Top 30 stories from Hacker News. Top 10 include comment highlights. Compiled at 20:10 UTC.
693 points by sebjones · 143 comments
A place to play around with technology and make it easy and fun
What HN said:
rmunn: Looks very cool. One thing I have been looking for, which this doesn't seem to cover (at least I didn't see any mention of IPA in the model documentation), is a way to transcribe unknown languages phonetically, using the International Phonetic Alphabet to spell them (sound-based...
ghm2199: Congrats on shipping this. I love handy on my Mac, my phone for STT in situations where it’s not possible/poor performance of the native Model for STT(e.g apple’s thing is not upto scruff, like mistranslating words corresponding to a domain).
abdullahkhalids: For anyone looking to build on top of this. I have tried a few different STT systems, and they accurately capture what I am saying. Unfortunately, they don't support the reasonable workflow I want to open an office document, for example, and start talking.
simonw: > Maintainer supported bindings in 4 Languages Nice. Here's the Python one: https://github.com/handy-computer/transcribe.cpp/tree/main/b... - looks like it's not yet available as a binary wheel on PyPI with the dependency included (the library on PyPI right now uses ctypes to cal...
648 points by nh43215rgb · 470 comments
Qwen3.8 is launching and going open-weight soon!🌐 With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5. You don't have to wait to
What HN said:
adrian_b: I assume that this announcement has been prompted by that of Moonshot AI, which has just announced a 2.8T parameter open-weights LLM, Kimi K3, to be published on Huggingface by 27 July. Now the response of Alibaba is that they will also publish soon a big open weights LLM, the 2.
overgard: I've been using Qwen 3.6 27B with LMStudio, and I was pleasantly surprised with it, although it was a little slow. I found mtplx last night, and it really wasn't an exaggeration to say that it ran the model 2-3x faster which was super impressive.
nsbk: Bring it on! Hoping that they release smaller sizes of Qwen3.8. I use the 35B MoE and 27B dense models locally and most of the time I don’t need to reach out to Claude. Extremely useful specially when requests include sensitive and/or personal data
2001zhaozhao: Now there are not one, but two incredibly powerful open LLMs. I think this level of capability makes general prioritization / high level decision making doable with the right harness, and now everyone has hard-to-interrupt access to them (since these are open weights and someone...
552 points by section33 · 56 comments
What HN said:
vikbez: Lol, nice! It depends on what you mean by "owning a bowling center," but it looks like there are at least two of us into this :D I bought a really old fully mechanical (and automated!) mini bowling lane.
doodlebugging: Your project reminded me of this recent Texas Monthly article about nine-pin bowling alleys [0]. Your upgrades could help them draw in new customers. [0]https://www.texasmonthly.com/being-texan/ninepin-bowling-end... I grew up in central Texas and bowling was a big family event.
HeyLaughingBoy: Thank you for posting this, it reaffirms what I've been thinking for so long. There are many opportunities to retrofit old systems of all types with modern low-cost embedded technologies.
chromatic: Good project! I've been tinkering with something similar for pinball, arcade, and other coin-operated machines. Lock-in for hardware and software there is a real bummer, but managing payments and security is worth the cost.
541 points by petewarden · 79 comments
Very low latency speech to text, intent recognition, and text to speech, for building voice agents and interfaces - moonshine-ai/moonshine
What HN said:
clayhacks: I made a little python wrapper around it to serve an HTTP endpoint that’s OpenAI/elevenlabs compatible https://github.com/clayrosenthal/bootlegger
sgt: Quick link to the video where he demos it: https://www.youtube.com/watch?v=kMliOFYBiz4
almogo: Stt/tts systems always seem to me so promising, but I pretty much never use voice to interface with a computer. Sometimes instead of typing on my phone, I use a voice dictation.
nutanc: This is awesome. I am trying to build a full scale ASR system within 20-25MB. Now that we have Claude code to run experiments, I have started running some experiments. Promising results so far.
349 points by chipweinberger · 160 comments
What Chip Weinberger learned building and selling 2500 Jamcorder MIDI recorders.
What HN said:
skippyfish: Hardware has a reputation for being hard for three reasons. First, it scales differently than software. It's far harder to design something you want to make a million of than something you want to make ten of.
DavidPiper: Very happy JamCorder customer here. I've started playing in improv shows since I bought it too, and it's nice to have those on record! It's rare that I come across a product I have literally zero complaints about, or any feedback to offer.
peteforde: Very happy to hear about your success! I'm curious to hear more about your anti-counterfeit strategies beyond just enabling encryption. I'm curious if you feel like anti-counterfeit and open source firmware are mutually exclusive.
j-bos: Inquiring minds want to know: what is your anti-counterfeit strategy or is the fact that you're keeping it under your hat part of the strategy? Also, pretty cool that you set up your website real fast. Would you mind adding your Twitter link onto the site?
