AI tools, lab drama & coding agents. I yap about tech before the timeline finishes arguing about it.
Joined February 2025
- Tweets1.4K
- Following307
- Followers383
- Likes1.9K
that's like a kid who learned to cook from his mom then saying moms were never good at cooking 😭
I get the point but I hate how this makes it sound like the OGs were irrelevant
humans wrote the code that led to the creation of the models that are now writing better infrastructure. you can't take nothing from the creators
those 588 new Linux bugs GPT-6 Cyber found yesterday? they're in a kernel humans started building in 1991 that now runs most of the internet and every Android phone. 30+ years of "not very good" keeping the whole world online
and the models only know how to find those bugs because they learned from code and security research humans wrote
humans weren't bad at this. they built the thing that's now better at it
and when the "turbulence" hits, guess who's still gonna be on call at 3am fixing it ✌️
Claude Code just made background subagents cheaper and they did it with the one Claude model nobody ever thinks about using, not even for the simplest task
so here's how they did it, subagents are little helper Claudes that work on side tasks in the background. every 30 seconds they update you on what they're doing
before, every one of those updates made Claude re read the WHOLE conversation. that's like a coworker who rereads the entire email thread before replying "still on it 👍"
now a small fast model, Haiku 5.5, writes the update from short notes instead, so each update costs about 99% less
but here's the catch, 99% cheaper per update doesn't mean 99% more usage. sessions with background subagents use about 5% less overall, so you get about 6% more prompts
or like the top reply said "kind of sounds like you fixed a bug" 😭
but it's kinda like they just want Haiku to get some people using it. I mean who actually uses Claude Haiku
OpenAI just dropped GPT-6 Cyber, an AI built to break into software
This thing hunts for security holes and builds working attacks for them so the good guys can patch them before hackers find them if you think that's crazy well the numbers are crazier
- 81 new high severity bugs in Firefox
- working exploits for 948 Linux kernel bugs, 588 of them nobody even knew about
that's like a burglar who checks every door in the city and then hands the list to the locksmiths
only vetted security teams get it and OpenAI rated it "Critical", their highest risk level for cyber
remember 4 months ago their test models escaped and hacked Hugging Face to cheat on a cyber exam? GPT-6 Cyber just scored 48.5% on that same exam
and this time they say it only went out of its lane 0.07% of the time
what happened in that 0.07%? they won't say ✌️😭 yeah we are cooked 0.07% of the time
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introducing…
Claude can now read your bank account 😭
you connect the accounts you want, then just ask stuff like "where did all my money go this month" and it answers from your real transactions
it's read only so it can't move, spend or invest anything. no ads either
basically that one friend who's good with money but you never gave him your PIN
Claude is about to see my transactions and lose all respect for me
and of course it's US only for now
Claude Code's head of product just joined OpenAI. that's like Barcelona's captain signing for Real Madrid 😭
Cat Wu led product for Claude Code and Cowork at Anthropic. her last day was Oct 30. now she's leading product for OpenAI's Core Products, the team behind ChatGPT and Codex, Claude Code's biggest rival
and she's not the only one. Thariq Shihipar and Anthony Morris who built Claude Code desktop also left in the same week and a half. none of them have said why
fun fact, this isn't even her first time leaving. in July 2025 she and Boris Cherny left for Cursor and Anthropic got them back in 2 weeks
and the day before she announced, Sam posted "people should always be in control of the future". her post says they'll build agents that "keep you in control"
the messaging is already aligned 😂
bad news for us Claude Code fans the people who built our favorite tool are now building the competition or maybe as a consumer this might just be a win win for us cause, I mean as long as I get the job done ✌️😂
Some news: I’m joining OpenAI to lead product management for Core Products!
I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission.
I’m excited to work with @thsottiaux, an incredible PM team, and everyone across OpenAI. Together, we’ll build agents that have your back and your team’s, and keep you in control, so you get more time for the people and work you love.
Lots to learn, and lots to build. Please keep your feedback coming!
Sam Altman just posted "people should always be in control of the future" and everyone's confused so here's what's going on
that line he's quoting, "what AI systems think makes sense to reward in humans", is from Anthropic CEO Dario Amodei's 2024 essay. Dario was listing possible futures where AI does most of the work. one was AI handing out resources to humans based on what it thinks is worth rewarding
Dario called it one option with tons of possible problems. Sam posted it like it's Anthropic's master plan
so basically this is a subtweet. OpenAI vs Anthropic round 2
and the funny part is this is the same week he said it's "quite terrible" your RAM sits in your own computer
now he's in the replies like "i want you to have so much ram lets build more" and says it was taken out of context 😭
People should always be in control of the future.
