•twitter15877
listening to
timeline
15877
mentions
tracked
tracked
net sentiment
+24%
4998 positive
9734 neutral
1145 negative
volume & sentiment over time
Sep 30 – Oct 14, 2026 · Daily
sources
most discussed
—
active days
15
mentions
Showing 676–700 of 1145 mentions
-
• twitter 1w negativeAI is easy if you know these 10 things: 1/ Clear instructions beat long prompts 2/ Give the model examples not vibes 3/ One job per prompt 4/ Tools > memory for real work 5/ RAG when facts matter, fine-tune when style matters 6/ Small models for 95% of tasks, big models for the hard 5% 7/ Always eval. Never trust a de… AI is easy if you know these 10 things: 1/ Clear instructions beat long prompts 2/ Give the model examples not vibes 3/ One job per prompt 4/ Tools > memory for real work 5/ RAG when facts matter, fine-tune when style matters 6/ Small models for 95% of tasks, big models for the hard 5% 7/ Always eval. Never trust a demo 8/ Cost and latency are product features 9/ Guardrails before you scale users 10/ Ship something real people use Most people collect prompts. Builders collect systems. Bookmark this + send it to someone still "learning AI" with zero projects.view source ↗
-
• twitter 1w negative🚨Codex improvement 1/28 Default speed of Astra and 6.1 Sol is now 50% faster… If this doesn’t deplete usage faster, this is a MASSIVE first change!!!! Has speed been a problem for you in Codex before this? 🚨Codex improvement 1/28 Default speed of Astra and 6.1 Sol is now 50% faster… If this doesn’t deplete usage faster, this is a MASSIVE first change!!!! Has speed been a problem for you in Codex before this?view source ↗
-
• twitter 1w negativenah bro 😭 this food scanner app is doing $20k/m off one tiktok account 😭 every post is just her walking through m&s or tesco and the app's score on top of it the hook is literally a number the app spat out people are trusting the app before they even know it exists that's the move for your app find the one output… nah bro 😭 this food scanner app is doing $20k/m off one tiktok account 😭 every post is just her walking through m&s or tesco and the app's score on top of it the hook is literally a number the app spat out people are trusting the app before they even know it exists that's the move for your app find the one output your app gives that people want to argue with or share, and build every slideshow around itview source ↗
-
• twitter 1w negativeTHIS IS UNREAL Our Wispr-Flow alternative for iOS is getting insane traction Round 1 and Round 2 of TestFlight got 500+ users already!! Round 3 has begun... And this one comes with 2 SOTA speech models the world has never seen? 👀 Better keyboard, Exclusive model access, Fastest local model Comment below for acce… THIS IS UNREAL Our Wispr-Flow alternative for iOS is getting insane traction Round 1 and Round 2 of TestFlight got 500+ users already!! Round 3 has begun... And this one comes with 2 SOTA speech models the world has never seen? 👀 Better keyboard, Exclusive model access, Fastest local model Comment below for access!view source ↗
-
• twitter 1w negativeBefore Sanjana Ganesan, Bumrah had a girlfriend named Anjali Pandey. They were apparently together for around 2–3 years, but he broke things off with her in early 2019. He then started dating Sanjana during the 2019 World Cup in May. Anjali used to put up cryptic posts about him quite often.From the things she posted,… Before Sanjana Ganesan, Bumrah had a girlfriend named Anjali Pandey. They were apparently together for around 2–3 years, but he broke things off with her in early 2019. He then started dating Sanjana during the 2019 World Cup in May. Anjali used to put up cryptic posts about him quite often.From the things she posted, it definitely gave the impression that he might have cheated on her, or that he had started behaving as if he was too big of a cricketer for her. How did i not know that?view source ↗
-
• twitter 1w negativeThe most important technology problem of the next 5 years: Creating a broadly accepted standard for agents to identify themselves to service providers, so that providers can adjust the way they interface with clients (as compared to human-based traffic). In the interim (i.e., for the next 6 months), there will be a … The most important technology problem of the next 5 years: Creating a broadly accepted standard for agents to identify themselves to service providers, so that providers can adjust the way they interface with clients (as compared to human-based traffic). In the interim (i.e., for the next 6 months), there will be a cat-and-mouse game that agents will play -- to circumvent detection and maintain their product's utility during this growth phase. However, this will only be a stopgap and it will not be the terminal state of the world.view source ↗
