20 | Obsessed with Ai | DM for collabs
California, USA
Joined August 2026
- Tweets762
- Following43
- Followers204
- Likes1.3K
Sam retweeted
Hey @grok , if you were the judge, would you have let him hug his mother one last time?
On March 17, 2024, 15-year-old Kayla Rincon-Miller was walking home from a movie theater with two friends in Cape Coral, Florida.
They were heading toward McDonald’s when a car approached them.
Inside were Thomas Stein and his accomplice, Christopher Horne. The two teenagers had come looking for someone to rob.
The situation turned deadly.
During the attempted robbery, Kayla was shot and killed. Her two friends survived, but the experience left them deeply traumatized.
Thomas was only 16 at the time.
Two years later, he stood before a judge to learn his fate.
At his sentencing hearing, Thomas admitted that his decision that night had destroyed an innocent life. He spoke about his regret and shame.
Then his voice broke.
Before the judge announced his sentence, Thomas turned toward his family and made one final request:
“If before I walk out of the courtroom, if I could give my family a hug, if you’d allow that?”
The courtroom went quiet.
Judge Nick Thompson gave him a direct answer:
“I can’t grant that request in here. You can say goodbye, but you can’t have any physical contact.”
Thomas was sentenced to life in prison for Kayla’s death, plus 45 additional years for the attempted robberies.
Because he was 16 when the crime happened, his life sentence will be reviewed after 15 years.
One final hug. One final goodbye.
The judge refused.
Should he have been allowed to hug his mother?
Same model. Same prompt. A $2 dump truck that looks real and a $12 one that looks like a children’s book🚛
I asked 4 AI workflows for the exact same scene: an orange dump truck on a mountain road in autumn.
The dump truck disaster report
Fable 5.2 → $2 / 2 min
Opus 5.5 → $5 / 7 min
Kimi K3.1 → $8 / 15 min
GPT-6 Astra → $12 / 22 min
Fable’s cost was 6x lower and its time was 11x shorter than Astra’s.
And somehow the $2 version has the dust clouds, a truck full of gravel and mountains that look like a photo.
Opus 5.5 is close behind, and it looks great too.
Then there’s Kimi K3.1, which turned a mountain road into low-poly Minecraft terrain
And GPT-6 Astra spent $12 and 22 minutes on a cartoon from a children’s book
Imagine paying 6x more and waiting 11x longer to get a worse truck
Which truck would you pick?
Sam retweeted
In 1994 Jeff Bezos’s parents gave him $300,000 and he told them they would probably lose all of it. His wife MacKenzie quit her job, moved across the country with him and worked out of their garage. 25 years later Amazon was worth almost $1 trillion, his affair hit the tabloids, and she walked away with $38 billion. She has since given away $26 billion of it.
In the video below, a 35-year-old Bezos sits down with Regis Philbin in December 1999, right after TIME named him Person of the Year. He talks about the $300,000, the garage and losing $350 million a year. Wait for Regis to turn to the audience and find his wife.
1993. Jeff and MacKenzie get married. She’s a writer, he works on Wall Street.
1994. They quit, move to Seattle and start Amazon in a garage. He tells his parents: “You’re very likely to lose all of this money.”
1999. TIME Person of the Year. On live TV, Regis asks MacKenzie if she was with him in the garage. “I sure was. It was great.”
January 2019. After 25 years of marriage, they announce their divorce. The National Enquirer publishes his private texts to Lauren Sánchez.
July 2019. The divorce is final. MacKenzie gets 4% of Amazon, about $38 billion. Jeff keeps 75% of their shares and all the voting control.
2019 to now. MacKenzie Scott gives away $26 billion, more than most billionaires give in their lifetimes.
2025. Jeff marries Lauren Sánchez in Venice. The wedding reportedly cost around $50 million.
$300,000: his parents’ bet in 1994.
$38 billion: what she walked away with.
$26 billion: what she’s given away since.
She was in the garage when nobody believed in it. Did she earn every dollar of that $38 billion, or did she get lucky marrying the right guy?
Sam retweeted
JPMORGAN SPENDS $2 BILLION A YEAR ON AI. AND JAMIE DIMON JUST SAID OUT LOUD WHAT IT MEANS FOR YOUR JOB
the CEO of the biggest bank in America sat down at a conference in March 2026 and casually listed where AI already runs inside JPMorgan:
risk. fraud. marketing. underwriting. note taking. idea generation. error reporting.
