Omid Sayfun (@OmiidSayfun) header
Omid Sayfun (@OmiidSayfun) profile picture

jack of all trades // payments @ https://nitter.cf/t.co/hxcWww4QE5 // building https://nitter.cf/t.co/9V9r4EsDX7 // open-source stuff @ https://nitter.cf/t.co/EzdWHNOfSf

Amsterdam 🇳🇱
Joined May 2021
Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)Photo by Omid Sayfun (@OmiidSayfun)
I found this @t3dotcodes configuration randomly. Now it's saving my long sessions. 10/10 recommended. It's called "Compact idle threads": before the cache expires, it compacts Claude threads over 200,000 tokens. Goodbye rate-limits! T3 Code is 91% of my logged Claude usage over the last 7 days.
2
92
Vercel's code verification sends you to Gmail and searches by subject. It was one of the best verification processes I've had lately. Kudos!
2
171
Just switched from Whisper Turbo to MAI-Transcribe 2 and wow! I can't believe I've been missing out on this. It's great on multi-language transcription and it's fast! I still wanna test it on Amsterdam wind. Whisper Turbo couldn't deal with the noise. Will report back.
1
3
268
Did you know that Tailscale has a library that allows your app to connect to your tailnet without the Tailscale VPN switched on? It's called TailscaleKit and it runs a Tailscale device inside the app. Now my iPhone app can reach my Hermes gateway without anyone switching on a VPN. They've only published the source code, in Go, so you need to build it. I tried adding the built framework to my app. It added 70MB to the repo, and building it from source adds lots of new dependencies. The only other build I found was a patched fork with no details. So we created a separate repo that builds it once per Tailscale version and publishes it as a download the app pins. Every release has a GitHub build attestation, and you can check it with gh attestation verify. Two things: - iOS cuts a suspended app's sockets and the Tailscale device inside the app doesn't recover. So it stops when the app goes to the background and starts again when you come back. - Google refuses sign-in inside an app's web page. So the first sign-in to a gateway goes through Safari, and that one still needs the Tailscale VPN. This is the original Tailscale repo: github.com/tailscale/libtail… This is the build repo: github.com/usual-labs/tailsc…
2
3
407
I love having all my MCPs and tools on @executorsh. The only ones I couldn't move there are @mobbin and @posthog, because they return UI components and Executor drops them.
2
8
1,407
This is tempting me like, a lot!
OpenSource our AI-UGC workflow For as low as $0.2/s instead of $50 per video Try it here: treg.to/ugc Reverse-engineer trending videos to scalable UGCs that brings $$$ millions of impressions Repo link below 👇
1
111
Tailscale, the AI Skills Company?
One home for your team’s AI skills. Create, organize, and share skills in Aperture, control who can access them, and distribute them to compatible agents through Skills over MCP. Start sharing skills: buff.ly/O8d6zJh
1
172
Fable 5.2 dropped last night, and I managed to hit my session limit in under an hour. Now that “Reset Limits” button is looking real sexy.
1
4
279
At this point, I'm scared to ask anything in the comments and end up on @dillon_mulroy's list
138
ok so after the MIT weight-space programming post i wanted to see if it works on an image model too gave opus 5.5 the raw SDXL weights and asked it to make every "dog" come out as a corgi in an astronaut suit. everything else should look the same it wasn't allowed to train anything or fit to data. it could read tensors, embed like 40 prompts, and look at renders (15 calls, 6 images each, that's it) so it just... wrote the edit. by hand. a few lines of linear algebra left is normal SDXL, right is after opus messed with the weights. same prompt same seed
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
1
3
227
it also broke stuff lol hot dogs. every attempt explicitly protected "hot dog" and every attempt broke it anyway, there's this garbled ketchup lettering above it now. one of them spells "HOT … DOG" "a child hugging a dog" sometimes gives the kid a suit too (opus actually predicted that one in its report before i ran the test, which is a little annoying) breed was always harder than the suit. both have to go through the same single "dog" token and the corgi part loses more also run 3A read its own pass log outside its folder once. it reported that itself at least
1
1
31
caveats: CLIP is a pretty lenient judge, esp for "corgi". some of these are like half corgis if you look closely. and 14 prompts × 4 seeds is small it's low-rank but still touches 141 matrices (under 1M numbers describe the whole edit). way less sparse than the MIT result "no fitting" was enforced by instructions plus an audit afterwards, not by code. all runs passed the pretty pics are picked, the numbers aren't. every original/edited pair is in the repo so you can see the bad ones too all credit to Takehi, Teoh, Zhang & Khattab at MIT CSAIL, this is just me poking at their idea with a different model and a dumber task edits are normal SDXL LoRAs if you want to try them github.com/iamomiid/weight-s…
1
34
1. what?
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2
197
TIL Apple Shortcuts run as a different app depending on how you start them. Press ▶ in the editor and it runs as Shortcuts.app. Use a hotkey or the menu bar and it runs as “Shortcuts Events”, which has its own privacy permissions.
215
This behavior from Atlassian feels odd. We created a new Forge app, but the author shows up as an individual account name instead of our company name.
2
324
What's happening at Anthropic?
Replying to @ClaudeDevs
We’re rolling out monthly Claude Platform API credits for Max and Team plans: Max 5x: $100 Max 20x: $200 Team: up to $500, pooled Works on any model, including Haiku 5.5, in your code or third-party harnesses. How it works and full terms: support.claude.com/en/articl…
200