[
  {
    "title": "Google Chrome 'silently' downloads 4GB AI model to your device without permission, report claims — researcher says practice may violate EU law, waste thousands of kilowatts of energy",
    "subreddit": "ArtificialInteligence",
    "score": 311,
    "comments": 45,
    "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1t5k7ng/google_chrome_silently_downloads_4gb_ai_model_to/",
    "selftext": "",
    "created": "2026-05-06T17:45:48.000Z"
  },
  {
    "title": "A Michigan farm town voted down plans for a giant OpenAI-Oracle data center. Weeks later, construction began",
    "subreddit": "ArtificialInteligence",
    "score": 155,
    "comments": 37,
    "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1t5hjiv/a_michigan_farm_town_voted_down_plans_for_a_giant/",
    "selftext": "In Saline Township, Michigan, as in most municipalities, homeowners who want to build a new house know what a complicated and lengthy process it can be: Navigating permit requirements, zoning changes, or variance requests for even a small construction project can take weeks or months. An error in the paperwork, a challenge from a neighbor, or a resistant local official can slow things even further, or kill a project entirely.\n\nSo it surprised many in this agricultural community of red barns and dirt roads that an enormous AI data center—at 21 million square feet, the largest construction project ever undertaken in the state and one almost universally opposed by local residents—seemed to race through the process from application in late summer to groundbreaking in November.\n\nEven more surprising: The $16 billion data center for OpenAI and Oracle’s Stargate AI infrastructure initiative, which will fundamentally reshape the area with its construction, traffic, electricity demand, and environmental impact, was flat-out rejected by both the town’s board and its planning commission in September. But those votes turned out to be only minor bumps on the project’s path: The developer quickly sued, the town settled, and the construction vehicles rolled in.\n\nThe story of how the mega AI data campus became an unstoppable inevitability—over the vocal objection of residents who picketed the vote and posted “no data center” signs outside their homes—reveals a broader dynamic of the nationwide AI data center boom: Once projects of this scale are underway, local governments often have limited leverage to block them.\n\nRead more \\[paywall removed for Redditors\\]: [https://fortune.com/2026/05/06/ai-data-center-michigan-saline-politics-farmland/?utm\\_source=reddit/](https://fortune.com/2026/05/06/ai-data-center-michigan-saline-politics-farmland/?utm_source=reddit/)",
    "created": "2026-05-06T16:14:25.000Z"
  },
  {
    "title": "Those of you who started Algotrading from zero - what do you wish someone had told you on day one? Looking for real, hard-won wisdom (not the generic version)",
    "subreddit": "algotrading",
    "score": 129,
    "comments": 110,
    "url": "https://www.reddit.com/r/algotrading/comments/1t5bh8x/those_of_you_who_started_algotrading_from_zero/",
    "selftext": "I'm asking this coz am seriously considering going deep into algorithmic trading as a career path. I've been doing my own research but I've realized that nothing beats hearing from people who've actually been through it.\n\nAll your advise will be highly valuable to me...",
    "created": "2026-05-06T12:30:17.000Z"
  },
  {
    "title": "It's working!!! Staying on top of the strategy and keeping discipline (with some help)",
    "subreddit": "algotrading",
    "score": 55,
    "comments": 32,
    "url": "https://www.reddit.com/r/algotrading/comments/1t5rm6f/its_working_staying_on_top_of_the_strategy_and/",
    "selftext": "Just wanted to share some progress, along with how I have been able to do it.\n\nDiscipline has been the hardest thing to clamp down over the years. The strategy I have adopted simply works, which is:\n\n1) Identify a strong trend  \n2) Look for a pullback/consolidation  \n3) Enter as soon as trend continues  \n4) Manage risk accordingly\n\nIt's simple, and it works. The hard part was sticking to it, especially in the midst of actually trading and emotions getting in the way.\n\nWhat did I do about it? I put together an automated algorithm (2nd pic) that uses my strategy to signal clean entries and exits. This was my first step to tackle my discipline issue. If the algorithm sees an entry and an exit, I should follow it because it is able to read and make sense of all the data better and faster than I can. After watching it perform, trusting it completely was easy because you can see how well it does in real-time.\n\nThe next thing I did was join a group with like-minded traders for moral support. Anytime I have thought about going against the strategy for any reason, the group is there to keep me in check.  \n  \nTrading is super stressful as it is. Doing it alone adds to that stress. A good supportive group is essential IMHO.\n\nThats it! It's that simple. Build a system, surround yourself with good, supportive, like-minded people who understand what you are going through and are a positive source who can tell it like it is and help you achieve your goals!\n\nAsk any questions you may have!",
    "created": "2026-05-06T22:09:58.000Z"
  },
  {
    "title": "Great, go use hermes if you want, but stfu in this subreddit",
