Andrew Hinson

@sama That's a no from me dawg, I don't want these LLMs making slop daily summaries for the family. What is this some Microsoft teams email synthesis utopia we're living in??

cool use case of chatgpt work i heard last night: connect your family calendars and explain your kids' interests. every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming birthday, some news, etc.

Your harness will suck because you won't support it for the next 5+ years. Just build extensions, plugins, for well established harnesses you want to see improve. Outside of learning and system design study they really are not that crazy of a project

Is everyone really trying to build a harness? I don't see the point outside of learning. They are all almost the same. A harness is more about the group or org behind it. You want the long term support and integration work with third parties (subscriptions, APIs, etc...)

@elpatox16 @coolcoder56 Yeah it would depend how these are. If they are applicable problems then sure but if it's some foreign hard problem then that's not fun. Also 8 rounds is ridiculous. Moving to the Bay area ain't worth it

@hinsonan @coolcoder56 the technical screen and onsite parts still have some random leetcode stuff

OpenAI Software Engineer Interview Experience

@habibislop @panPytlik @coolcoder56 You ain't wrong

@panPytlik @hinsonan @coolcoder56 it would be economy destroying lol and yeah they would but that doesn't make tech interviews any less of a humiliation/hazing ritual. millions of people are literal cucks too

@wagslane Competition for remote is dumb. You have to beat the filter from 1000+ foreign applicants. It's a mess.

Pro tip: if you're having trouble finding an interview, apply for in-person roles Remote is playing on giga hard mode

That's a crazy reduction in price. The bang per buck on Luna was already pretty great. Getting the "lightweight" and "middleweight" models cheaper and better have large impact

@UnrealSensei ...this is a joke right?

UE6 should ditch UMG and replace it with HTML and CSS. AI is unbelievably good at frontend web development. We can take all the improvements for web generation and apply it to game UI and menus.

@ThePrimeagen It's pretty concerning this company and anthropic are failing in this area. It's not that hard. It's a real "emperor has no clothes moment"

better post: OpenAI doesn't know what a sandbox actually is and is currently one of the leaders in technology

Is OpenAI incapable at programming some simple, slightly robust safe guards?

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Kim… Tech report: github.com/MoonshotAI/Kim… Tech blog: kimi.com/blog/kimi-k3

@unclebobmartin @TomWatkins1994 It's more about the search space and ability to find examples. One positive from your experiments is making sure people know how and what to test and the steps needed for verification. That is valuable.

@hinsonan @TomWatkins1994 From now until eternity, this complaint will be raised: "Not fair, it was trained on that." My response: "What hasn't it been trained on?"

Friendly reminder that Uncle Bobs current project is a remake of an old arcade game not critical piece of production software. Not to say it's not impressive what he's doing, but when he tells you that you don't need to read code anymore just bear this in mind.

@unclebobmartin @TomWatkins1994 It's a well understood and studied example though. It's in distribution and there are many great examples for agents to crawl online and find. This is the major issue and why such strict notions of "don't read the code" come off as odd to many

Missile Command was, at one time, a critical piece of production software. In this case I have exposed it to the whole world for review and invited criticism. Notice the similarity between the words "critical" and "criticism". Don't play the game game with me. Every program ever written is a game.

Don't worry I know the world was waiting for this

@divaagurlxw In simpler terms have you tried learning how a forward and backwards pass works?

