Updated
Updated
Showing low-relevance items too. Hide low-relevance items
Awni Hannun@awnihannunAI score13AIAwni Hannun suggests stating your own understanding of a topic before asking an AI, then having the model correct your mistakes. He argues this approach is a much better way to learn than asking the question directly.
NVIDIA AI Developer@NVIDIAAIDevAI score20AINVIDIA and Google DeepMind experts will host a live session on Friday, April 24 at 11:00 AM PDT showcasing Gemma 4 running on DGX Spark. The demos cover vision translation, long-context document Q&A, and real-time code generation, with audience questions welcome.
Tri Dao@tri_daoAI score42AITri Dao points to Wentao Guo's explanation of a mathematical rewrite of the MoE backward pass, which reduces activation memory and speeds up training, especially for fine-grained MoE. He also notes the work leverages new Blackwell features, including 2CTA MMA and CLC, to build fast MoE kernels.
AICognition reports that multi-agent setups work best when writes stay single-threaded and extra agents contribute intelligence instead of actions. It describes a code-review loop where a clean-context review agent catches bugs in Devin-written PRs, averaging 2 bugs per PR with roughly 58% severe. The post also says the smart-friend pattern, pairing a smaller primary model with a stronger one, has not yet worked well with asymmetrically weaker primaries and is an open training problem.
Why it matters: The post gives concrete findings on which multi-agent setups work, including clean-context code review and smart-friend escalation, and where they still fail.
NVIDIA AI Developer@NVIDIAAIDevAI score13AINVIDIA AI Developer announced a live broadcast on contributing to open source, featuring NVIDIA NemoClaw and Nemotron Labs. The post links only to the broadcast and gives no further details on the content or speakers.
NVIDIA AI Developer@NVIDIAAIDevAI score26AIA developer trained a 12M-parameter LLM on a custom ML framework built with a Rust backend and CUDA kernels, including Flash Attention, fused LayerNorm, and fused GELU. The framework claims 3x throughput gains, WebGPU fallback for non-NVIDIA devices, and a TypeScript API installable via npm.
NVIDIA AI Developer@NVIDIAAIDevAI score46AINVIDIA's developer account published a step-by-step tutorial for building a fully local, sandboxed, always-on AI agent. The guide uses OpenClaw with NVIDIA NemoClaw and runs on NVIDIA DGX Spark.
Awni Hannun@awnihannunAI score23AIA blog post by @pcuenq describes a new skill and test harness for automating the porting of new models from Transformers to mlx-lm. The post is shared by Awni Hannun, who highlights it as a useful guide for this workflow.
Andrej Karpathy@karpathyAI score62AIKarpathy describes using LLMs to compile raw source documents into a markdown wiki that he views in Obsidian, with the LLM writing and maintaining most of the wiki. He reports that at about 100 articles and 400K words, the LLM agent can answer complex questions directly from the wiki, and he also runs LLM health checks to find inconsistencies and gaps. He shares the underlying idea as an "idea file" that users can give to their own agents to build a customized version.
Andrej Karpathy@karpathyAI score49AIAndrej Karpathy describes using LLMs to compile raw research sources into a markdown wiki of about 100 articles and 400K words, viewed in Obsidian. He says an LLM agent answers complex questions against the wiki without RAG, with outputs filed back to enhance it. He also suggests the workflow could become a product rather than a collection of scripts.
AIAnthropic describes Claude Code auto mode, which delegates approval of agent actions to model-based classifiers instead of manual prompts or skipped permissions. The classifier reviews tool calls before execution and a separate probe screens tool outputs for prompt injection. Anthropic reports a 0.4% false positive rate on real internal traffic and a 17% false negative rate on real overeager actions.
Why it matters: The post explains the layered classifier design and its measured tradeoffs, showing how autonomous coding agents can cut approval fatigue without fully removing risk.
AIAnthropic's Labs team describes a three-agent harness with planner, generator, and evaluator agents for building full-stack applications over multi-hour autonomous coding sessions. The evaluator uses Playwright to test the running app against sprint contracts, and a retro game maker built with the harness worked end to end where a single-agent run's core feature did not. The author later removed the sprint construct and kept only the components still needed on Opus 4.6.
Why it matters: The post shows how a generator-evaluator loop, with explicit grading criteria and a tuned QA agent, turned a solo run's broken output into a working app, and how the harness was pruned as models improved.
Nano Banana 2.1@NanoBananaAI score62AINano Banana 2, also called Gemini 3.1 Flash Image, adds visual grounding with Google Search, 512px resolutions, and extreme aspect ratios of 1:8 and 1:4. The guide advises using it as the default for new projects, with Nano Banana Pro reserved for complex prompts it fails, and keeping Thinking mode off by default.
