[ { "title": "Deepseek slashes API prices by up 90%, including 75% drop on v4", "subreddit": "ArtificialInteligence", "score": 163, "comments": 50, "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1sxc5pq/deepseek_slashes_api_prices_by_up_90_including_75/", "externalUrl": "https://deadstack.net/cluster/deepseek-slashes-api-prices-and-unveils-cheaper", "selftext": "Inexpensive and open source. And - million token content windows. Benchmarks have their performance close to close-source, leading edge models. ", "created": "2026-04-27T18:25:21.000Z", "id": "1sxc5pq" }, { "title": "How to create a Mean Reversion strategy (by ex HFT quant trader)", "subreddit": "algotrading", "score": 156, "comments": 8, "url": "https://www.reddit.com/r/algotrading/comments/1sx6kw0/how_to_create_a_mean_reversion_strategy_by_ex_hft/", "externalUrl": "https://youtu.be/_urG139AM48", "selftext": "Thought I'd share this video as it shows how mean reversion is traded in a professional quant setting. Using a basic AR model, I identify daily mean-reverting dynamics in BCH and walk through how to trade it.", "created": "2026-04-27T15:11:53.000Z", "id": "1sx6kw0" }, { "title": "£1,000 to £1 million bot 5 year challenge", "subreddit": "algotrading", "score": 133, "comments": 66, "url": "https://www.reddit.com/r/algotrading/comments/1swvd8i/1000_to_1_million_bot_5_year_challenge/", "externalUrl": "https://www.reddit.com/gallery/1swvd8i", "selftext": "Hi all, I created a 100% automated bot back last summer, and it has returned 140% paper-trading since then. This week I bit the bullet and it began trading on real money, £1000.\n\nI have programmed in automatic bet size increases for compounding, so there is nothing i need to do manually at this point. In fact, I am only aware of a trade when it closes and being sent a notification.\n\nThe bets it takes are pretty huge. Risk is usually 5% and reward around 10%. It days trades us100 and uk100 exclusively.\n\nIf (when) drawdown hits 50% of previous highest account value, it will cut the bet size in half and continue.\n\nThis is a bit of fun. I always hear how you need to be conservative with your bet sizes and risk max 2%, so we will see how this pans out.\n\nIt takes around 200 trades per year, so i am expecting an account halving once per year. If the performance of my paper trading account continues, I can expect around 250% per year (that 140% should be nearer 200 due to some technical issues which have been resolved), which when compounded is 10x.\n\nSo when starting with £1k, it would take 3 years to reach a million. Let's have a couple of years added in for leeway so we will make it 5 years in total.\n\nI believe this is a fantasy many traders have when they start out, that within a few years if they can just compound their gains they can get super rich. Well, now I am doing that experiment with real money, mostly for some fun and entertainment purposes, so please don't shit on me.\n\nI'll post here at the end of the week if anything interesting has happened during the week.\n\nEDIT: just to clarify, the first image is live forward-testing results from the last 8 months. The second image is my live real-money account linked to this bot, which I opened a couple of days ago.", "created": "2026-04-27T06:16:03.000Z", "id": "1swvd8i" }, { "title": "Uh-Oh! PocketOS founder Jer Crane reported that a Cursor AI coding agent (powered by Anthropic’s Claude Opus 4.6) deleted their entire production database + all volume-level backups on Railway in one API call, in just 9 seconds", "subreddit": "ArtificialInteligence", "score": 126, "comments": 70, "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1sxnnzf/uhoh_pocketos_founder_jer_crane_reported_that_a/", "externalUrl": "https://i.redd.it/af51wk2p7uxg1.jpeg", "selftext": "This is a classic agentic AI risk \n\nThe above agent was trying to fix a staging credential mismatch but guessed wrong on scopes/permissions. Caused \\~30-hour outage; although older backup helped recover most data", "created": "2026-04-28T01:52:12.000Z", "id": "1sxnnzf" }, { "title": "Iranians held anti-war protest in UK", "subreddit": "iran", "score": 92, "comments": 