X (Twitter) threads examples

BREAKING: Apple just proved AI "reasoning" models like Claude, DeepSeek-R1, and o3-mini don't actually reason at all. They just memorize patterns really well. Here's what Apple discovered: (hint: we're not as close to AGI as the hype suggests) https://t.co/9A4Lg5aQWO Instead of using the same old math tests that AI companies love to brag about, Apple created fresh puzzle games. They tested Claude Thinking, DeepSeek-R1, and o3-mini on problems these models had never seen before. The result ↓ All "reasoning" models hit a complexity wall where they completely collapse to 0% accuracy. No matter how much computing power you give them, they can't solve harder problems. https://t.co/nQJWpOjxBz As problems got harder, these "thinking" models actually started thinking less. They used fewer tokens and gave up faster, despite having unlimited budget. Apple researchers even tried giving the models the exact solution algorithm. Like handing someone step-by-step instructions to bake a cake. The models still failed at the same complexity points. They can't even follow directions consistently. The research revealed three regimes: • Low complexity: Regular models actually win • Medium complexity: "Thinking" models show some advantage • High complexity: Everything breaks down completely Most problems fall into that third category. https://t.co/EvkTg4DPq6 Apple discovered that these models are not reasoning at all, but instead doing sophisticated pattern matching that works great until patterns become too complex. Then they fall apart like a house of cards. If these models were truly "reasoning," they should get better with more compute and clearer instructions. Instead, they hit hard walls and start giving up. Is that intelligence or memorization hitting its limits? This research suggests we're not as close to AGI as the hype suggests. Current "reasoning" breakthroughs may be hitting fundamental walls that can't be solved by just adding more data or compute. Models could handle 100+ moves in Tower of Hanoi puzzles but failed after just 4 moves in River Crossing puzzles. This suggests they memorized Tower of Hanoi solutions during training but can't actually reason. https://t.co/g8VOH7gzOs While AI companies celebrate their models "thinking," Apple basically said "Everyone's celebrating fake reasoning." The industry is chasing metrics that don't measure actual intelligence. Apple's researchers used controllable puzzle environments specifically because: • They avoid data contamination • They require pure logical reasoning • They can scale complexity precisely • They reveal where models actually break Smart experimental design if you ask me. What do you think? Is Apple just "coping" because they've been outpaced in AI developments over the past two years? Or is Apple correct? Comment below and I'll respond to all. If you found this thread valuable: 1. Follow me @rubenhassid for more threads around what's happening around AI and it's implications. 2. RT the first tweet

Wan 2.2 Animate Lip syncing Test. Definitely way better than Runway Act2 in my opinion. Takes about 8 minutes for a HD video to be processed at 720p vertical (reels style) https://t.co/OEo8MRwMJY @NoiseesoiN Rtx 6000 pro @Dork_sense 15 sec for 8 minutes. Now available inside➡️ https://t.co/ntZmJxY3p8 📷🫡 https://t.co/RM79GoXDB0

LinkedIn has 310 million active users. But 99% of people don't know how to use it. Here are 5 powerful ways you can use it: 1. Steal your competitor's audience - Use the search bar - Input your target persona (e.g. CEO) - Go to ‘all filters’ - Insert your ‘competitors name’ in the ‘followers of’ field - Click on ‘show results’ Save time on scrolling and searching, find who you need in 60s https://t.co/FbvfpjLtAK 2. Showcase your services LinkedIn lets you add services to your profile Here’s how to enable it: 1/ Visit your profile 2/ Click on "Open to" and then "Providing Services" 3/ Fill in the form https://t.co/gAvTxGUMI6 3. Soft-sell your audience Use your LinkedIn profile as a landing page. Showcase: - A free offer (newsletter) - A low ticket offer (Ebook) - Higher ticket offer (Main offer) Make it easy for people to buy from you (without you even asking). https://t.co/vDSPFJ2T7i 4. Send personalised requests Unlike all other socials, on LinkedIn you can ‘connect’ with others. A good personalised connection request can make 10x the impact. Here’s the template that has a 99% acceptance rate for me: [im https://t.co/nVPjBboF7R 5. Community building Turning followers into fans is the best long term play. LinkedIn is a networking app, use it as such. - Comment on peoples content - Add value - Connect - Have fun Make real connections. I started on LinkedIn in 2022 today I have: - 68K followers - 6-figure biz - A loyal community Want the same? Book a power hour with me and start building your personal brand too: https://t.co/qmHJvPDi4e https://t.co/xJguYa3yOM