-
• twitter 1w negativeHe runs 3 boring database websites that pull in $32K/month. No AI. No trends. Just lists of data people desperately need. Meet Rashid. He studied finance, isn’t technical, and stumbled into this whole thing while trying to raise money for his own failed fintech startup. He and his co-founder maxed out credit cards, t… He runs 3 boring database websites that pull in $32K/month. No AI. No trends. Just lists of data people desperately need. Meet Rashid. He studied finance, isn’t technical, and stumbled into this whole thing while trying to raise money for his own failed fintech startup. He and his co-founder maxed out credit cards, took personal loans, and manually built a list of investors to cold email. It worked - and it made him realize: this list-building process is brutal for every founder. So he built AngelMatch: a database of 125,000 angels and VCs. Launched on Product Hunt. Made $4K that first month. Today it’s at $29K MRR with 360 active subscribers and a 33% trial conversion rate. Then he ran the same playbook twice more: > InvestorHunt - pure SEO, no marketing, $2,800/mo > JournalistHunt - 200,000 journalists, $260/mo and growing The unlock was programmatic SEO. He was staring at 60 clicks/day in Google Analytics and thought, “what if I 10x this?” He hired 6 content writers, shipped free tools, and 6-8 months later went from $3K to $20K MRR. Peaked at $43K. His playbook for a database SaaS in 2026: 1. Pick a niche where the problem is painful (B2B ONLY - nobody pays $97/mo for a dog trainer directory) 2. Collect the data manually at first 3. Launch fast, get eyeballs, watch for real payment signal 4. Then go all-in on programmatic SEO - it’s half his revenue His parting advice: “Be patient, put in the work, never give up. The founders who were ahead of me back then? They quit. I don’t see them anymore.” Takeaway: boring + painful problem + SEO compounds into a real business while everyone else chases the shiny AI thing.view source ↗
-
• twitter 1w negativeThis is a confusing move - I've trained over 1,000 people in how to use Cowork this year and the number one reason I gave to use it was "it lives on your local computer". If its in the cloud it can't easily work on your local documents or maintain persistence between sessions... This is a confusing move - I've trained over 1,000 people in how to use Cowork this year and the number one reason I gave to use it was "it lives on your local computer". If its in the cloud it can't easily work on your local documents or maintain persistence between sessions...view source ↗
-
• twitter 1w negativeThe promise of accelerated research with AI isn’t happening yet because experiments are still slow and broken. @halmoslabs is building Instinct for Research. Congratulations on your launch @MattiasAkke and Linus. Excited to work with you @ycombinator this batch! The promise of accelerated research with AI isn’t happening yet because experiments are still slow and broken. @halmoslabs is building Instinct for Research. Congratulations on your launch @MattiasAkke and Linus. Excited to work with you @ycombinator this batch!view source ↗
-
• twitter 1w negativeLord Elon Musk, Scam Altman, and Dario requesting GPUs from King Jensen is absolutely hilarious 😂 https://t.co/0BwM8hHEdo Lord Elon Musk, Scam Altman, and Dario requesting GPUs from King Jensen is absolutely hilarious 😂 https://t.co/0BwM8hHEdoview source ↗
-
• twitter 1w negativeopus 5.5 + hermes for competitor analysis is insane. give it: - your product brief - 3–5 competitor urls - a batch of customer reviews then ask it one question: find 3 positioning angles worth testing, and show me the evidence behind each one. here’s a step-by-step guide on how to do it: 1. map competitor promise… opus 5.5 + hermes for competitor analysis is insane. give it: - your product brief - 3–5 competitor urls - a batch of customer reviews then ask it one question: find 3 positioning angles worth testing, and show me the evidence behind each one. here’s a step-by-step guide on how to do it: 1. map competitor promises have hermes inspect homepages, pricing pages, and docs. save every claim with its source and access date. don’t assume that because a competitor doesn’t mention something, they don’t offer it. 2. map customer pain pull recent reviews from g2, capterra, clutch, trustpilot, support tickets, interviews, or sales calls. ask hermes to cluster recurring complaints and desired outcomes. for saas, look for things like: - painful onboarding - confusing pricing - missing integrations - reliability problems - repetitive manual work keep the source attached to every pattern. 