Hacker News: bytesandbits

New comment by bytesandbits in "We must pace the frontier"

bytesandbits — Sun, 13 Sep 2026 09:41:10 +0000

This is just bait for regulatory capture. It’s the only way in which you can find alignment of Elon, Sam and Dario within the same 24h.

New comment by bytesandbits in "Linux Zoom client proactively reading everything written to X11 clipboard"

bytesandbits — Sun, 13 Sep 2026 09:40:07 +0000

Zoom is malware, has always been.

New comment by bytesandbits in "Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot"

bytesandbits — Wed, 12 Aug 2026 20:02:20 +0000

No Small Actors.

New comment by bytesandbits in "Show HN: Immersive Gaussian Splat tour of grace cathedral, San Francisco"

bytesandbits — Tue, 21 Jul 2026 17:32:26 +0000

I imagine it is based off World Labs

New comment by bytesandbits in "Show HN: Immersive Gaussian Splat tour of grace cathedral, San Francisco"

bytesandbits — Mon, 20 Jul 2026 23:20:03 +0000

absolutely crazy level of detail... in awe

New comment by bytesandbits in "VS Code inserting 'Co-Authored-by Copilot' into commits regardless of usage"

bytesandbits — Sun, 03 May 2026 07:40:18 +0000

I thank Microsoft deeply for the forced copilot crap, almost impossible to remove, that they have put into vs code. Finally after 5 years I have deleted vs code in my Mac! that was the last piece of windows software I still had around. VS code was great years ago, until Microsoft started to push crap into it – and afaik they also made the fully open-source, telemetry-free fork difficult to use with many extensions.

Really, thanks for forcing me into deleting it. turns out vim + Claude Code or codex was much better all along, it really works well for me.

New comment by bytesandbits in "GPT-5.5"

bytesandbits — Thu, 23 Apr 2026 19:15:44 +0000

I know a 10x engineer when i see one.

New comment by bytesandbits in "Show HN: Ghost Pepper – Local hold-to-talk speech-to-text for macOS"

bytesandbits — Wed, 08 Apr 2026 21:09:29 +0000

fine. I will port it myself. Real-time, sub 100ms latency. Here

https://testflight.apple.com/join/myNP5XvU

New comment by bytesandbits in "Show HN: Ghost Pepper – Local hold-to-talk speech-to-text for macOS"

bytesandbits — Tue, 07 Apr 2026 02:38:53 +0000

Handy with parakeet is pretty awesome by the way!

New comment by bytesandbits in "Artemis II crew take “spectacular” image of Earth"

bytesandbits — Sat, 04 Apr 2026 02:29:02 +0000

here the original NASA photos at high resolution without unnecessary ads.

https://www.nasa.gov/gallery/journey-to-the-moon/

New comment by bytesandbits in "Artemis computer running two instances of MS outlook; they can't figure out why"

bytesandbits — Fri, 03 Apr 2026 06:56:12 +0000

why is a mail client needed in an onboard space computer at all?

New comment by bytesandbits in "Day 1 of ARC-AGI-3"

bytesandbits — Fri, 27 Mar 2026 07:58:06 +0000

anything that doesn't touch the model parameters at all once it has been compiled. for example, in streaming ASR of an encoder-decoder you can get gains in accuracy just by enhancing the encoder-decoder orchestration and ratio, frequency of fwd passes, dynamically adjusting the length of rolling windows (if using full attention). Prompting would be part of this too, including few-shot examples. Decoding strategy is also part of this (top-k, nucleus, speculative decoding, greedy or anything else). Applying signal processing or any kind of processing to the input before getting it into the model, or to the output. There are a lot of things you can do.

New comment by bytesandbits in "Day 1 of ARC-AGI-3"

bytesandbits — Fri, 27 Mar 2026 06:13:54 +0000

we constantly underestimate the power of inference scaffolding. I have seen it in all domains: coding, ASR, ARC-AGI benchmarks you name it. Scaffolding can do a lot! And post-training too. I am confident our currently pre-trained models can beat this benchmark over 80% with the right post-training and scaffolding. That being said I don't think ARC-AGI proves much. It is not a useful task at all in the wild. it is just a game; a strange and confusing one. For me this is just a pointless pseudo-academic exercise. Good to have, but by no means measures intelligence and even less utility of a model.

New comment by bytesandbits in "Apple's 512GB Mac Studio vanishes, a quiet acknowledgment of the RAM shortage"

bytesandbits — Mon, 09 Mar 2026 03:34:42 +0000

https://machinelearning.apple.com/research/exploring-llms-ml...

New comment by bytesandbits in "Apple's 512GB Mac Studio vanishes, a quiet acknowledgment of the RAM shortage"

bytesandbits — Sun, 08 Mar 2026 18:14:02 +0000

Apple

New comment by bytesandbits in "MacBook Pro with M5 Pro and M5 Max"

bytesandbits — Wed, 04 Mar 2026 09:34:57 +0000

4x faster PREFILL not decode. Decode is bandwidth-bounded. Prefill is flops-constrained.

New comment by bytesandbits in "iPhone 17e"

bytesandbits — Tue, 03 Mar 2026 05:49:45 +0000

do you run two eSIMs when traveling and if so how is stability / battery life?

New comment by bytesandbits in "Inside the M4 Apple Neural Engine, Part 1: Reverse Engineering"

bytesandbits — Tue, 03 Mar 2026 05:39:06 +0000

incredible work

New comment by bytesandbits in "Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3"

bytesandbits — Sun, 01 Mar 2026 10:16:31 +0000

I am not sure what "edge" device you want to run this on, but you can compress parakeet to under 500MB on RAM / disk with dynamic quants on-the-fly dequantization (GGUF or CoreML centroid palettization style). And retain essentially almost all accuracy.

And just to be clear, 500MB is even enough for a raspberry Pi. Then your problem is not memory, is FLOPS. It might run real-time in a RPi 5, since it has around 50 GFLOPS of FP32, i.e. 100 GFLOPS of FP16. So about 20-50 times less than a modern iPhone. I don't think it will be able to keep it real time, TBF, but close.

regardless, this model with such quantization strategy runs real time at +10x real-time factor even in 6-year old iPhones (which you can acquire for under $200) and offline at a reasonable speed, essentially anywhere.

You get the best of both worlds: the accuracy of a whisper transformer at the speed and footprint of a small model.

New comment by bytesandbits in "Microgpt"

bytesandbits — Sun, 01 Mar 2026 10:09:01 +0000

sensei karpathy has done it again