BREAKING: Stanford just surveyed 1,500 workers and AI experts about which jobs AI will actually replace and automate. Turns out, we've been building AI for all the WRONG jobs. Here's what they discovered: (hint: the "AI takeover" is happening backwards) https://t.co/2nX8UMdaBh → Workers don't want full automation. They want partnership. 46.1% of tasks got positive automation ratings, but here's the kicker: Workers prefer "H3 equal partnership" with AI over full replacement. https://t.co/RTkxmNR4iF → Scientists divided AI adoption into 4 zones based on worker desire vs. technical capability. 41% of startup investments are going to the WRONG zones, i.e. areas workers don't want automated or that have low capability. https://t.co/WqrbBZKL1x → High-paying "analyzing information" jobs are becoming less valuable. "Interpersonal communication" skills are skyrocketing in importance. It's the complete opposite of what every expert predicted. https://t.co/xWIq2cCY7J → The startup problem everyone's ignoring is that Y Combinator companies are flooding the "Low Priority" and "Red Light" zones while ignoring the "Green Light" zone where workers are BEGGING for help. The market is building what sounds cool, not what people need. https://t.co/H9gErtZ0sL → Comparison with usage data from Claude reveals the real problem ↓ The top 10 jobs that want automation most represent only 1.26% of actual AI usage. We're solving the wrong problems while the biggest pain points get ignored. https://t.co/5FfoistvSe → Creative workers drew a hard line in the sand. Only 17% of creative tasks got positive automation ratings. Workers are saying: "automate my boring stuff, not my creative work." The message is crystal clear, but tech isn't listening. https://t.co/o1njp5ODZq Art directors, editors, and designers don't want AI replacing creativity, they want it handling admin work. "I want workflow maximized, not content creation," said one art director. Tech is backwards. → 45.2% of occupations want H3 partnership over full automation. This reveals a massive opportunity in human-AI collaboration that everyone's ignoring. The future is teamwork, not replacement. https://t.co/ss6vfcJGMi → The top most common fears that workers have about AI automation in their work ↓ The trust problem is bigger than the job problem. https://t.co/JFQf2lWk6R → The wage reversal is already happening. Skills ranked by current wages vs. future human agency needs show a massive shift: • "Analyzing data" dropping down • "Training others" rising up • "Interpersonal skills" becoming premium https://t.co/nFeycEKS0j → In audio interviews, Stanford found that when workers describe their ideal AI future, 23% want "role-based support", i.e. AI that acts like a specialized teammate, not a replacement. "Set up my AI assistant" not "replace me with AI." https://t.co/VDcrs5vDeN → Here's what workers actually said they want: • "Automate the task so I can focus on high-value work" - 69.38% • "Handle the repetitive stuff" - 46.6% • "Reduce stress and mental drain" - 25.5% Freedom to do meaningful work, not unemployment. https://t.co/mlisecjpQ0 The "AI takeover" narrative is backwards. Workers want AI teammates, not replacements. The real opportunity is collaboration, not elimination. Trust matters more than capability. I believe it's time to build what people actually want. The future of work isn't about humans vs. AI. It's about humans WITH AI.

BIG NEWS: I sold Backlinko https://t.co/8sDWQrgK8B I also want to say a big "thank you!" to everyone who supported me over the years. Growing Backlinko has been one of the proudest achievements of my life. And it wouldn't have been possible without your support. Thank you for that 🙏🏻🙏🏻🙏🏻!!!

I love whatever is wrong with this viral AI companion named Merv https://t.co/MKdwfF9WU4 Imagine entering the AI companion lottery expecting a waifu and you end up with this guy (Though tbh the dance moves impressed me) https://t.co/kgapyQHu4A

In the next version of Claude Code, Claude’s WebFetch tool automatically adds Accept: “text/markdown, *” to requests which helps docs sites provide token-efficient docs https://t.co/uYthd1m9RP @0xRaduan Summarization is one thing we do to reduce prompt injection risk. Are you running into specific issues with it?