Nano Banana 2.1@NanoBananaAI score23AIGoogle's Nano Banana 2 image model lets users add exact details to outputs by writing short sentences in prompts. The post contrasts a bare prompt for a snow leopard portrait with a detailed one specifying pose, melting snow, flowers, a sun dog, lighting, and blue eyes.
AIPublic AI benchmarks such as MMLU, SWE-bench Verified, and GPQA Diamond are saturating or showing contamination, prompting OpenAI to call SWE-bench Verified "no longer suitable" in late February and recommend SWE-bench Pro. OpenAI's audit found 59.4% of the problems its best model failed had flawed test cases, and GPT-5.2, Claude Opus 4.5, and Gemini 3 Flash could reproduce original fixes from memory. The article argues that behavioral tests, such as Vending-Bench's simulated vending machine business, may be more useful for everyday model choice.
Nano Banana 2.1@NanoBananaAI score18AIGoogle's Nano Banana account says users can turn anything into an enamel pin with Nano Banana 2, either by selecting the "Enamel pin" style in the Gemini app or by using a prompt. The example prompt asks for just the subject as a gold enamel pin shown as a minimal photo on a desk.
Max Woolf@minimaxirAI score38AIMax Woolf published a blog post summarizing what he learned over several months working with agents including Opus 4.5 and beyond. The post also describes an optimization trick he considers promising.
Nano Banana 2.1@NanoBananaAI score10AIGoogle's Nano Banana 2 is presented as the best model for a prompt that asks for a cute, nearly spherical animal with oversized eyes and no separate head, shown in its natural habitat. The example prompt specifies cartoonishly cute features with the eyes looking up cheekily.
AICognition reports merging 659 Devin PRs into its own codebase last week, up from 154 in its best week in 2025. The post describes internal workflows across web, Slack, Linear, CLI, and API, including Devin Review for PR diffs and bug catching, a daily design system audit, automated bug triage on Linear, and DANA for data analysis.
Why it matters: The post shows concrete workflows for using Devin across Slack, Linear, and code review, with specific usage figures that help teams judge fit for their own engineering processes.
Nano Banana 2.1@NanoBananaAI score10AINano Banana, the account linked to Google and Gemini, shares a prompt asking an image model to imagine the image as a simple cookie cutter and show the resulting unfrosted homemade cookies on a plate without showing the cutter. The post is a creative prompt example rather than an announcement of a new feature.
AIThe article argues that engineers are splitting their work between building a harness of constraints, tools, and documentation for agents and directing the agents' work. It cites OpenAI, Stripe, and Anthropic examples, including architecture guardrails, custom linter messages, AGENTS.md updates, and plan-first execution. The author notes that open problems remain around code maintainability, verification at scale, and adopting these practices in older codebases.
Eugene Yan@eugeneyanAI score27AIAnthropic's Claude API now supports automatic prefix caching, enabled by setting one cache control field at the top level of the request body. Developers still need to structure prompt templates so the reusable prefix stays consistent to benefit from caching.
Oriol Vinyals@OriolVinyalsMLAI score5AIOriol Vinyals shared a prompt asking an AI model to generate an SVG of a pelican riding a car in France, with a cat beside it and the Eiffel Tower in the background. The post offers no model name, results, or benchmark figures.
Oriol Vinyals@OriolVinyalsMLAI score4AIGoogle DeepMind researcher Oriol Vinyals posted a simple prompt asking for an SVG image of a rollercoaster. The post includes no model name, output, or benchmark result.
Nano Banana 2.1@NanoBananaAI score6AINano Banana, Google's Gemini image account, shares a prompt for a 9:16 wallpaper with negative space near the top and bottom, a subtle top-to-bottom gradient, no blur, and no cropping. The post advises keeping the scene simple and placing the subject and style at the end of the prompt, with example images attached.
Nano Banana 2.1@NanoBananaAI score14AINano Banana (@NanoBanana) posts a sample prompt showing how specific image-generation instructions can be, describing a San Francisco cafe scene with a couple by the window, a sleeping black-and-white French bulldog, and a mirrored "Nano's Cafe" gold lettering on the glass. The post offers the detailed prompt as an illustration of how far users can go with specificity, covering clothing, props, lighting, and poses.
Max Woolf@minimaxirAI score14AIMax Woolf reports that experiments using GPT-5.3-Codex to hyperoptimize batch prediction are going well. The post gives no benchmark scores, speedups, or further technical details.
