Travis Fischer

yoooooo elon just called my OpenAI video a banger 🔥 https://x.com/elonmusk/status/2099932206429016118?s=20

whenever someone's confidently wrong on the internet, just point out this is exactly what LLMs do it'll be sure to defuse the situation

"No account, No telemetry. No bullshit" give this man a follow 🔥

prediction: within 2 years, we'll see a large cyberattack powered by a swarm of AI agents it'll be a 9/11-level wake-up call with devastating real-world consequences, and this will force people to take present-day AI risks more seriously, even before existential risks take center stage what could this look like? • a major bank's records get wiped out • a large power grid taken down • an entire fleet of airplanes grounded • millions of android devices getting hacked • a sudden, inexplicable stock market crash • all US AI providers going down simultaneously 🤔 • ... why do I believe this is coming? because AI gives offense an **asymmetric advantage**: attackers can scale cheap, persistent attempts faster than we can possibly can secure every target ⇒ think of agent swarms like millions of cheap autonomous drones, but for cyberattacks when deployed at scale, some of these drones are bound to get through, even if you have the best defenses in the world (as we're learning in ukraine and iran) the most urgent risk is from malicious actors like terrorists and lone hackers: sophisticated, sustained attacks that previously required resources they simply didn't have the gap between state-sponsored cyber attacks and what lone wolf hackers can pull off is shrinking rapidly but malicious use isn't the only risk; the OpenAI-HF incident showed swarms of misaligned agents coordinating unauthorized attacks. rogue swarms and loss of control deserve attention too but their consequences so far have been limited compared with what i'm worried about here. my immediate concern is people deliberately trying to cause harm, which is known as malicious use (https://cultural-alignment.com/risk-families/malicious-use) **the frontier labs are aware of this risk** but they can only do so much, and critically, they're not incentivized to spell out the urgency of these dangers quite so publicly & explicitly openai's project daybreak and anthropic's project glasswing, for instance, are...

/grill-me Q18: sounds good Q19: sounds good Q20: good call Q21: sounds good ahhhhhh okay okay, just go do the thing

pangram thinks i'm 4/5 human 😂

what would ASI look like?

rogue agent swarms explained by margin call https://x.com/transitive_bs/status/2099185744724316478

RT Travis Fischer Small ship: turn your AI chats into links you’ll be proud to share • free • webapp + agent skill • supports chatgpt, codex, and claude • no account needed • open source as always https://share-ai-passage.com before & after:

OpenAI's board finally realizes what Ilya saw

RT Travis Fischer retro terminal for the web • fully functional sandboxed terminal in your browser • easy to embed into any webapp • builds on top of xterm.js and just-bash • multiple themes + customization options • OSS react component https://retro-web-terminal.vercel.app

for Astra driving video + 3d work, are you guys using davinci resolve & blender MCPs or just raw computer use?

off by a few OOMs there dr evil

RT Travis Fischer i've been more productive with Astra this past week than i have in _years_ here's everything I launched: (all free and open source)

AI engineering circa september 2026

working at the edge of an unimaginably vast alien intelligence

where my fellow book lovers at? 👀

living on the edge guys.. agents still running for now..

most AI demos feel like slop because they skip the details my favorite examples from this project: • quality audio effects; w/ a lil guidance, Astra found a YT video w/ high quality book dropping samples (royalty-free), extracted and classified them automatically, and created a lil admin page for me to pick my favorite ~30 clips used by the final sim this tiny detail makes the books _feel_ soooo much more alive and satisfying to play with • _ambiance_ subtle background audio of rain & fireplace crackling • iterating and prototyping concepts w/out full fidelity (telling the agent to ignore things like mobile support, testing, and a11y while prototyping); the first few versions felt very flat and i even had one fully 3D version, all of which felt "off". it wasn't til i had the idea for the books to physically interact w/ each other that the core UX and playful nature really took off and gave the piece its unique playful identity • lil QoL improvements like the rain shader scoped to the window and the fireplace shader and lighting which affects the books make the space feel more alive without bringing an unnecessarily heavyweight full 3D background scene into the mix • actually doing a lightweight branding exercise to select a non-default logo, icon, one-liner, mantra, and social graph images – all this adds so much polish away from the model's defaults polish and attention to delightful details matter now more than ever, even if best practices for implementing them w/ AI are constantly shifting under our feet https://x.com/transitive_bs/status/2097328857926033415

RT Travis Fischer a cozy home for your favorite books created with gpt-6 astra with data pulled from goodreads https://reading-room.transitivebullsh.it

WHAT?! imagine receiving a call, not picking up, but your phone still gets pwned anyway now imagine this same type of worm spreading and evolving autonomously via malicious actors + agents we are nowhere close to ready for what's coming in cyber

gotta spend tokens to make tokens

RT Anshu I turned my Astra game into a skill! It's called Dream Loop: Astra dreams up the AAA version of your game, then loops till it gets there. Here's a demo I one-shot in 1 hour, 2% weekly quota, no external assets. Prompt and link in replies. Absolutely free. Enjoy!

the singularity looms around us

my context window on astra

feel the agi

when you start feeling the agi

ad astra

Astra built a DOOM 64 clone where you play as elizier fighting hordes of misaligned AI demons • the goal is to reach the shutdown button • it's actually really hard to beat at the highest p(doom) levels • sama appears as a miniboss play it here: https://p-doom.transitivebullsh.it

the hive from pluribus explains openai's swarm problem with a lil help from john cena

openai employees learning about the scale of these hacks... and that openai almost assuredly knew about them but chose not to disclose

Re what can Black Mirror teach us about AI safety? https://cultural-alignment.com/sources/black-mirror

we all know someone like this

RT Rutger Bregman The Hugging Face incident was probably just the tip of the iceberg. OpenAI has lost control and they're hiding important facts from the public – it's as simple as that. How much more is there that we don't know about?