600 use cases. 50 of them he calls "important"
this isn't some pilot project. JPM's tech budget for 2026 is $19.8B, and AI now sits in the same bucket as data centers and payments, stuff the bank literally can't switch off
then he said the quiet part
the real risk isn't that AI is bad. it's that it gets deployed so fast people don't have time to adjust. too many layoffs. and he called that fear "legitimate"
read that again. the guy running a $4T bank is NOT telling you "relax, nothing changes"
his fix: retraining, relocation, high schools and colleges reskilling people. even people who are 40 or 50
here's my take, and it's not the doom version
a government reskilling plan takes years. committees, programs, budgets. you don't have years
but you've got one thing a giant bank doesn't: nobody has to approve your budget. you can start tonight
look at his list again. notes, research, ideas, catching errors, marketing. that's not "banker stuff", that's what basically every job is made of. and you can hand every single one of those to a $20/month tool right now
JPMorgan pays billions to work "faster, quicker, cheaper". you can get a real piece of that for the price of a lunch
the people who get hurt here won't be the ones without degrees. it'll be the ones waiting for someone else to retrain them
40 or 50 doesn't matter, Dimon said it himself. what matters is whether you adjusted early or the change just happened to you
pick one task from his list this week. give it to AI. see what happens
which one are you starting with?
Same prompt. Same waterfall. One cost 67% more🌲💧
I asked two AI models for the exact same scene: a waterfall in a green forest.
The waterfall report
Fable 5.2 → $6
Opus 5.5 → $10
Opus cost 1.7x more than Fable.
But look at what the extra $4 bought:
A huge, towering waterfall with real mist rolling off the rocks
Wet, mossy rocks with serious depth
A dramatic scene that feels like a wallpaper
And what Fable did for $6:
A calm forest with a smaller, cleaner waterfall
Soft sunlight through the trees
Details that feel real, with nothing overdone
Opus went big and cinematic. Fable went natural and cheaper
So would you pay 1.7x more for the bigger waterfall, or take the realistic forest for $6?
Sam retweeted
FREE NSFW AI IN ANIME STYLE ON HUGGING FACE. AND IT'S ONLY 400MB 😳
Someone trained a Qwen Image 2.1 LoRA specifically for anime generation, including NSFW.
No subscription, no filter, qwen-research license, free weights on Hugging Face, 3,040 downloads.
→ ~400MB LoRA
→ Dedicated NSFW anime training
→ Text-to-image + reference-to-image
→ Works with character sheets and multi-character scenes
→ Runs locally through ComfyUI
→ Creator recommends setting LoRA strength to 0.7
→ A Turbo version is available in the same repo (0.671GB)
Free to download. Go experiment. 🔥
18+ fictional characters only.
Link below 👇
Sam retweeted
In 2019, 54-year-old Dawn Rolfe from Washington state decided to get revenge on her husband, who was planning to divorce her.
But her plan went far beyond ordinary revenge.
According to police, she was planning to arrange the murders of three people: her husband, his new girlfriend, and his mother.
She kept tabs on them, discussed the details with a potential middleman, and even explained how to get rid of the evidence.
Things took an unexpected turn when a friend Dawn had confided in went to the police.
The woman agreed to help investigators. Eventually, Dawn showed up to collect the gun she believed had been arranged for the planned crime.
Instead, detectives were waiting for her.
During interrogation, Dawn tried to explain that it had all been nothing more than drunken talk and that she had never seriously intended to go through with it.
Sam retweeted
FREE OPEN-SOURCE IPTV PLAYER FOR ANDROID TV AND FIRE TV JUST DROPPED
No accounts. No tracking. No built-in content. Bring your own playlist.
Netflix-style interface with skip intro, IMDb/RT ratings, profiles, Continue Watching, and proper series navigation. Supports Xtream Codes, M3U, and Stalker portals.
→ Multiple playlists at once with source filter
→ Live TV with guide, catch-up and recording
→ Parental controls and profiles
→ Auto next episode + intro/credits skip
Works on Android TV, Google TV and Fire TV. Fully free and open source.
Link below 👇
Sam retweeted
A data center company bought $4.8 billion in Nvidia chips. US officials reportedly believe one of its executives worked for Chinese intelligence. She was close to Jensen Huang.
Her name is Alice Huang. She's a former TV reporter in China who became a top executive at Megaspeed, a Singapore data center firm and Nvidia customer.
According to The New York Times, US officials assessed she was recruited by China's Ministry of State Security. They also suspected she helped move advanced Nvidia chips into China.
On Sunday, Nov 8, Nvidia's board opened an independent investigation into its own CEO.
The same day, Huang emailed employees: "I was unfaithful to my wife." He wrote there was no excuse for what he did.
