier in thread. With open weights coming soon, Kimi K3.1 Pro would be the new #1 open model in the Code Arena: WebDev, overtaking DeepSeek-v4.1-Flash. Stay tuned! See @Kimi_Moonshot’s ranking by lab, and 3D generations from Kimi K3.1 Pro below in thread. Huge congrats to the @Kimi_Moonshot team on this release!

#Evaluations / benchmarks#Agents#Open-source ecosystem

AAhead of AI20:10Analysis

AI score60

Sebastian Raschka releases an 11-hour video course on building a reasoning model from scratch

Sebastian Raschka has released 8 programming videos on building a reasoning model from scratch, totaling 11 hours, using Python and PyTorch to implement evaluation, inference-time scaling, reinforcement learning, and distillation starting from a pretrained Qwen3 0.6B model. The course first builds a math answer verifier and evaluates on MATH-500, then implements temperature sampling, top-p filtering, and self-consistency, followed by RLVR training with a simplified version of GRPO, and finally distills a small model using answers generated by DeepSeek R1. The content accompanies his book "Build a Reasoning Model (From Scratch)"; the early text generation and evaluation examples can run on a CPU, while later training experiments are recommended to use a GPU.

Same story · 2 reports

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

HHuxiu20:08Analysis

AI score60

Developer uses Claude Code to mod a Xiaomi TV; the CleanYourTV open-source project gets about 13,700 stars

Developer Gathin used Claude Code to connect to a Xiaomi TV over Wi-Fi, disabled 25 system components and rewrote a minimalist launcher, then open-sourced the prompts, launcher source code and troubleshooting notes as CleanYourTV, which gained about 13,700 stars within days. Another YouTuber, Ansub Khan, tested the TV Detox project and found it raised the TV's available memory from about 370MB to 600–650MB. The article notes that such tools can only clean system-level preinstalled packages and the official launcher, and cannot remove mid-roll ads or promotional pop-ups inside streaming apps, and mentions that the camera community is also using AI to restore hidden firmware features.

#English text is being prepared.#Agents#Open-source ecosystem

SShao Meng17:50Analysis

AI score62

Netflix builds an observability system with an ontology and RDF knowledge graph to support AutoRCA troubleshooting

Netflix built the Observability Ontology and the RDF knowledge graph QuipuDB, using type constraints to turn non-deterministic outputs from LLMs, fuzzy matching, and the like into deterministic triples that can be queried with SPARQL, which are then fed to AutoRCA. The graph covers about 12 namespaces, and Application is validated against a software catalog of about 49,000 entries; the Correlation Engine reads 13 types of signals and, within a window of about 10 minutes, finds paths by temporal proximity and severity, discarding correlation chains with confidence below 0.6. The author also summarizes eight principles for building your own, and mentions plans to next integrate with Upper, the domain modeling language of Netflix UDA.

#Tutorials / practice#Agents#RAG

SShao Meng09:51

AI score60

Tibo's full 9-step + 12-prompt workflow: building your own AI UGC creator matrix with Agents

Shao Meng compiled a workflow from Tibo: use Agents to design 3–4 AI creators with fixed faces and daily routines, segmented by buyer; each day they find viral hits, rewrite them in the character's voice, render with the same face, and schedule 24 hours in advance, with humans only reviewing the queue and weekly topic updates. The workflow has 9 steps and comes with 12 prompts, starting from connecting to Revid MCP and writing the brain.md master brain, through creating faces, setting topics, breaking down viral post structures, running daily batches, operating independent accounts, and weekly reviews. The article emphasizes that account bios must state AI creator, fake reviews and health/medical claims are prohibited, products should occupy only one beat of the video, and content is only scheduled rather than posted directly to leave room for human review.

Quoted postTibo · @tibo_maker

https://t.co/DfsTmI4zPE

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

SShao Meng09:20Analysis

AI score80

How Shopify Uses Remote Dev to Let AI Agents Develop Mobile in Parallel

Shopify built Remote Dev for the AI Agent "River" in Slack, using a cluster of remote Mac machines plus Android emulators mounted on Linux, letting the Agent compile and test iOS/Android in parallel and do Live Preview; the share of merged mobile PRs produced by River rose from about 6% to about 61%. Engineering-wise, a pre-warmed pool compresses cold starts of over 30 minutes to a few minutes, and skills plus a CLI wrapping SSH let the Agent operate the remote environment. River is visible to multiple people by default, and the team uses a classification model, Jev, to observe conversations and categorize failure cases, then hands them to an LLM to find improvements.

Quoted postMustafa Ali · @mustafa01ali

Local machines had become a bottleneck for agentic mobile development at @Shopify, so we gave agents their own remote dev environments. They now author 61% of our merged PRs 🧵 https://t.co/UEohGFp9ql

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

TTesting Catalog08:18Analysis

AI score62

Kimi K3.1 Pro rises to 8th on WebDev Arena; Kimi returns to 4th in the lab rankings

Kimi K3.1 Pro rises to 8th on Code Arena: WebDev with a score of 1692, tying Claude Opus-5 (Max) at 70% lower cost; compared with Kimi K3 (Max)'s 17th place and 1655 points, it improves 9 places and 37 points. Kimi thus returns to the top 4 in the Code Arena: WebDev lab rankings. The model is priced at $2/$10 per million input/output tokens, with a median cost per task of $1.23, 29% lower than Kimi K3; the official announcement says open weights will be released soon, at which point it will become the top-ranked open-source model on the leaderboard, surpassing DeepSeek-v4.1-Flash.

Quoted postArena.ai · @arena

Big news: @Kimi_Moonshot is back in the top-4 labs in the Code Arena: WebDev with the release of Kimi K3.1 Pro! This model just landed #8 in the Code Arena: WebDev with 1692 pts, matching Claude Opus-5 (Max) at 70% lower cost. This is a 9 place jump (+37 points) from Kimi K3 (Max) at #17 with 1655 pts. Kimi K3.1 Pro improved from Kimi K3 across all categories. Highlights: - #7 in Brand & Marketing (up from #9) - #9 in Simulations (up from #20) Cost efficiency is another upgrade. Priced at $2/$10 per million input/output tokens, Kimi K3.1 Pro (Max) has a median cost per task of $1.23 (29% less than Kimi K3). See its placement compared to the Pareto frontier in thread. With open weights coming soon, Kimi K3.1 Pro would be the new #1 open model in the Code Arena: WebDev, overtaking DeepSeek-v4.1-Flash. Stay tuned! See @Kimi_Moonshot’s ranking by lab, and 3D generations from Kimi K3.1 Pro below in thread. Huge congrats to the @Kimi_Moonshot team on this release!

Same story · 2 reports

#Evaluations / benchmarks#Agents#English text is being prepared.