Mark Rober posted a video ripping @Tesla’s FSD/AutoPilot technology. The problem? FSD was never engaged, and nothing was on when he crashed into the wall. The whole video was designed to bash Tesla and promote Luminar, a LiDAR company. https://t.co/WbEdKt5nIi @oulu420 @Tesla Because the screen shows it’s not turned on at all. https://t.co/e1gwtaIdFk Are you wondering why Luminar would want the public to distrust FSD/Autopilot? Because if they convince enough people that Teslas aren’t safe, they can push for regulations on autonomous vehicles requiring LiDAR. This would increase their value while setting Tesla back https://t.co/H6qDQ0dUWm @elonmusk you should sue. @elonmusk More on these tests from @techAU: @elonmusk @techAU @SawyerMerritt Luminar is no longer promoting Mark Rober’s video on their website. https://t.co/4D4HYLQ94e

1. Mercedes-Benz, 2... The 10 best Christmas ads of all time: 1. Mercedes-Benz, 2013 https://t.co/ce6rjPlV3h 2. Jordan Brand, 1985 https://t.co/BnXgzTc0hm 3. Adidas, 2012 https://t.co/mD7ZqgKATL 4. Durex, 2016 https://t.co/ZaHh2N5ePK 5. Pepsi, 2016 https://t.co/oDgMqIzS1f 6. British Airways, 2007 https://t.co/taFmByENsB 7. M&M, 2013 https://t.co/bOD8C9GdpS 8. Heineken, 2009 https://t.co/B9OCYyX7F2 9. McDonalds, https://t.co/gamDpTzwGg 10. Tide, 2017 https://t.co/EQmWTRVWRc

This is Mark Rober He spent years building one of the strongest reputations in social media. Loved by tens of millions of followers Yesterday he obliterated it all with one 18 minute video It's a story of greed, deception, and cowardice Here's the story of what happened: https://t.co/FQTl0FAtrb Mark is a long time Youtuber who spent years accruing over 65 million subscribers He filmed wildly entertaining videos testing the limits of physics and tech He was Mr. Beast BEFORE there was a Mr. Beast Yesterday he let money, politics, and greed get in the way of all that https://t.co/OG2St5nXWn Mark put out a video titled "Can You Fool A Self Driving Car?" The concept of the video is simple. Drive a Tesla in Full Self Driving mode on a road where there is a wall painted like a road See if the Tesla cameras see the wall and cause the car to stop Harmless enough right? https://t.co/e1X4anl7CV Wrong. The video was filled with deception, lies, and illegal undisclosed advertising In the video the Tesla does not detect the wall and goes right through it On top of that there is a kid mannequin that gets absolutely nailed But there's a few things Mark doesn't tell you https://t.co/Tcuzgnyixu First of all, he lied about Full Self Driving. He didn't even use FSD. He used Autopilot, which even Tesla admits is an older, not as strong technology But even worse, he deactivates it right before he hits the wall. In a sneaky maneuver, he moves the wheel to deactivate https://t.co/QyX0EonGux But why would he so purposely slander and attack Tesla like this? O yeah, he is working with a Lidar company which he ILLEGALLY never discloses In fact, on its first trading day since the video was posted, the Lidar company's stock is up over 27 PERCENT! (over 50 million $) https://t.co/R0s03rifZY On top of that, he lies about other sponsors He regularly shills Google Pixel phones In the video itself he shows you a Pixel filming the video. Look close though in the reflection of the Tesla, he's actually filming with an iPhone and edited the shot to look like a Pixel https://t.co/xyJR1q1Ofu 65 million subscribers. 500,000 X followers. A pristine reputation. All of that destroyed for a pay day If you are a creator this is the ULTIMATE lesson in ethics and responsibility It takes YEARS to build a reputation, and moments to destroy it forever The backlash since has been unrelenting His video is being bombarded with dislikes. Social media is in an uproar about his lies He was even forced to put out a follow up video. One issue, the video had COMPLETELY different footage from his original video More lies https://t.co/exHpb7PU8V He sold his soul He betrayed his country by attacking the most American made car company in the world He deceived his entire audience. All for a paycheck Let this be a warning for all creators out there: it is NEVER worth risking your reputation for a couple bucks I hope you learned some lessons about reputation from this post If you found it at all interesting, I highly encourage you to follow me @alexfinn for more Also feel free to hit like and RT on the post linked below https://t.co/yisna3hJTz @greg16676935420 Greg IMO is the most trustworthy account on X. When he says he's going golfing and eating hot dogs, you can count on that being the truth

The CIA couldn’t have dreamed up a oneshotting as effective as sending thousands of American girls to RedNote https://t.co/nKMHjxtDJp It’s remarkable - the app is full of this now. These are just a few of dozens of screenshots I took in ~10 min of scrolling. https://t.co/opnIsXtVLU I appreciate that they’re not afraid to shoot their shot with long and poetic messages https://t.co/V4TQ79SZS5

