Sam Butler

The more I use Linux, the less I like macOS

@chrisdeweese_ so that’s what the stargate is used for

As an ex-openai engineer I’d like to share some more on this. This new Images V2 uses a novel approach. They invented a way to transport a standard camera through a multiverse dimension. By prompting the image you want, OpenAI teleports a camera to that specific dimension to snap the photo then return the data Every time the prompt fails, we fail to teleport the camera back into our dimension, and it’s costly So we have a large volume order with Sony to prepare for the anticipated demand when Images V2 launches Unfortunately our multiverse engineers were poached recently to work on @Meta’s metaverse

@morqon @SemiAnalysis_ I’d been wondering how it was worth the effort to rewrite their training/inference stack for TPUs, but this makes a lot more sense. they didn’t

openai’s deal with google cloud is for GPUs, not TPUs, and a large share of those are redistributed from coreweave

@benhylak the convergence will be companies who can figure out how to create a product for one group and expand into the other. not sure which direction is easier/preferable the most relevant example I can think of is @SunoMusic buying the WavTool AI DAW

in ai, it feels like there's broadly two trends: 1. make experts faster (e.g. claude code) 2. make anyone an expert (e.g. v0) it'll be interesting to see how this converges.

@jakemclain_ @muse_art_ a super cool use of Max for Live can’t wait until it’s available as a VST plugin ever since WavTool disappeared, I’ve been waiting for someone to make this

Today I'm launching @muse_art_ It's an AI tool for music producers that lets you generate MIDI and synth presets, available for Ableton Live and on the web.

@DhravyaShah @natolambert’s @interconnectsai @dylan522p’s @SemiAnalysis_ @GergelyOrosz’s @Pragmatic_Eng Matt Levine’s Money Stuff Bloomberg column. he’s effectively the Ezra Klein of financial columnists, in terms of readership

I want to consume better content. What’s the best substack / blog / newsletters you follow? Where do you find good content to consume?

@togethercompute Seems like a missed opportunity that your own model is not available on your platform

Announcing DeepSWE 🤖: our fully open-sourced, SOTA software engineering agent trained purely with RL on top of Qwen3-32B. DeepSWE achieves 59% on SWEBench-Verified with test-time scaling (and 42.2% Pass@1), topping the SWEBench leaderboard for open-weight models. Built in collaboration with the @Agentica_ team. 💪 DeepSWE is trained with rLLM, Agentica’s modular RL post-training framework for agents. rLLM makes it easy to build, train, and deploy RL-tuned agents on real-world workloads — from software engineering to web navigation and beyond. 🤗 As always, we’re open-sourcing everything: not just the model, but the training code (rLLM), dataset (R2EGym), and training recipe for full reproducibility. 🔥 Train DeepSWE yourself. Extend it. Build your own local agents. No secrets, no barriers. DeepSWE and rLLM mark our major shift: from training language reasoners to building language agents that can truly learn from experience. We believe the future of AI lies in experience-driven learning — and we’re here to democratize it. Welcome to the era of experience. 🌍

@Angaisb_ Claude Max > Cursor Pro > Claude Pro

Do you prefer Cursor Pro or Claude Pro (with Claude Code included)?

@cluely, but for everyone who doesn’t trust them enough to use their product

@togethercompute chat has free, seemingly unlimited access to these open weight models I’ve been paying 4 cents an image on @FAL for flux kontext [pro] chat.together.ai

@zCat_App Mulvad has a much more robust solution for completely disassociating your personal information from your VPN account accounts are identified by an anonymous account number, instead of being associated with your email

I trust Proton more than Google. It does not mean it is perfect. But what is better?

@inerati as a Vision Pro owner, yes a bit for most people, it’s not comfortable to wear for extended periods of time and unless you’re using the Vison Pro as a virtual display for your Mac laptop, it’s not a great device for productivity support.apple.com/en-us/118521

lowkey want a vision pro but only for doing work on planes. is this retarded idea or

@katzenzeitungen @SIGKITTEN fail2ban deals with the malicious login requests that an open ssh port will receive tailscale creates a software defined private network between your devices, which eliminates the need to worry about login requests as the port is no longer publicly accessible

@samdcbu @SIGKITTEN Is fail2ban too complex of a solution?

easiest claude-code (or any other cli coding thing) iPhone setup imo - termius (or whatever ios ssh client you like) - ssh to your laptop > tmux > claude thats it

@SIGKITTEN I was just asking to confirm that you didn’t port forward the ssh port on your laptop adding tailscale would be a simple addition that would let you ssh in from anywhere

@samdcbu um yes

@ARandomNeko21 @HSVSphere yeah that would be fine, one of my email aliases is s@butler.cx if it’s still available buy that domain immediately

@HSVSphere is me@<myname>.com valid?? cuz somehow <myname>.com was on cloudflare for $12 a year or smth

One great thing about having your own domain is that you're no longer a gmail pleb

@ShinyQuagsire yeah it’s definitely USB2 because the cable between the battery and headset is actually a jumbo lightning cable

iperf3 says 340Mbit so it's definitely still USB2 I think, but it's still a hell of a lot more stable than WiFi

not to rub salt in the wound, but visionOS 26 supports USB ethernet over the dev strap, unlike Quest

@allisonology @xeophon 🥺👉👈

@xeophon 🤝

my claude code will talk to your claude code

I see this mentioned constantly on my timeline if true, it should be trivial to prove it. just run SWE-bench every hour and a cyclical pattern should quickly emerge in the data until then, I think the most likely explanation for this phenomenon is mass hysteria

@marimo_io is the first OSS notebook with a modern UI/UX like DeepNote/Hex it reminds me of Helix and @typstapp they’re all great software, but they’re probably doomed bc they’re trying to displace deeply entrenched, incumbent software (marimo=Jupyter, Helix=vim, typst=LaTeX)