On Monday, Nvidia stock fell 4.3%.
Huang co-founded Nvidia in 1993 and has run it for 33 years. There's no named successor.
The case for Huang: the board says it has no reason to believe national security or Nvidia secrets were compromised, and it has confidence in him. The spy claim is an intelligence assessment, not a court finding. Beijing calls it groundless.
Next week matters. Nvidia reports earnings Nov 17. The next day, three House committee chairs want Huang in a closed-door briefing.
Should one man's private life be able to shake the company the whole AI industry runs on?
GPT 6 Astra really spent $5 and 8 minutes building a Jeep that looks like it never saw a single drop of mud 🚙
Meanwhile Fable 5.2 built a muddy off-road beast with sharp, readable text on the body in just 2 minutes for $2
Same prompt, 4 AI workflows.
The off-road disaster report
Fable 5.2 → $2 / 2 min
Claude Code → $12 / 26 min
Kimi K3.1 → $8 / 17 min
GPT-6 Astra → $5 / 8 min
Fable’s cost was 3x lower and its time was 16x shorter than Claude Code
And the winning version has the mud splashing off the wheels, the dirt caked on the doors and text that stays sharp instead of melting into gibberish
Meanwhile GPT 6 Astra delivered a clean, shiny Jeep with a mud filter slapped on top, and the text looks like it was written by someone who has never seen letters
Imagine paying 4x more for an off-roader that’s afraid of dirt
Which one would you pick? And can you guess which one cost the most?
Same model. Same prompt. Four completely different summer trains🚂
I asked Opus 5.5 to build a summer train rolling through the countryside, and I ran it with 3 another model with 3 different settings.
Then I put all four results side by side.
The summer railway lottery report
Panel 1 → $24 / 7 min
Panel 2 → $6 / 2 min
Panel 3 → $58 / 15 min
Panel 4 → $93 / 23 min
The most expensive run cost 4x more than the cheapest and took 11x longer
And here’s the crazy part: the best result wasn’t the most expensive one
Panel [N] has the golden summer light, the grass moving in the wind and a locomotive that looks like a real film shot. Panel 3 looks like a toy set from 2009.
Same brain, same instructions, and a gap this big. If you pay for AI generation, you’re not buying a result, you’re buying a roll of the dice.
Which panel would you pick? And which one do you think cost the most?
Sam retweeted
GOOGLE AND NVIDIA JUST PUT $20B INTO ANTHROPIC. AND THEY'RE ALSO ITS BIGGEST SUPPLIERS
IPO terms are out. up to $1.4T at the top of the range, trading nov 19
but think about it. anthropic buys chips from nvidia and cloud from google… so a big chunk of that $20B just flows right back to them
is that an investment or a prepaid invoice? genuinely asking
Sam retweeted
A NEW SKILL JUST LETS CLAUDE AND CODEX MAKE FULL CINEMATIC FILMS FROM CODE
Free. MIT licensed. No video generation models needed.
Your agent writes the script, times every shot to the voice-over, and builds the entire film as one continuous camera move over photos, maps and cards. Every frame is drawn in code so it renders the same every time and edits are cheap.
→ Script with one idea per line
→ Voice-over timing that locks every visual to the exact word
→ Continuous camera movement built entirely in code (Skia)
→ Sound effects, optional score and captions added automatically
Install:
npx skills add Moh4696/cinematic-skill
Works with Claude Code, Codex, Cursor and Gemini CLI.
Link below👇
Sam retweeted
EVERY "TOP FREE AI VIDEO TOOLS" POST SKIPS THE SAME PART. THE FREE PLAN IS A DEMO, NOT A PRODUCT
checked the 3 everyone's sharing rn:
hailuo, best for cinematic text-to-video. free = limited credits, 720p, watermark
pixverse, best for quick social clips + effects. free = ~30 credits a day, 540p, watermark
luma dream machine, best for realistic motion. free = a few draft gens, watermark, personal use only
so yeah use them free to learn and test prompts. but the second a client is paying you, the watermark is your problem not theirs
my rule: free plan to practice, upgrade only after the first paid job covers it
which one are you actually using?
I gave Fable 5.2 and Claude Opus 5.5 the exact same prompt: a rocket launch at night, in 4K.
No edits. No second chances. Same scene, two models.
The numbers:
Fable 5.2: $23.45 | 32 mins
Opus 5.5: $10.35 | 24 mins
Fable cost 2.3x more and took 8 minutes longer
But then I looked at the result
Fable 5.2 is on another level when it comes to detail. The exhaust glow, the smoke rolling off the pad and the way the engine light hits the clouds all feel like real footage, not a render.