    "subreddit": "openclaw",
    "score": 48,
    "comments": 59,
    "url": "https://www.reddit.com/r/openclaw/comments/1t5j69y/great_go_use_hermes_if_you_want_but_stfu_in_this/",
    "selftext": "please!",
    "created": "2026-05-06T17:10:49.000Z"
  },
  {
    "title": "Why is everyone still using Sharpe ratio?",
    "subreddit": "algotrading",
    "score": 48,
    "comments": 36,
    "url": "https://www.reddit.com/r/algotrading/comments/1t5co3u/why_is_everyone_still_using_sharpe_ratio/",
    "selftext": "I see two clear problems:\n\n1. It assumes a normal distribution, but it’s not uncommon to find fatter tails and skewness.  \n2. It penalizes upwards volatility.\n\nCalmar ratio seems much more appropriate.\n\nWhy still use Sharpe?",
    "created": "2026-05-06T13:17:26.000Z"
  },
  {
    "title": "i tracked every dollar my openclaw agent spent for 2 weeks. heres exactly where the money goes",
    "subreddit": "openclaw",
    "score": 16,
    "comments": 11,
    "url": "https://www.reddit.com/r/openclaw/comments/1t5ojmr/i_tracked_every_dollar_my_openclaw_agent_spent/",
    "selftext": "after almost getting a surprise bill i started logging every interaction by model and task type. ran this for 14 days on my telegram + discord agent\n\nheartbeats (every 30 mins, 672 total)... 38% of my token usage. was running on opus. genuinely insane waste for a status ping\n\nfile reads and summaries... 29% of usage. also on opus. flash handles this identically\n\nactual conversations where model quality mattered... 22% of usage\n\ncomplex tasks where opus was genuinely better than flash... 11% of usage\n\nso 67% of my spend was on tasks where the cheapest model (v4 flash at $0.14/M) would have been identical quality to opus ($6.75/M effective after tokenizer)\n\nthe fix... switch your primary model to deepseek/deepseek-v4-flash in your openclaw.json under agents.defaults.model.primary. then use /model anthropic/claude-opus-4-7 mid-session only when you actually need it for somthing hard. switches instantly, no restart, same session. type /model deepseek/deepseek-v4-flash when youre done with the hard part and go back to cheap\n\nwent from \\~$170/month to about $35 with this approach. the quality difference on heartbeats, file reads, and simple questions is genuinley zero\n\nhonestly the most frustrating part was spending 2 weeks manually logging everything just to find this out. i run my gmail agent on betterclaw free tierwith BYOK and they recently added an update that shows exactly how your api key is spending per task which is genuinley a great update... caught my heartbeat waste there instantly instead of 2 weeks of manual tracking. but yeah switching your primary to flash and /model-ing up to opus only when needed is the move",
    "created": "2026-05-06T20:17:05.000Z"
  },
  {
    "title": "Anthropic launched Claude Security into public beta: it scans your code, finds vulnerabilities, and proposes patches.",
    "subreddit": "ArtificialInteligence",
    "score": 13,
    "comments": 1,
    "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1t62ige/anthropic_launched_claude_security_into_public/",
    "selftext": "Anthropic just pushed Claude Security into public beta for Enterprise customers (Currently Enterprise-only, with Team and Max access coming later). It scans your codebase like a security researcher would: traces data flows across files, understands business logic, finds vulnerabilities that pattern-matching tools miss, and proposes patches you review and approve.\n\nReference: [https://claude.com/product/claude-security](https://claude.com/product/claude-security)\n\n**What it actually does:**\n\n* Parallel scanning of code with multi-file context\n* Adversarial self-verification on every finding to cut false positives\n* Suggested patches that match your existing code style\n* Pushes findings to Slack, Jira, or webhooks\n* Scoped scans (subdirectory level) and scheduled scans\n* Powered by the same models Anthropic uses internally for its own security\n\n**The good:** This is genuinely a leap. Traditional SAST tools drown teams in false positives and miss anything that needs cross-file reasoning. An LLM that actually understands what the code is doing, then writes the fix, is the right shape of tool for the problem. The fact that Anthropic eats its own dog food on this is a real signal.\n\n**The uncomfortable part:** Same capability that finds bugs for defenders finds bugs for attackers. Anthropic published their own research on \"LLM-discovered 0-days\" so they're clearly aware of it. Their bet is that defenders deploying this first creates an asymmetry in favor of the good guys. Maybe.\n\nWhat I keep coming back to though: a successful Claude Security deployment produces a concentrated, validated, well-explained list of exactly where your software is broken. If that list leaks (compromised Slack webhook, an insider, an exported CSV in the wrong S3 bucket), an attacker gets a pre-built attack plan. The product doesn't create new attack surface against random websites, but it does create a very high-value internal artifact that needs to be guarded like crown jewels.\n\nAnyone here from a security team actually trying it? Curious whether the false positive rate holds up in practice and how teams are handling the finding-storage problem.",
    "created": "2026-05-07T06:37:39.000Z"
  }
]