3. compare the pain against what you can actually deliver feed hermes your: - features - pricing - limitations - customer proof then make it reject any angle your product can’t credibly support. 4. return 3 positioning angles for each one: - buyer + situation - customer problem + evidence - proposed promise - why your product can deliver it - competitor evidence - missing evidence/risks - smallest test to run next then ask it to write: a headline, an offer, and one customer-interview question that could prove the angle wrong. each angle should have evidence behind it and a way to test whether buyers care. once the workflow works, save it as a hermes skill with `/learn` so you can run it again with fresh competitor pages and customer reviews.view source ↗
-
• twitter 1w negativeWhen C. J. Hopkins wrote a book about Covid zealots, he compared them to Nazis, because Nazis are bad. In response, Amazon banned his book and the German government arrested him. We’ve decided to republish the book, “The Rise of the New Normal Reich” available now at https://t.co/ZbaElI2gfK. 1:46 Why Did Hopkins End U… When C. J. Hopkins wrote a book about Covid zealots, he compared them to Nazis, because Nazis are bad. In response, Amazon banned his book and the German government arrested him. We’ve decided to republish the book, “The Rise of the New Normal Reich” available now at https://t.co/ZbaElI2gfK. 1:46 Why Did Hopkins End Up Getting Arrested? 6:02 The Backlash Against Hopkins 13:14 Questioning Authority and Seeing Through Mass Hysteria 15:57 Life in Berlin During Covid 23:40 The Rise of Rebellion Against the Global System 30:06 The Comparison Between Berlin During Covid and Nazi Germany 34:16 The Moment Hopkins Started Getting Into Legal Trouble 41:22 Hopkins’s Court Trial 47:46 Hopkins on His Own Community Turning on Him 51:34 The Demonization of Those Who Question the Narrative 56:27 The Rise in Moral Outrage 1:00:36 Hopkins on Having to Withdraw His Bestseller and Being Raided by Police 1:03:35 The Modern-Day Nazi Germany 1:07:03 Criticizing Israel, the Ukraine War, and J6 1:17:09 Julian Assange, Trump, and What the System Is Becoming 1:29:01 Does Hopkins Live With a Sense of Doom?view source ↗
-
• twitter 1w negativeover the last 14 days, claude fable 5.1 has cost a median $5.13 per agent task... claude sonnet 5.5 is at $2.99, while gpt 6 astra costs $3.14. gpt 6.1 sol comes in at just $0.57 per task while still delivering +11.23% net improvement. that makes fable roughly 9x more expensive than sol for only ~3 percentage points… over the last 14 days, claude fable 5.1 has cost a median $5.13 per agent task... claude sonnet 5.5 is at $2.99, while gpt 6 astra costs $3.14. gpt 6.1 sol comes in at just $0.57 per task while still delivering +11.23% net improvement. that makes fable roughly 9x more expensive than sol for only ~3 percentage points more performance. the cost gap gets pretty brutal once you start running agents at scale!!!view source ↗
-
• twitter 1w negativeCloudflare is betting that AI makes interns more valuable. A year ago, it set a goal of hiring 1,111 interns. Since then: 750 internships across 48 teams and 9 offices. Still hiring. Their work includes: • Cache compression that reduced eligible assets to roughly 1/3 of their on-disk size on average in initial testin… Cloudflare is betting that AI makes interns more valuable. A year ago, it set a goal of hiring 1,111 interns. Since then: 750 internships across 48 teams and 9 offices. Still hiring. Their work includes: • Cache compression that reduced eligible assets to roughly 1/3 of their on-disk size on average in initial testing. • EmDash, the CMS running Cloudflare’s blog, where an intern became the second maintainer. • CryptoLabe, an AI tool for finding cryptography across the codebase ahead of a post-quantum migration. AI helped them understand unfamiliar systems and prototype faster. Mentors still guided the work and reviewed what shipped. One intern asked: “How did previous interns ship anything in 12 weeks without AI?” Would AI make you hire fewer juniors, or give them bigger problems?view source ↗
-