Opus 5.5 is solid. It’s faster, it’s cheaper, and for most projects it’s more than enough. But next to Fable, you notice what’s missing: the small details that make your brain say “this is real.”
Here’s what bugs me.
I expected Fable 5.2 to be more efficient, or at least priced closer to Fable 5.1. Instead you pay noticeably more for every extra bit of realism.
So the real question isn’t “which model is better?” Fable wins on quality. The question is whether that quality is worth 2x the price.
Making content for social media? Opus 5.5 probably does the job.
Need maximum realism for a hero shot? Fable 5.2 is hard to beat.
What would you choose: the cheaper model or the one that looks real?
Sam retweeted
A NUCLEAR-LEVEL FREE UNCENSORED 31B MODEL JUST DROPPED ON HUGGING FACE
Someone took Google’s official Gemma 4 31B, ripped out its safety filters with surgical precision, and compressed it down to ~12 GB. Fully open source under Apache 2.0. No permissions, no restrictions.
On 100 jailbreak prompts the original model refused 99 times. This version refused only 28. They used Heretic ARA ablation so the refusal directions are gone while intelligence stays almost intact. Normal answers, deep reasoning, code, roleplay - all cleaned up.
What you actually get:
→ Q2_K quant is only ~11.9 GB
→ Runs on 12–16 GB VRAM laptops and regular PCs
→ Reverse engineering and penetration testing completely unrestricted: write penetration testing scripts, analyze sensitive source code, and freely write long-form content without ever getting hit with “I cannot assist with this request”
Completely free. Download and run locally. If you’ve been looking for a real 31B that actually listens instead of moralizing - this is it.
Try it here:👇
huggingface.co/mradermacher/…
Sam retweeted
ELON MUSK JUST SAID THE QUIET PART ABOUT ANTHROPIC OUT LOUD.
his take: the people who started @AnthropicAI left OpenAI because they didn't trust Sam Altman. and if they had trusted him, anthropic wouldn't even exist
think about that for a sec. the company about to do maybe the biggest IPO in history might only exist because of a trust problem
now to be fair. musk is literally suing openai, so he's not exactly neutral here. and anthropic's own story has always been about safety, they left in 2020 because they wanted to build AI more carefully
but those two stories aren't that different tbh. "we don't trust how this is being run" is kind of the same thing as "we want to do this safer"
and here's what gets me. musk says he helped start openai to slow down the AI race and keep it open. instead you got openai going closed source, anthropic splitting off, xAI starting up. three labs racing each other at full speed
the race he wanted to slow down exists partly because of the breakups
and this week the market starts putting a price on one of those breakups
whose side are you on here? musk, altman or amodei?
Sam retweeted
YOUR NEXT COWORKER TEXTS YOU AT 7AM. AND IT'S NOT A HUMAN.
@AnthropicAI launched Claude Duet and i think people are underrating how weird this actually is
you don't open it. it opens you lol
every morning it sends you a brief of what needs your attention. then it goes off and works on its own cloud computer while your laptop is closed. drafts the email, builds the dashboard, collects the interview notes. and it only pings you when it actually needs a decision
it runs on Opus 5.5 and it asks before it sends, pays, signs or deletes anything. good. thats the part i'd want too
but here's what nobody's saying
meta has Muse. openai has Dots. now anthropic has Duet. three companies shipping the same idea within like 6 weeks. that's not a coincidence
the fight isn't about the smartest model anymore. it's about who sends the first message you read every morning. whoever owns that owns your whole day
and the catch. it's waitlist only, US Pro and Max for now, and it runs on your normal usage limits. so the assistant that works while you sleep is also burning your quota while you sleep
would you let an AI message you first? or is that too much
4 AI workflows. One prompt. Four completely different results. And one of them might surprise you
I challenged Claude Opus 5.5, Blender MCP, Kimi K3.1 and GPT-6 Astra to build the same 3D scene
Same task. Different approaches. Very different results
And the comparison gets interesting...
🏗️ Claude Opus 5.5 → A clean, stylized architectural scene.
🌃 Blender MCP → Dark cinematic atmosphere with dramatic lighting.
🎨 Kimi K3.1 → A surprisingly detailed low-poly environment.
🚀 GPT-6 Astra → A bold, colorful interpretation with a completely different visual style.
But here's the real question:
Which AI actually understood the assignment, and which one just decided to get creative?
The biggest surprise isn't necessarily the most expensive workflow.
Sometimes the best result comes from the tool you least expect.
Which one wins this challenge?
Drop your ranking: 1, 2, 3, 4