AINicholas Carlini of Anthropic's Safeguards team describes an agent-team setup where 16 Claude instances worked in parallel on a shared codebase without human intervention to write a Rust-based C compiler. Over nearly 2,000 Claude Code sessions costing about $20,000 in API fees, the team produced a 100,000-line compiler that can build Linux 6.9 on x86, ARM, and RISC-V. The post focuses on harness design, including high-quality tests, lock files for task claiming, GCC as a reference oracle for the kernel, and the limits the project reached.
Why it matters: The post shows concrete harness design choices for long-running agent teams, including test design, locking, and parallel work division, that readers can adapt to their own autonomous projects.
AIAnthropic's performance engineering lead Tristan Hume describes how a take-home test for hiring performance engineers was repeatedly defeated by successive Claude models. Claude Opus 4 outperformed most human applicants within the 4-hour limit, and Claude Opus 4.5 matched the best candidates in 2 hours. Anthropic is releasing the original take-home as an open challenge, with the best known Claude result at 1487 cycles.
Why it matters: The post traces how each Claude model defeated the take-home test, showing concrete redesign tradeoffs for evaluating engineers when AI assistance is available.
Nano Banana 2.1@NanoBananaAI score18AINano Banana (@NanoBanana) suggests adding the word "risograph" to image prompts for a distinctive print effect. The example prompt asks for a 2x2 grid of icons on pure white with thick black outlines, stochastic stippling, and vibrant, unfaded color, here themed "cute but weird animals (that are real)."
AITim Dettmers, a professor who has used Claude Code for eight months to automate his own work, argues that more than 90% of code and text should be written by agents. He says the coding-focused hype on Twitter overstates parallel sessions and autonomy, which translate poorly to most non-software tasks. The post offers a balanced guide to what actually works in agent-based automation.
Nano Banana 2.1@NanoBananaAI score12AINano Banana posts a prompt template for generating a 3x3 grid of colorful, tactile 3D icons on a white background with no text. The post lists example themes: dogs with different emotions, bananas, January, and the same cat in different emotions.
Nano Banana 2.1@NanoBananaAI score13AIGoogle's Nano Banana account says Nano Banana Pro can search for how the Gemini 3 Flash launch was received and turn those findings into a short, headed article. The demo prompt also asks for a rendered image of a glossy magazine cover and two-page spread.
Nano Banana 2.1@NanoBananaAI score4AINano Banana, Google's Gemini image account, posts a sample prompt for a traditional Christmas card featuring a couple in Christmas hats. The prompt specifies the text "Happy Christmas 2025" and a classical, friendly design in a winter wonderland setting.
Nano Banana 2.1@NanoBananaAI score14AINano Banana, the account owned by Google and Gemini, thanks @Arminn_Ai for inspiring a post showing 3D miniature isometric room images made with Nano Banana Pro. The post says users can upload their own photo to generate a personalized chibi-style figure in a cube-shaped room and offers to share the full prompt on request.
Nano Banana 2.1@NanoBananaAI score24AINano Banana, the account of Google's Gemini team, shares a prompt template for generating cute isometric 3D cube diorama lamps. The prompt specifies internal lighting, chibi figurine styling, matte PVC material, a neutral background in a dark room, many small details, and subtle dust and scratch textures.
Nano Banana 2.1@NanoBananaAI score4AINano Banana, Google's image model account, shares a prompt for a photo of a miniature chocolate shaped like a cartoon image shown as an advent window backdrop. The chocolate is held after being taken from a chocolate advent calendar.
AICognition says its autonomous coding agent Devin can complete a .NET Framework to .NET Core migration in as little as two weeks, using a Strangler Fig approach adapted from Jimmy Bogard's guide. The post says Devin handles planning via Ask Devin and DeepWiki, dependency sharing, controller and view conversion, and session state adaptation through a remote app.
AICognition rebuilt its Devin coding agent for Claude Sonnet 4.5, reporting 2x faster performance and 12% better results on its Junior Developer Evals, now available in Agent Preview. The team found the model is aware of its context window, which led to premature wrap-up behavior that they countered with repeated prompts and a 200k usage cap within a 1M token beta.
Why it matters: The post explains which agent behaviors changed under Sonnet 4.5, such as context-window awareness and note-taking, that forced a rebuild rather than a simple model swap.
AIEight Sleep integrated Cognition's Devin into its data workflows, letting staff tag Devin in Slack to query Snowflake, dbt, and Looker and check Amplitude. The company says it is now shipping 3x as many data features and investigations each week, with its ad-hoc data request queue near zero. Devin was used to trace a suspicious revenue spike to a better-than-expected email campaign.