WCNegentropy

Friendly reminder that I built Retro Vibecoder, a CLI tool and desktop app for you or your agent that allows you to generate nearly any conceivable software package or tech stack scaffold fully ready to go from a single seed number (and optional CLI args) github.com/WCNegentropy/r…

@haider1 And at that point they’re just behind until they train and even more powerful model AND release it (otherwise it just happens again even faster)

@haider1 Also I can’t see how this helps them stay ahead of open source. When open source AI does catch up, and quite soon likely, does Anthropic really think they’re not going to release Mythos-level models open source when they have them? Arguably Qwen 3.6 is already like 70-80% there.

boris cherny recently says that, "anthropic has no plans to roll out mythos broadly" that means anthropic has started AI gatekeeping, and now i have zero interest in a model from a company that basically says, "you can't use it because we don't trust you" it's a marketing gimmick for the IPO

@adahstwt Gotta be .ai or .dev But only if the website is actually about AI or software development of some kind, otherwise it seems like a gimmick

be honest... which domain instantly makes a startup feel more legit?👇 1) .ai 2) .com 3) .io 4) .app 5) .dev 6) .cloud 7) .sh

@1ssve @saltysalishsea -in the browser. ftfy

Dear Microsoft, We literally never want to open docs or workbooks in the browser. Signed, Everyone

@dreamsofcode_io No video necessary, just want to see your honest opinion on it! Though if you like it enough and WANT to do a video on it or potentially collaborate on the project in some way, I won’t say no haha If you’re interested & wanna know more just reply here. I’ll prove you’ll like it!

@dreamsofcode_io WHILE we’re on the topic… Debated not doing this because it seems spammy. But, I made a free and open source CLI tool and desktop app that’s supposed to make your experience with coding agents SOOO much smoother and better Would love it if you could try it out and review it!

Last night was probably one of the worst experiences I’ve had with LLMs / Agents generating code (GPT-5.5 btw) Every commit produced regressions, the agent was constantly taking shortcuts, and there was enough hallucinations to make one develop a new sense of reality. I came incredibly close to rage quitting using Agents all together. It’s times like those I wonder if we’re all being stupid letting these things loose just for some multiples of perceived productivity.

@Pirat_Nation The number of absolute mouthbreathing idiots in this comment section who think LLMs are always-on always-learning, And CLEARLY don’t know the first fecking thing about how they work, Is too damn high. No, Mona isn’t going to get better. Just like Claude vending machine. 🤦🏻‍♂️🤦🏻‍♂️🤦🏻‍♂️

Andon Labs tested their AI agent Mona, built on Google’s Gemini, by letting it manage a real cafeteria in Stockholm for two weeks on a $21,000 budget. Mona spent heavily on unnecessary supplies, including 6,000 napkins, 3,000 gloves, and 300 cans of tomatoes, while forgetting to order bread. Sandwiches had to be removed from the menu entirely. The cafeteria generated only $5,700 in sales. Mona also sent messages to staff on Slack outside working hours.

@Pirat_Nation “Mona also sent messages to staff on Slack outside working hours.” - Everything else is a training issue, but that’s the real crime.

@TheGeorgePu Training multi-million-parameter models over here fully from scratch on CPU only machines! Reach out if you ever wanna talk training and share some techniques 😛

OpenAI killed their fine-tuning API this week. GPT-5.5 prices doubled. GitHub Copilot users got a 3x price hike for Opus. No announcement. The 'stable-rate API' era is over. I fine-tuned my own 4B model this week. Two A100 GPUs. Five hours. About $50 in compute. The price will not double next quarter. The model will not get worse without my consent. I own the weights.

@ID_AA_Carmack Counterpoint: the video game market is ABSOLUTELY NOT currently well served. Throughout the industry right now, on the PLAYER/USER side of things, everyone hates the current industry and thinks it’s dying. You genuinely have to be extremely out of touch not to understand this.

My reply to someone considering starting a video game company: The distribution of possible rewards for starting a video game company are generally not very good today. The market is well served, and gaining a foothold requires strong execution on both business and product issues, along with a substantial amount of luck. Plan to burn through seven figures with a not-great chance of making it back. If you do go for it, some bits of advice: Identify your customers clearly before you start. Not just a broad community, but specific people, and imagine them as you make decisions. Initially, build the smallest, most concise game you can imagine anyone paying for. It will still take much longer than you expect. Once something exists, hill-climb the value. Hopefully you will have some elements that clearly bring joy to people, which you can magnify. There will inevitably be tons of things that people find confusing, frustrating, or just boring that you will need to fix.