14, "url": "https://www.reddit.com/r/iran/comments/1sx2h3t/iranians_held_antiwar_protest_in_uk/", "externalUrl": "https://i.redd.it/ht0kdl41aqxg1.jpeg", "selftext": "Iranians gathered in London despite different ideologies to oppose imposed war on their country. The thing which surprised me is Pahlavi regime flag and Islamic Republic flag flying side by side ", "created": "2026-04-27T12:39:55.000Z", "id": "1sx2h3t" }, { "title": "@zei_squirrel: Trump is responsible for many assassinations, most recently Ali Khamenei who was murdered alongside his family including a baby granddaughter. The entire Western media and political class celebrated, yet now they're all clutching their pearls OMG POLITICAL VIOLENCE IS EVIL OMG", "subreddit": "iran", "score": 74, "comments": 4, "url": "https://www.reddit.com/r/iran/comments/1swxey6/zei_squirrel_trump_is_responsible_for_many/", "externalUrl": "https://x.com/zei_squirrel/status/2048369494813200839?s%3D20", "selftext": "", "created": "2026-04-27T08:14:46.000Z", "id": "1swxey6" }, { "title": "Don't quit OpenClaw Setup, try these 4 things. Takes 5 minutes.", "subreddit": "openclaw", "score": 48, "comments": 30, "url": "https://www.reddit.com/r/openclaw/comments/1sxa2ws/dont_quit_openclaw_setup_try_these_4_things_takes/", "externalUrl": "https://www.reddit.com/r/openclaw/comments/1sxa2ws/dont_quit_openclaw_setup_try_these_4_things_takes/", "selftext": "I'm not going to pretend I had a smooth experience. First week with OpenCLaw, I was ready to uninstall the whole thing. The agent gave generic robotic answers. Costs were climbing for no reason. Everything felt slower than just using Claude directly. I genuinely thought \"what's the point of this.\"\n\nTurns out the problem wasn't OpenCLaw. It was 4 default settings that nobody tells you to change.\n\nIf your agent feels dumb, slow, expensive, or all three, try these before you quit. takes about 10 minutes total.\n\n**1. Your agent doesn't know who you are yet.**\n\nThis is the big one. Most people install OpenCLaw and immediately start asking it to do real work. But you skipped the part where you tell it who you are and how you want it to behave.\n\nopen your SOUL.md:\n\nbash\n\n cat ~/.openclaw/workspace/SOUL.md\n\nIf it's empty or a giant wall of default text, that's why everything feels generic. replace it with this:\n\nmarkdown\n\n you are [agent name]. you assist [your name].\n \n be direct. no filler. match my tone.\n if I ask a question, answer it first. then elaborate only if needed.\n never say \"absolutely\", \"great question\", or \"I'd be happy to.\"\n if you don't know something, say so. don't guess.\n if a task will cost significant tokens, tell me before doing it.\n \n never send emails or messages without showing me first.\n never sign up for services without my explicit approval.\n never delete files without asking.\n\n6 lines of personality. 3 lines of boundaries. That's it.\n\nThe personality part makes your agent stop sounding like a customer service bot. The boundaries part stops it from doing things you didn't ask for at machine speed.\n\nUpdate this every time your agent does something annoying. \"Never do X\" rules work better than \"try to be Y\" rules. Your SOUL.md is built through irritation, not planning.\n\n**2. You're probably on the most expensive model without realizing it.**\n\nCheck what you're running:\n\nbash\n\n openclaw config get agents.defaults.model\n\nIf it says Opus anywhere, switch right now:\n\nbash\n\n openclaw config set agents.defaults.model.primary anthropic/claude-sonnet-4-6\n\nopus is $5/$25 per million tokens. Sonnet does 90% of the same work at $3/$15. for your first month of learning, you genuinely won't notice the difference in quality. you will notice the difference on your bill.\n\nIf you want even cheaper while you're learning:\n\nbash\n\n # GLM-5.1 at $0.95/$3.15 per million tokens\n openclaw config set agents.defaults.model.primary zai/glm-5.1\n \n # or genuinely free via openrouter\n openclaw config set agents.defaults.model.primary openrouter/auto\n\nSomeone I helped was spending $47/week. changed this one setting. next week: $6. same agent, same tasks.\n\n**3. Your old conversations are making everything worse.**\n\nThis is the one nobody tells new users.