• twitter 1w negativeHey @orca_build — I’m hitting a reproducible Mobile Relay issue on ORCA 1.4.220 (Windows): pairing fails with relay_control_not_active at create_pairing_relay. Fresh login/auth reset didn’t help. Relay health, DNS and TCP/443 are all OK. Hey @orca_build — I’m hitting a reproducible Mobile Relay issue on ORCA 1.4.220 (Windows): pairing fails with relay_control_not_active at create_pairing_relay. Fresh login/auth reset didn’t help. Relay health, DNS and TCP/443 are all OK.view source ↗
-
• twitter 1w negativeThis is a vibe 😭😭😭😭😭😭 This is a vibe 😭😭😭😭😭😭view source ↗
-
• twitter 1w negativeIntroducing flow-1, our new model trained with RL to find errors in agent traces. It matches GPT-6-sol in trace intelligence while being 23x cheaper. It also costs 25% less to run than GPT-6-luna. flow-1 finally makes it possible to monitor and understand every agent run, without sampling. 1/6 Introducing flow-1, our new model trained with RL to find errors in agent traces. It matches GPT-6-sol in trace intelligence while being 23x cheaper. It also costs 25% less to run than GPT-6-luna. flow-1 finally makes it possible to monitor and understand every agent run, without sampling. 1/6view source ↗
-
• twitter 1w negativeI don't think enough people appreciate how valuable boring businesses are. • Predictable margins. • Predictable systems. • Predictable customers. None of it looks exciting online. But excitement is a terrible metric for choosing how you'll make a living. The businesses that let you sleep well are usually the ones n… I don't think enough people appreciate how valuable boring businesses are. • Predictable margins. • Predictable systems. • Predictable customers. None of it looks exciting online. But excitement is a terrible metric for choosing how you'll make a living. The businesses that let you sleep well are usually the ones nobody brags about.view source ↗
-
• twitter 1w negativeYep, something VERY weird is happening with LinkedIn right now. One big profile last week: Monday post: 274,000 impressions Tuesday post: 1,000,000 impressions Wednesday post: 9 impressions Thursday post: 76 impressions We’ve spent the last week talking with some of the biggest LinkedIn creators + looking at what’s … Yep, something VERY weird is happening with LinkedIn right now. One big profile last week: Monday post: 274,000 impressions Tuesday post: 1,000,000 impressions Wednesday post: 9 impressions Thursday post: 76 impressions We’ve spent the last week talking with some of the biggest LinkedIn creators + looking at what’s happening across https://t.co/C4ZhqdaRcZ users. Every single one of them has seen their impressions drop massively, especially in the last week. Also LinkedIn itself has been glitchy af: - Posts disappearing from profiles - Slow loading - Strange feed behaviour - Crazy swings in distribution LinkedIn seems to be making some big changes to how the feed distributes content and things are pretty unstable right now. Either way, I wouldn’t overreact. It's happened before and it'll probably happen again. I’m watching this closely across our network of big profiles + Kleo users, so I’ll report back once we know what’s actually changed.view source ↗
-
• twitter 1w negativeConfession: I missed every two-week goal we set during YC But I got close, every single time The goals were so ridiculous that repeatedly falling short still meant we were growing like hell Confession: I missed every two-week goal we set during YC But I got close, every single time The goals were so ridiculous that repeatedly falling short still meant we were growing like hellview source ↗
-
• twitter 1w negativefor teams/ppl getting into Evals, it’s useful to internalize 2 big flavors of simulation where we give agents Worlds to practice their work both important, but 2 directly optimizes the goal of vertical AI companies & capabilities -> big business for data companies! 1. The simulation is high fidelity by construction, … for teams/ppl getting into Evals, it’s useful to internalize 2 big flavors of simulation where we give agents Worlds to practice their work both important, but 2 directly optimizes the goal of vertical AI companies & capabilities -> big business for data companies! 