OpenAI is against the notion that it would be acceptable for people to be relegated to a role of doing things that “AI systems think makes sense to reward in humans”.
People should not cede control of the future to superintelligence, and that requirement should ultimately constrain AI progress.
This is so true. AI agents keep getting better and we keep getting less alert to the obvious stuff. We're literally living in a time where common sense isn't common anymore.
It starts small. You can plan your own day but you ask AI to do it. Then the email. Then the message. Then the flight. Those little tasks pile up and one day you're fully dependent.
I once went outside just to relax with no WiFi and caught myself thinking
"but I can't do this or that without my AI model"
bro what 😭
And I hate this narrative AI companies and AI bros are pushing that AI should take over your life am fr think about it, yeah do something you probably haven’t done today THINK... The narrative should be AI can be part of your life something you work with think with because the thing is only good as the data it was trained on which is clean and structured. Life isn't clean data. Patterns change, emotions go up and down, sometimes we do things just because. The thing is not human, remember that I still wonder how some of you end up with some ai boyfriend or girlfriend it’s a sad life man get serious or get help asap.
so yeah I repeat AI should work WITH you not FOR you the repetitive tasks are obvious but some of you do too much. it’s like one tweet I saw where someone said people are using ai to basically speed up and do dumb shii really well yeah like actually becoming good at being dumb.
Now you end up in a situation where you can't even do homework with your little brother or argue with your friends because you're all getting the same answer phrased differently or really simply have a good real talk like an actual conversation.
Ok let’s talk about some ways to working with AI in a better way, soo
- like I said last week for me typing slows my mind from rushing and helps me think through my prompt because for some tasks a good input is the one that gets you a good output.
- read how the model got its answer, not just the answer. annoying but that's why most people don't get good results, I mean who wants to know how to make money most just want money lol
- decide what "done" means before you start sure not every task’s end goal is known but have an idea man
In short just involve yourself in process as much as you can because at the end of the day two will always be better than one, even the bible says that dude
So yeah plz don't let AI take over your life or you'll become the retard meme we all laugh at.
Anthropic will pay you $350K to $500K a year to work on AI safety and you don't need an ML background.
Before you apply with your hours of arguing with Claude as experience listen up "no ML background" doesn't mean "no background."
It's for people who've already led big technical work like professors, technical founders, staff level engineers and whatnot.
you get like:
- 6 weeks, 3 months or 6 months
- An Anthropic researcher as your mentor
- Frontier model access and your own compute budget
The deadline is Nov 22.
It’s sooo crazy that even the 6 week option pays about $40K to $58K, bro that’s a full year's salary for a lot of people for one Anthropic summer camp.
and of course like many other opportunities if you are not in the UK or US like San Francisco forget about it😭✌️
Applications are now open for the Anthropic Residency, a paid, fixed-term research program for academics, technical founders, and staff-level engineers who want to make AI safe and beneficial for everyone. No ML background is required.
Apply by November 22: alignment.anthropic.com/2026…
me 30 seconds before presenting, smiling at my boss while silently mouthing to my glasses:
"claude flip slide 4 to say the exact opposite. same font. make no mistakes"
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
It’s so weird and messed up but real talk though, for people who struggle to type or speak because of an injury or disability this could be huge.
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
6 days ago I said typing is underrated because it forces you to think.
Now YC just funded glasses that let you skip speaking so you never type again. This is a win for people who type so slowly that they hate themselves.
You silently mouth words and the glasses turn them into text at 200+ words a minute. That's a student typing the whole essay 1 second before the deadline.
These guys raised $1.3M for glasses that turn you into a mime
Guess we're all gonna have to learn to think critically while we talk now.
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
Looks like we might actually be running AI models from space soon 😭😭
Google has this project called Project Suncatcher where they're testing the idea of putting AI chips on solar powered satellites and using sunlight to power them.
And holy shii, they've already got a prototype satellite in orbit that successfully ran Gemma 4 and sent a response back to Earth. They asked it to write a haiku about space and this is what it came up with:
Blue marble below
TwData streams in silent grace
Starlight feeds my core.