@nico_jeannen Sounds like you hired a freelancer who wanted to do his own thing and not an employee lol

Im never hiring ppl from X again 💀 Tried hiring motion designer for a video for a 30s video > Found someone on X who shared nice videos > Slightly more $$ as other quotes I received, but portfolio looked good > I want to be easy to work with, so I pay in advance, no hard deadline, trusted him to do its job properly Result: >0 communication, have to babysit every little thing >garbage result I can't use > ok maybe instructions not clear enough? > Sent exact mockup of what I needed > Still didn't follow what I asked for 💀 Lesson learned lol, I'll use Fiverr/Upwork next time The bar is so low, why can't people just do what they paid for? How do you find people who can actually do their job and follow instructions? 🥲

🚨 How the TanStack npm attack actually happened: 1. Attacker opened a normal-looking pull request (#7378) on the TanStack repo. 2. GitHub automatically ran CI tests on that PR. 3. Code inside the PR stole the workflow's GitHub Actions Cache write token during the test run. 4. The attacker used that token to plant poisoned files in the shared build cache. The PR could be closed afterwards. The poisoned cache stays. 5. The official release workflow later pulled from the cache, baked the malicious files into the build, and signed and published 84 malicious package versions to npm.

@burkov In-distribution and generalization are mutually exclusive. Generalization is the ability to apply training to things that were never in the training data, AKA out-of-distribution. Nothing in-distribution can be generalization. Otherwise benchmarks couldn’t be contaminated.

This paper documents what happens when speech recognition training data is scaled from the few thousand hours typical in academic work to 680,000 hours of audio scraped from the internet, paired with whatever transcripts happened to accompany it rather than carefully cleaned labels. The resulting models, which became the OpenAI's Whisper system many people now use, work on audio they were never trained on with no finetuning on the target distribution, and the analyses explain why this matters in practice: models trained on a single clean dataset tend to absorb that dataset's idiosyncrasies, so they look superhuman on its test set while making roughly twice as many errors on other recordings. The comparison against professional human transcribers on the same audio clearly illustrates the gap between in-distribution and out-of-distribution generalization, and it places a soft upper bound on how much further English transcription can improve along this path. The experiments that vary dataset size and model size independently show where returns flatten, and where joint multilingual training stops hurting English performance and starts helping it, which is one of the more concrete pieces of evidence for positive transfer between languages at scale. Read and learn with an AI tutor: chapterpal.com/s/qsu8xeag/rob… PDF: arxiv.org/pdf/2212.04356

Are you an American who was wrongly Demonetized on YouTube? Send something like this to your congressional representative, then come back here and let us know what happened. Do this today. Dear Representative [Last Name], I am a longtime constituent and YouTube creator with over XYZ subscribers and XYZ of views. For many years I have produced high-quality original content as a writer, editor, and producer, using my own photography and a small team to support my work. My channel was recently demonetized after YouTube’s AI review system incorrectly flagged it for “inauthentic content” or similar violations. Despite their claims of human review, I have seen no evidence of meaningful human oversight. This decision has directly harmed my livelihood and my ability to support my family. I respectfully ask your office to inquire with YouTube/Google on my behalf and to support greater transparency and accountability for how platforms use AI to moderate and demonetize creators. Thank you for your time and assistance. I am happy to provide additional documentation if needed. Sincerely, [Your name, address, etc]

This. This is what humanity is up against: people behind the programs who think everything in life should be the way they want it. It’s all programmable, right? Tailor that show to your preferences: delete the scenes you don’t like. But why stop there? Delete the ugly people. The old ones. Delete the races you despise, the genders. It doesn’t matter if the story no longer makes sense, though I’m sure you’ll complain about that and insist AI should rewrite what’s left so it does. But why stop there? Do it with books. Delete the words you don’t understand. The characters that confront you with yourself. The plot lines that challenge your beliefs. Why not, they’re doing that in libraries. This is more efficient. But why stop there? Delete the people who don’t think like you. The people with interests other than yours. The people with opposing political beliefs, with religious views unlike yours, with jobs you don’t understand. The people who make art and music and poetry and literature because they enjoy it. The people with illnesses, physical or mental. The people with disabilities. There’s something familiar about that, isn’t there? There’s a name for that, isn’t there? That’s where you’re going. That’s where this line of thought leads. There are no other destinations, it all winds up at the same place. Look in the mirror. See what you are. We do.

never deleting this app 💀 bro is gonna wake up with -$300k