It is officially game OVER! Wan Animate (fully open source) is definitely a SOTA model. The fidelity of the image is like nothing I have seen so far. The relight LoRA does a phenomenal job as well with enhancing environmental lighting & integration. What’s missing? https://t.co/nM1yThtah4 Tutorial incoming 👇🏼 https://t.co/PnsUjr7tpy Access it now inside Enhancor (tools/animate) 👇🏼 https://t.co/KhwPmd5AnQ @MountainMn2022 Dropping a tut tmr on yt with the opensource code Tutorial is now live. Check it out. (i have attached the scripts and workflow to install it locally for free) https://t.co/krbNk7PKzD

What if Studio Ghibli directed Lord of the Rings? I spent $250 in Kling credits and 9 hours re-editing the Fellowship trailer to bring that vision to life—and I’ll show you exactly how I did it 👇🏼 https://t.co/IqUeBSH4H0 Here’s a side-by-side comparison of the two trailers: https://t.co/u3ynXa6bCS My process was: Screenshot all 102 shots in the trailer Remix them to Ghibli style in @OpenAI 's Sora Animate in @Kling_ai and @LumaLabsAI Re-edit in sync with trailer. https://t.co/7jq1EZ7YOn In Sora, I used prompts like: "Recreate this in the style of Studio Ghibli, intricately detailed. Make sure the composition, colors and vibe is similar. The scene pictured shows black cloaked figures on black horses riding away from a massive wave of water on a riverbed that is https://t.co/KME6C0Zzem I didn't have the time to lip sync them, so in Kling and Luma I just used prompts like: "a man yells, fearful" "a woman talks, concerned, brows furrowed" I used about half Kling and half Luma. When one didn't work how I wanted it to, I used the other. https://t.co/bChwf2mSdI If you remember, a few months ago, I went viral for doing the opposite of this. I took a Studio Ghibli film and turned it into live action. https://t.co/i2hFAt1Ipm I have also done other Lord of the Rings anime videos before: https://t.co/MHIXr9iPIr I'm an AI filmmaker and do fun projects like this between bigger projects like this recent anime music video we made for @Timbaland https://t.co/7e5k9XHjLr

Super excited to release Plugins for Claude Code. Plugins are the easiest way to share and install bundles of agents, slash commands, MCP servers, and hooks. Something we saw people doing already, now we made it even easier. @thecat Working on it

1. Durex, 2010 http... The 10 best marketing ads of all time: 1. Durex, 2010 https://t.co/WvgKJtWmg6 2. Porsche, 1967 https://t.co/zFQqHomjgx 3. McDonald's, 2021 https://t.co/PwVCquAXFi 4. BIC, 2021 https://t.co/47Uj9coz42 5. Nike, 2018 https://t.co/Dy9XVG5rrE 6. Nike, 2008 https://t.co/R0Uajpjd0v 7. Timex, 2022 https://t.co/rirvHFwyVD 8. Dentistry Group, 2019 https://t.co/PebUA0xVfL 9. SmileAtTheWorld, 2020 https://t.co/cT4im8fAM0 10. KitKat, 2016 https://t.co/NdEEadZBEw

I tested Sora vs. the new Google Veo-2. I feel like comparing a bike vs. a starship: https://t.co/YcHsVcUyn2 I generated videos from text, using Sora. And the results are dicey. But Veo-2 is: > Ultra-realistic physics > Complex camera motion > Characters with vibrant emotions 2025 is the year of AI video becoming a thing. It's no longer a meme. https://t.co/iHaJNVGe4w

SAN FRANCISCANS: my studio costs $2700 and I stepped on human feces & 1 used syringe on my morning commute CITY: we hear you, action must and will be taken. Scooters are now illegal SF-ANS: what CITY: no more delivery robots SF-ANS: but CITY: workplace cafeterias are forbidden LUNATIC THROWING CATS: *garbles unintelligible slurs* CITY: this man is right. Laundromats are historic buildings that must be protected

People still don’t realize how close we came to a full-blown cultural revolution I saw a woman in a slack channel at uber complain about sexism because a guy had told her « good morning, » (this is an exact quote), and she fell like he wouldn’t have said that if she’d been a guy I saw a PM get reprimanded because he suggested we put a « good looking model » in one of our ads