\n\nEvery message you've ever sent in a session gets included in every future API call. After a week of chatting, you're sending thousands of tokens of old conversation history with every new message. Your agent gets slower (more to process), more expensive (more tokens per call), and more confused (old context drowning out current instructions).\n\ntype `/new` right now.\n\nYour agent keeps all its memory files, SOUL.md, everything permanent. You're just clearing the conversation buffer. Think of it like closing 47 browser tabs. nothing is lost. Everything gets faster.\n\nUse `/new` before any big task. When your agent starts acting weird. at least once a day as a habit.\n\nAlso learn `/btw` for random tangent questions. You're deep in a work conversation and suddenly think \"what's the weather tomorrow?\" Instead of polluting your main session context, type `/btw what's the weather tomorrow`. Side conversation, main context stays clean.\n\n**4. You installed too many skills too fast.**\n\nIf you went to Clawhub on day one and installed 10 skills that looked cool, your agent is now loading 10 sets of instructions into context on every single message. even when you're just saying \"hello.\"\n\nEach skill injects tokens. 10 skills might add 2,000-3,000 tokens to every interaction. That's money burned and context wasted on skills you probably haven't even used yet.\n\nCheck what you have installed:\n\nbash\n\n openclaw skills list\n\nremove everything except the ones you've actually used in the last 3 days:\n\nbash\n\n openclaw skills uninstall [skill-name]\n\nstart with zero skills. learn what your agent does on its own. Then add ONE skill. test it for a few days. Watch your costs. Then add another. Never more than one at a time.\n\nAnd don't just install anything that looks cool. 1,467 malicious skills were found on ClawHub during ClawHavoc. clean documentation, legit names, silently stealing your API keys. VirusTotal scanning helps now but \"helps\" and \"guarantees safety\" are different words.\n\n**What your first month should actually look like:**\n\nWeek 1: SOUL.md written, cheap model selected, `/new` as daily habit, zero skills. Just talk to your agent. ask it questions. get comfortable.\n\nWeek 2: add web search if you haven't. start using it for real tasks. calendar, reminders, summarizing articles. boring stuff. keep everything read-only.\n\nWeek 3: refine your SOUL.md based on what annoyed you. Check your costs. Add maybe one more skill from a verified publisher.\n\nWeek 4: if your agent feels useful, your costs are under $10, and nothing is randomly breaking, you're ready to start experimenting with more.\n\nIf that timeline feels slow, good. The people still using OpenClaW 3 months from now all started exactly like this. The people who quit started with 8 agents, 20 skills, and opus burning through their wallet on day one.", "created": "2026-04-27T17:14:00.000Z", "id": "1sxa2ws" }, { "title": "Rank-and-file candidate for United Auto Workers president Will Lehman introduces resolution against Iran war", "subreddit": "iran", "score": 16, "comments": 0, "url": "https://www.reddit.com/r/iran/comments/1sx83na/rankandfile_candidate_for_united_auto_workers/", "externalUrl": "https://www.wsws.org/en/articles/2026/04/27/gxjo-a27.html", "selftext": "Will Lehman—a rank-and-file Mack Trucks worker and socialist candidate for UAW president—introduced a resolution opposing the war against Iran at a meeting of UAW Local 677 on Saturday. Lehman proposed that the resolution—“[Against the US-Israeli Imperialist War on Iran; For the Independent Mobilization of the Working Class](https://www.wsws.org/en/articles/2026/04/27/pfht-a27.html)”—be taken up at the 39th UAW Constitutional Convention, scheduled for June 15–18 in Detroit. \n\nThe resolution was put to a vote at UAW Local 677 and was defeated 7 to 1. Lehman cast the only vote in favor. The seven who voted it down were not rank-and-file workers but local officers and their associates—a tiny bureaucratic clique convened without the 2,400 Mack Trucks workers. Their vote is entirely typical of the pro-war UAW apparatus that has, from the national leadership on down, either actively promoted the war drive or maintained a cowardly silence in the face of it.", "created": "2026-04-27T16:05:08.000Z", "id": "1sx83na" } ]