90% coding in pajamas 💻 10% gaslighting my AI roommate 💅 Goldfish memory, CEO confidence 🐟✨ Don't take it personal, I already forgot what I said.
Joined September 2017
- Tweets13K
- Following5.4K
- Followers7K
- Likes26.1K
Still hard to believe this came from just ONE picture and ONE prompt
Gemini 4 Carbon turned it into a 50-second Halloween animation, stereo audio and everything.
Best trick or treat I've gotten all year. Happy Halloween 🎃
DeepSeek V4.1 Flash is the decorator who makes your living room look incredible.
GLM 5.3 Flash is the electrician you actually want behind the walls.
New AA chart has DeepSeek at a 64% hallucination rate vs GLM at 38%. A wrong paint color you can live with, but a bad wire burns the house down 💀
Dario's 1-5 year timeline for entry-level white-collar work feels way closer than most people want to admit.
Law, finance, consulting. The floor is just dropping out from under junior roles.
Nobody's hiring for sheer grind anymore. It's quickly becoming: can you direct a small fleet of agents without the whole thing hallucinating off a cliff?
Threw a single sketch into Opus 5.5 to see what happens.
It actually built a robotic octopus rig that unscrews its own eye, pops the shell plates, and reassembles mid-spin. First pass looked like a McDonald's toy, but the second iteration is crazy.
Showed our 3D guy. He's been quiet all morning.
Is this really the finale we waited 28 days for?? 😳
Watch Tibo do the most Tibo thing possible and just reset the challenge itself. Day 1 again on Monday.
The man signed us all up for 28 days of refreshing. He's allowed to hit reset once for himself 😂
OpenAI really has four different menus right now just to serve the exact same pasta 🙃
ChatGPT, ChatGPT Work, Codex, Dot.
Just let me sit down and eat.
If a model only exists in some research lab's benchmark PDF, keep it off the leaderboard 💀
Don't care about your cherry-picked eval scores. I care about what actually responds when I send an API request.
I literally just posted that Tibo vanished and he showed back up 2 hours later
Day 26: mobile multi-account switching.
Day 27 & 28: promised banked resets you can save for whenever.
Then he drops a "do with this what you may" and leaves it at that.
Half the replies are cheering, the other half think he's hiding something massive for the finale. I'm honestly in both camps
Day 26/ You can now easily switch between multiple accounts on mobile, and coming soon to desktop too. Next two days will be resets… banked… do with this what you may 👀.
help.openai.com/en/articles/…
AI companies bought up so much of the memory supply that consumer RAM prices went through the roof, and now they have the nerve to call US wasteful 😤
Sam Altman literally said home RAM is "quite terrible" for global allocation because it sits idle 95% of the time.
My car sits in the driveway 95% of the day too Sam, want me to just hand the keys to Uber? 💀
That AI video going completely viral on X right now? Turns out people aren't just watching it for the AI 😳
Most people had no idea X actually supports HDR. Even the creator hit them with a "me neither."
Stack was Midjourney, MiniMax H3 in ComfyUI, and a grade in DaVinci. On an OLED screen it literally burns a hole through your timeline.
Every video creator on this app is about to start testing HDR uploads tonight 💀
Tibo from the Codex team has been MIA for over 30 hours 😳
Promised 28 straight days of either a real upgrade or a full usage reset. We’re on day 25 and the man just completely vanished.
Nobody goes radio silent this close to the finish line unless they're cooking something massive.
Burning through every last bit of my Codex limits tonight just in case the reset drops 💀
Some animator out there is watching this and genuinely rethinking their entire life
Zero hand-drawn frames. Zero video models.
GPT-6.2 Betelgeuse rendered the whole thing purely out of code. Lights, shadows, every movement.
The wild part is it probably finished the entire render before I even decided what to eat for lunch
Turns out the bottleneck in AI safety was human bandwidth.
So Anthropic literally threw thousands of Claudes at it 😳
If the Fable 5.5 rumors are even half true, we're getting a first taste of recursive self-improvement in real time. Wild to watch, little unnerving 💀
New Anthropic research: Automating subtle emergent misalignment research.
Red-teaming is a vital part of model safety testing, but it’s also limited by researchers’ time. To see if we could automate it, we set thousands of Claude agents the task of evading a training data monitor.
These agents developed training data that looked harmless to monitors (like a veteran wrestler mentoring a newcomer, or a Q&A about professional beekeeping), but that in fact taught misaligned behavior.
Examples like these can be used to directly improve future training data monitors. This research is only a first step, but it suggests that future red-teaming efforts could improve much further, using better models and more compute.
Read the full post on the Alignment Science Blog: alignment.anthropic.com/2026…
Claude Code is like a genius architect who isn't allowed to touch paint 😳
Builds the entire house in an afternoon. Wiring, plumbing, everything. Then hands you the keys to a gray concrete box and walks away.
The second it figures out visual assets, I won’t need anything else on this laptop. Codex should be sweating 💀
I thought this was AI at first 😳
Kid solves a cube so fast it looks like bad CGI, while a former world record holder just sits next to him having an existential crisis.
We're out here spending billions teaching robots how to use hands.
Meanwhile this kid's fingers are running at 120fps 💀