Wan 2.2 Animate is CRAZY and it actually excels at 3 things from my tests: 1. Lip syncing (so far the best open source I have seen, beating Runway Act2) 2. Consistent lighting & shadows with color tone replication when you swap a character 3. It keeps the replacement character https://t.co/c8v2PeJoOd Full body replacement example https://t.co/CPKv6R8VYW Consistent lighting and shadows https://t.co/ozMFR2YJFd @AlejandroL70403 Yes. Dropping tomorrow. Available now inside ➡️ https://t.co/ntZmJxY3p8 🫡 (not quite sure why opengraph thumbnail does not show on x)

"It's called Ghibli vibe prompting. There's an art to it." https://t.co/4e6cKWoMbU (Original memes referenced) https://t.co/HssaFBORsT I just spent 9 hours making a Studio Ghibli Lord of the Rings trailer. Check it out! https://t.co/P6FFRQPxKQ

@sdrzn 👋 Boris from the Claude Code team here. Compact behavior is the same as before -- the new ⛝ boxes in /context are just a cosmetic UI change that gives people more transparency into auto-compact. We always auto-compacted near 155k tokens so there's enough buffer. We do that for @sdrzn If you're seeing lower rate limits than what we publish, or if you're seeing auto-compact earlier than ~155k tokens, that's a bug. /bug to report. @EricBuess @sdrzn We auto-compact for you when using the SDK, so you don't need to. To grab stats yourself, read assistant_message.usage. @mattrubens @sdrzn Yep just output. We used to have a lower max output limit but people asked for a higher limit @julie_bush @DataWhispurr @sdrzn We recommend sonnet 4.5 for everything — you get more rate limits with it, and it’s more intelligent for coding tasks. Re:hostile, specific examples would be helpful to debug @Danishafzalkhan @sdrzn @BorisJohnson Tool call errors happen with every model provider — in claude code we show errors while others hide them, so it may feel like they are happening more. The reason we show them is claude code is the same tool we use internally at Anthropic, and seeing the errors helps us debug

Yesterday, this account had 0 followers and 0 videos. Today: 120k followers, 8M views, just from posting four Veo 3 videos. If you're not making Veo 3 content, you’re skipping a once-in-a-generation shortcut to the top .001% of creators. Steal this prompt 👇🏼 https://t.co/o1Ykzz4vCo A cinematic handheld selfie-style medium shot, set on a snowy battlefield at dusk. A stormtrooper in full white armor holds the camera at arm’s length, his helmeted face perfectly framed as snowflakes swirl gently in the cold air. His armor is lightly dusted with frost and ash. https://t.co/rjNBsA9zU0 I created this video last week and posted it to my tiktok with 300 followers--the next day it hit 3 million views: So yes, Veo 3 rips with small accounts. https://t.co/iFWzFTzayj

Facebook has turned into an endless scroll of AI photos and the boomers don’t appear to have noticed https://t.co/qvkIbEGgQw The photos don’t even have to be that realistic, they’re just happy to see them https://t.co/VJGTigWjv6 Even when told an image is AI, it just…doesn’t register https://t.co/sfIlO3zEDR

Things I know at 41, I wish I had known at 21. 🧵 The biggest difference between success & failure is getting started: The majority of people I know fantasize about things that actually can be accomplished. They just never get started. If you get started and play the long game, you have a great chance of winning. The 2nd biggest difference between success & failure is persistence: Successful friends & peers of mine have almost always been doing their "thing" for decades. Not years. Most people give up in the "I suck at this" phase without considering the journey from a 30k ft view. Comparing yourself to others is the easiest way to get distracted: You can only control how successful you are, not how successful someone else is. Be the best version of yourself, not a better version of someone else. The most important skill to learn is how to learn on your own: Mentors are great, but they are overrated. Mentors work well when you've reached a ceiling of teaching yourself everything you could possibly know about a subject. Learn to learn and you're unstoppable. Don't spend time with people who don't reciprocate: My therapist put it best when she told me this: "Reciprocity is an indicator that someone is invested in the relationship." Thanks, Erin.

Tried the GDPR data export from Spotify. By default, you get like 6 JSON files with almost nothing. After many emails and complaining and a month of waiting, I got a 250MB archive with basically EVERY INTERACTION I ever did with any Spotify client, all my searches. Everything. https://t.co/ALVw6Auief They even store the brand of headphone I use. How do you even get that data, digging deep in CoreBluetooth? https://t.co/LnuDKxtW6v