As an AI Engineer. Please learn >Harness engineering, not just prompt engineering >Context engineering, not just long prompts >Prompt caching vs. semantic caching tradeoffs >KV cache management, eviction, reuse, and memory pressure at scale >Prefill vs. decode latency and why they optimize differently >Continuous batching, paged attention, and throughput optimization >Speculative decoding vs. quantization vs. distillation tradeoffs >INT8, INT4, FP8, AWQ, GPTQ, and when quantization hurts quality >Structured output failures, schema validation, repair loops, and fallback chains >Function calling reliability, tool contracts, argument validation, and idempotency >Agent guardrails, loop budgets, tool budgets, and termination conditions >Model routing, graceful fallback logic, and degraded-mode UX >RAG architecture: chunking, embeddings, hybrid search, reranking, and freshness >Retrieval evals: recall, precision, grounding, attribution, and citation quality >Evals: golden sets, regression tests, adversarial tests, LLM-as-judge, and human evals >LLM observability as a first-class discipline: traces, spans, tokens, latency, errors, and drift >Cost attribution per feature, workflow, tenant, and user journey not just per model >Safety engineering: prompt injection defense, data leakage prevention, and permission boundaries >Multi-tenant isolation, cache safety, and cross-user context contamination prevention >Fine-tuning vs. in-context learning vs. RAG vs. distillation and when each is the wrong tool >Latency, quality, cost, and reliability tradeoffs across the full inference stack >Production failure modes: hallucinated tool calls, malformed JSON, stale retrieval, runaway agents, and silent eval regressions

@mfishbein This seems outlandish

This is my personal software factory. It turns ideas into working products while I sleep. No babysitting coding agents with prompts all day. Cursor and Claude Code made writing code way easier. The harder problem is building a system that can context engineer and manage itself. My factory starts with a Skill called `/factory`. It's the foreman that remembers where the project stands and sends in the right worker for the job. Factory runs this assembly line: 1. `/factory-plan` - the interviewer Reads the existing codebase (if there is one), extracts missing context from me via interview, and writes a product brief. 2. `/factory-plan` - the planner The same skill turns the approved brief into small, testable features and development tasks. 3. `/factory-tests` - the professor Before anyone writes code, every task gets an exam. This skill defines the success criteria for each task, and how the coding agent can prove to itself that what it built works or needs iteration. 4. `/factory-explain` - the presenter Explains the plan to me like I'm 10, with visual metaphor and mermaid charts. Now that coding agents can write more code, faster than any human ever could, the new bottleneck is human understanding of the code. This skill solves that. 5. `/factory-handoff` - from CTO to SWE This packages the brief, plan, tests, safety rails, and stop conditions into one work order. Factory uses the best models for the planning in the previous steps above, then hands the work order to a lower token usage model like Grok 4.5 for execution. 6. Cursor or Claude Code `/loop` - the coffee The night shift picks one task, builds it, takes its exam, records what happened, iterates if needed, then moves onto the next task. If it gets stuck, circuit breakers stop it from confidently digging a deeper hole while I sleep. 7. `/factory-review` - the teacher grades the homework The student doesn't grade it's own homework. A fresh agent that never met the builder tries to break the result. The reviewer rereads the original plan, reruns tests, and finds anything that's broken. 8. `/auto-loom-proof` - shows the evidence Uses browser use and screen records itself performing the tests and adds an 11labs voiceover explaining what's being proven. It sends me the narrated demo video. 9. `/factory-explain` - the code The factory updates a plain-language owner's manual explaining what actually got built. I understand my own codebase, so I can make decisions without becoming the bottleneck or outsourcing my thinking to AI. NOTE ON BUILDING AI Cursor and Claude Code have made writing code dramatically easier. But getting AI to work reliably and at scale for you can't be fully automated. LLM-as-judge helps, but a judge needs a rubric, examples, and input from someone with subject matter expertise. You still need a human reviewing the work and teaching the system how to perform better. You can check out my factory on github in the post below.

Multi billion dollar company with the smartest ai in the world can’t make a valid chart

Developers should be ashamed of themselves for the moronic takes y'all publicly display here. Our field is due for a purging. Too many people who never should have been here to begin with.

@IamAroke Yes the code they write is not good and has a lot of pitfalls. There are times where it's fine to have some slop. But for things that you care about and want to function properly you have to read

Genuinely. Do you think code reviews are still necessary when AI can catch most issues automatically?