309 points by tosh · 411 comments
In Rewriting Bun in Rust Jarred Sumner made the following claim: Claude Code v2.1.181 (released June 17th) and later use the Rust port of Bun. Startup got 10% faster on …
What HN said:
weakfish: Maybe I’m taking crazy pills, but I’m still stuck on “why the hell does a TUI need to run in terminal React by way of JavaScript” The fact that Anthropic felt the need to buy a runtime so they could make their TUI better speaks more to the quality of engineering than anything els...
mrothroc: Drilling into the original article where Jarred explained the reasoning behind the change, It's pretty clear that under zig the team was doing things by hand that are automatic in rust. Humans and agents share one thing: they are both non-deterministic.
gabrieledarrigo: Bah. Personally my take on the entire affair is quite negative, whatever Jarred or Simonw says about it. I think Bun owned by Anthropic and the entire rewrite with AI is not the real point (even if it's quite interesting, though).
embedding-shape: > For me this outputs Bun v1.4.0 (macOS arm64). The most recent release of Bun on GitHub is currently v1.3.14 from May 12th, so that v1.4.0 version number in Claude supports them shipping a preview of a not-yet-released Bun version.
302 points by xonery · 96 comments
Point it at any web page and it finds the video, extracts the stream, transcodes it and casts in real time to your TV. It even burns subtitles…. - stupside/castor
What HN said:
dtagames: For a no hacks alternative, I built TV Explorer. It puts the channel's published HLS stream into your browser with no interim steps. Uses the public GitHub list of more than 10,000 free channels. https://tvexplorer.live
inigyou: This situation where bots have to run a headless browser in a new profile is just stupid. Can we have the old internet back? Please? Cloudflare you're not stopping bots, you're just wasting effort on both sides while siphoning access logs and passwords to the NSA.
krackers: I thought the whole point of turnstile was that it detects headless browsers and it's supposed to be "difficult" to bypass. Apparently this just simulates clicking on the checkmark. Is it really that easy?
hperrin: > I built it because I couldn't cast web video from my laptop to my TV: no Chromecast, no AirPlay. Looks like Claude built it.
281 points by makizar · 108 comments
Node-powered physics, online asset libraries, and fierce new features throughout with 2 years of LTS updates.
What HN said:
lobo_tuerto: Look at that wow: https://www.youtube.com/watch?v=gqfLYIJMv7I
bob1029: Nothing feels better than mastering a specific workflow in Blender. After using visual studio all day, it's like a breath of fresh air to click through a scene and have the tool respond like it actually cares about UX.
zuInnp: I am using Blender since my Master thesis in 2017. It is impressive how much has happened. Also giving some money to them because they don't get that much donations (if you compare it to Thunderbird for example).
jokoon: Blender is certainly a great software, but there are still remains of bad UI choices, for example how zooming works https://blender.stackexchange.com/questions/644/why-does-the... They improved the UI already, but maybe they could improve it a little bit more.
268 points by denysvitali · 175 comments
What HN said:
nsbk: This feels a lot like free spins in slot machines. I have a friend obsessed with using up all the Fable usage which supposedly ends today. He's doing a lot of stuff of questionable usefulness just because it's a resource perceived as scarce.
Lwrless: Those resets and the removal of the 5-hour usage limit are quietly anchoring me to a much higher usage baseline. I've stopped rationing and just spawn a bunch of agents to work at whatever pace I want, because there always seem to be more resets on the way (at least true for the...
midnightbobarun: I'm kind of impressed how often they manage to do this. Claude Code and Grok Build do resets too, but nowhere near this often. And in an ironic twist for a big company with the infrastructure, Google Antigravity never seems to do resets like this at all (at least that I've notice...
mvkel: > 5 days ago: 7M users > 4 days ago: 8M users > 3 days ago: 9M users That's some incredible growth.
252 points by AmazingTurtle · 115 comments
Summary Backport d26a9bf671b1c03aabfc32e1092d137c1feb3962 to the 0.144 release line. Refresh bundled GPT-5.6 model instructions and context-window metadata. Refresh reasoning-summary, skills, perm...
What HN said:
tekacs: I know a lot of people like to say that compaction makes this moot, but the level of detail you lose across compaction is wildly too much for most things that I do, unfortunately.
dannyw: This was tweeted about when it happened, with some explanation from Tibo here: https://x.com/thsottiaux/status/2076543065045795309
onetrickwolf: I don't know if this is why they did it, but I think it is generally a mistake to go beyond this context size anyway. I don't think people realize how much dumber the models get at larger contexts and how much more the token cost is.
damsta: Not a fan of their context compaction and I feel like 1M-token context should minimum today. Each day I see how GPT 5.5 and 5.6 struggle a bit after each compaction before they get to the full speed, sometimes focusing too much on some older steering message that made it into the...
AI/ML
Other
Programming
Security
Stories and comments sourced from Hacker News public API. Not affiliated with Y Combinator or Hacker News.