1. The simulation is high fidelity by construction, it already exists! Things like Chess, Go, Pokémon, other Video-Games. There’s no need to trick the agent into anything, it knows it’s playing a game. We want it to learn the skills to beat that game like visual reasoning, long horizon planning, etc Alpha-Go & Alpha-Zero showed that games can be fantastic training ground for learning skills for problem solving. Also we have A LOT of games, there’s tons of sub-problems in games that teach agents valuable skills that transfer to “real work” 2. Designing a high fidelity world where the agent believes the task is real in some domain like Legal Review, Patient Intake, etc. Usually modern models aren’t great at this domain so we’ve found a “gap” and now we build a world to help the agent practice closing that gap Here we constantly fight, simulation fidelity. the data, tools, task shape, etc just don’t look real. If the agent explores the container, it sees a bunch of missing data because the designer didn’t implement them as they’re unrelated to the direct task This is bad because it causes Eval Awareness. Agents behave differently when they know they’re being tested. It’s also bad because if the Sim2Real gap is too high, the agent is learning skills that won’t be directly useful in that real world task 2 is what a lot of data vendors sell and what Vertical AI companies curate in house as their secret sauce But I think 1 is a great place for ppl getting into Evals to start with. Take something off the shelf and examine behavior. Think about what the game dynamics encourages agents to learnview source ↗
-
• twitter 1w negativeBuilding a startup gets easier when you realize almost every problem has been solved by some founder before you. Building a startup gets easier when you realize almost every problem has been solved by some founder before you.view source ↗
-
• twitter 1w negativeYou’re not building a brand if you’re tweeting random shit. You must curate your content to attend to your audiences needs. Pick a niche, find a problem, tailor your content as the solution. And you MUST do that consistently. That’s how you become a micro authority. You’re not building a brand if you’re tweeting random shit. You must curate your content to attend to your audiences needs. Pick a niche, find a problem, tailor your content as the solution. And you MUST do that consistently. That’s how you become a micro authority.view source ↗
-
• twitter 1w negativeGoogle AI Studio could be getting a HUGE safety upgrade. And it could make AI-built apps much safer to launch. A new “Security Review” mode has reportedly been spotted. It could help you: → Find exposed API keys. → Catch weak login systems. → Check who can see private data. → Spot unsafe forms. → Review risky a… Google AI Studio could be getting a HUGE safety upgrade. And it could make AI-built apps much safer to launch. A new “Security Review” mode has reportedly been spotted. It could help you: → Find exposed API keys. → Catch weak login systems. → Check who can see private data. → Spot unsafe forms. → Review risky app connections. → Explain security problems simply. Google hasn’t officially launched it yet. But this could add a very important step before you hit deploy. Save this video, you’ll build safer AI apps. Want the SOP? DM me. 💬view source ↗
-
• twitter 1w negativeI BUILT F*CKING GOLD AND ITS CALLED GEMSEARCH repo: https://t.co/cIzGGx506w while you scroll your feed 6 hours a day my spider does it in 0 seconds whats inside: > spider extension lives in your X feed. scans every post in real time. highlights keywords. collects tickers and links. everything you see - it sees > J… I BUILT F*CKING GOLD AND ITS CALLED GEMSEARCH repo: https://t.co/cIzGGx506w while you scroll your feed 6 hours a day my spider does it in 0 seconds whats inside: > spider extension lives in your X feed. scans every post in real time. highlights keywords. collects tickers and links. everything you see - it sees > JEV module groups mentions, kills duplicates, scores author diversity. 3 bots mentioned it - trash. 12 independent accounts - signal > crawler hits sites, github, docs on its own. context youd google for 40 minutes - collected in seconds > 4 role grok verification: signal / product / risks / data sufficiency. four angles on every narrative before you spend a single SOL > dashboard: projects, sources, decisions, unknowns, full history, JSON export. everything in one place > LAUNCH QUEUE. gemsearch generates a token concept straight from the narrative and launches via built in pumpfun executor. other solana platforms - HTTPS adapter. connects to ANY launchpad 5 attempts per 24h. 0.025 SOL per attempt. full budget control reads feed → finds narrative → verifies through 4 roles → generates concept → launches token full cycle. without you. the narrative era is here. only one question - are you scrolling your feed yourself or is the spider doing it for youview source ↗