atharva

@atharvaraykar.com

116 Followers 198 Following

i write at atharvaraykar.com i work @nilenso.com yes-anding the world.

72 upvotes. Title: AI as a Fascist Artifact | 47 comments
Our tasks typically use environments that do not significantly change unless directly acted upon by the agent. In contrast, real tasks often occur in the context of a changing environment. […] Similarly, very few of our tasks are punishing of single mistakes. This is in part to reduce the expected cost of collecting human baselines. This is not at all like the tasks I am doing. METR acknowledges the messiness of the real world. They have come up with a “messiness rating” for their tasks, and the “mean messiness” of their tasks is 3.2/16. By METR’s definitions, the kind of software engineering work that I’m mostly exposed to would score at least around 7-8, given that software engineering projects are path-dependent, dynamic and without clear counterfactuals. I have worked on problems that get to around 13/16 levels of messiness. An increase in task messiness by 1 point reduces mean success rates by roughly 8.1% Extrapolating from METR’s measured effect of messiness, GPT-5 would go from 70% to around 40% success rate for 2-hour tasks. This maps to my experienced reality.
screenshot: PART I: 2025 KEY INITIATIVES Toxicity Filtering a Toxicity is a persistent challenge for all large-scale social apps. As communities grow, maintaining space for both friendly conversation and fierce disagreement requires intentional design choices. Our community doubled in size over the past year, and with that growth came tension: how to preserve healthy discourse while respecting genuine debate and diverse user preferences. Toxic and inflammatory discourse appears across all forms of social media; and almost universally, it's the case that a small percentage of people contribute disproportionately to causing this problem. A tiny number of users can have an outsize impact on conversation quality and on people's willingness to participate. In 2023-2024, anti-social behavior, such as harassment, trolling, and intolerance, consistently ranked among our top complaints reported by users. This content drives people away from forming connections, posting, or engaging, for fear of attacks and pile-ons.
clawdbot star history showing hockey stick growth, inflection point on Jan 20-ish
SWE-bench Verified and SWE-bench Pro What it measures How well a coding agent can submit a patch for a real-world GitHub issue that passes the unit tests for that issue. The specifics There are many variants: Full, Verified, Lite, Bash-only, Multimodal. Most labs in their chart report on SWE-bench Verified, which is a cleaned and human-reviewed subset. Notes and quirks of SWE-bench Verified: It has 500 problems, all in Python. Over 40% are issues from the Django source repository; the rest are libraries. Web applications are entirely missing. The repositories that the agents have to operate are real, hefty open source projects. Solutions to these issues are small—think surgical edits or small function additions. The mean lines of code per solution are 11, and median lines of code are 4. Amazon found that over 77.6% of the solutions touch only one function. All the issues are from 2023 and earlier. This data was almost certainly in the training sets. Thus it’s hard to tell how much of the improvements are due to memorisation.
My Quarterly System Health Check-in It is essential to periodically take a few steps back from the day to day and reflect on where we are against our strategic goals. If you’re an engineering leader, a head of engineering, a director, or a VP, you likely have a recurring meeting to this effect. In this post, I propose a structure for this operational exercise (complementing a business review) that lasts 2-4 hours, every month or quarter. I see quality as solving for the Pareto front with the tangible dimensions of reliability, performance, cost, delivery and security, and the more intangible dimensions of simplicity and social structures. For each dimension, go through the list of questions below and try to answer them together.
It’s a recursive paradigm shifting paradigm shift Code is data is code. Lispers get this, its turtles all the way down. Let’s walk through it. 1. Software can use AI. We write pieces of text in between code that uses AI’s thinking or knowledge access capability. Like with autocompletion, or chatbots. 2. AI can use software. It can access the filesystem and run programs on your OS, if you let it. It can access the web through search engines, or web browsers, if you let it. 3. AI can build software. It writes and executes small python scripts to analyse data when thinking. The agency and autonomy needed to build full fledged meaningful software isn’t there with AI yet, but it’s advancing quickly. AI assisted coding is pretty big, you know this. 4. AI IS software. AI is a trained neural network, and with sufficiently advanced capabilities, it can build itself.
Image description While Kuhn doesn’t go into it, the technological diffusion, and the economic impact of scientific revolutions are better studied through the GPT (general purpose technology) paper from Bresnahan & Trajtenberg in 1995. > “General Purpose Technologies (GPTs) are technologies that can affect an entire economy (usually at a national or global level). They have the potential for pervasive use in a wide range of sectors and, as they improve, they contribute to overall productivity growth.” And Calvino et al in June 2025, finds that AI meets the key criteria of a General Purpose Technology. It’s pervasive, rapidly improving, enables new products, services and research methodologies, and enhances other sectors’ R&D and productivity.
wikipedia screenshot of brooks's law page.
Table of Contents General guidelines Starting Points Official announcements, blogs and papers from those building AI High signal people to follow News and Media Esoterica Do I chug water from a firehose?
Andrej Karpathy on the YCombinator startup school stage talking about my AI-assisted coding blog post
In brief, directly or indirectly, I will offer help and happiness to all my mothers, And secretly take upon myself All their hurt and suffering.
demo of tldraw computer program
google's menu highlights for a thindi joint. the first highlight is half eaten masala dosa accompanied by a picture.
dreamy surreal distortion of a bicycle framed outside hae kum gang korean restaurant gents toilet
chai darkening in action