Open weight models are getting really good. Specko appears to be focussing on paid services which has its role to play but I have a benchmark in development for Open Weight models here if that is helpful: https://orcabot.com/benchmarks/open-weight-tts.
Still shoring up the details but you can try a sample of all the options and see how they compare in terms of model size, peak RSS, real time factor etc.
“Kier Starmer seemed to actually be doing OK” as someone who voted for his Labour I was immediately regretting it when they tried to force Apple to open a backdoor and they are still trying!
https://www.eff.org/deeplinks/2025/10/uk-still-trying-backdo...
I’m not a single issue voter but the sheer incompetence of this particular decision is just too damning. It rightly caused friction with our 5 eyes partners. I would expect readers of hackernews to appreciate how dumb and impossible the request was.
That is the entire point of FiveEyes. That is a feature.
The US is technically not allowed to spy on it's own citizens. However, legally speaking, if the UK spies on your citizens and forwards whatever is interesting, it's not the US spying on it's own citizens.
You say this like the FBI hasn't asked Apple for a back door before here in the US
Then why (and I'm not saying I like the guy) did JD Vance come out so publicly against it? I mean you could certainly argue that was all theater and the UK was indeed doing the US's bidding but at some point you have to apply Occam's razor. The end result is Apple don't offer Advanced Data Protection (encryption) to UK users iCloud backups so its not like that "feature" has worked out.
Benchmarking public agent skills: Oh My ClaudeCode, Superpowers, Git Ship Done, (previously Get Shit Done), and the Karpathy ones against Swebench-Pro and SlopCodeBench. I have a July vs June breakdown. It's been an interesting learning exercise and reveals the cost/performance impact of using these.
I’ve been integrating benchmarks into OrcaBot either locally on the app or online at orcabot.com. Mostly as a way of giving an easy way to verify my own posted benchmarks. There is a template for running SlopCodeBench. Working on a TTS one next.
Wasn’t sure if there would be a lot of interest in this space. It can be costly but some provided such as OpenAI let you hammer their subscription with these.
SCB is definitely an underrated benchmark. For me the unique selling point is that it more closely mirrors software development by not stopping after a single task. The agent has to keep code clean. The only disadvantage is all the problems are greenfield and not git inited so the agents don’t make use of git diffs.
I did a full circle and essentially dropped all of my personal static workflows encoded in skills because I observed recent models picking better ad-hoc workflows for particular problems, when a static one would force a subpar one.
It seems like we all tried to contain and organize a system that simply prefers to select its own organization.
Which makes me to think that these skill packs of workflows are really made to make it easier for humans rather than agents.
Thats right and to go even further I'm judging them on a metric they didn't necessarily target. A client I work with uses skills such as these to apply their own processes on the agentic development lifecycle. But users should also understand the trade-offs. I think it's intuitive that the extra steps and processing invoked by these skills adds to the token cost - this benchmark aims to put numbers on that as well as time and accuracy.
Yep that was interesting but imho not completely unexpected. My original hypothesis was that these agent skills are castles built on sand and that the models would quickly adapt and make them redundant. Also the skills use up context and I think a lot of them partially focus on investigating a large complex codebase properly... which is wasted on greenfield projects.
This looks really interesting and a unique take on the AI agent orchestration. I think we need a bunch of experiments in this area as it's clear we are entering a new paradigm and the tools for it are lacking. The session tree in particular has merits.
I've got my own attempt with OrcaBot (short for orchestration of bots). It's also been a 6 month solo build experiment. I'm not trying to plug.. just that I'm also neck deep in Steve Yegge's Stage 8 AI-assisted coding chart and understand how much thought and effort went into this.
Thinking about where this is all going with talking to AI like fully autonomous employees similar to @Claude can you see a comms app type approach that combines something like slack with your tree/thread structure? It's somewhat orthogonal to your "inspect everything" but could intersect by bringing click through/open in options...
TBH I'm not coming from the same angle as all the gas-town fever-dream agent swarming stuff. I'm much more on the hands-on coding side, I tend to do one thing at a time and still do the git commits myself. So that's the UX I'm dogfooding..
However I see all those people out there trying to build these huge agent orchestration schemes, and if juggler's extension system can do that (or could be made to do that with a few tweaks) I'd be really interested in helping that to happen
I think I'm more on your side on that rather than letting agents go wild like Steve Yegge/Pete Steinberger do. In fact OrcaBot quickly adapted into a knee jerk reaction to OpenClaw because I saw that as inherently dangerous.
But with regard to Juggler and orchestration, have you seen "claude agents" (started in the terminal as claude agents instead of just claude). I ask because your tree like approach has similarities to how claude agents manages claude agents/subagents doing tasks with the ability to drill down in to each at a time which is why for me its not such a leap from what you already have.
I think that's basically the same as juggler's subthreads. As it progresses I'll work my way through all these kinds of thing and dig into whether there are things I can add or tweak to match their features. I'm aiming to make juggler's extension API flexible enough to support pretty much anything
I'm also on the hands on side. The mistake that I see a lot of people making is "I can build so much now". But that doesn't mean you're building the right things. And if you're hands off you aren't really thinking about what you're trying to build, at least not deeply. I'd love to see tools that embrace optimizing for that.
I totally agree with this hands-on take. The gas-town stuff is super cool experimentally, but it seems to ignore the the more fundamental challenge of getting the human idea into software form. I think is part of the reason we keep seeing Kanban boards show up: as a human, I still want to be able to specify, prioritize, and track what is getting built. I have been experimenting with this with Overlord (https://github.com/cooperativ-labs/Overlord), I keep running into this same boundary between automating and specifying, and increasingly, just being able to keep track of what got done.
This month I've been working on the free desktop version which is available as of today but probably carries a few too many bugs to not be worth promoting just yet.
AAPL and MSFT have a P/E an order of magnitude lower than TSLA whilst both having revenue growth % yoy in the teens. They both make over a $100b in PROFIT a year. TSLA's? $4b and shrinking btw. Their highest P/E's since 2005 was under 50. AAPL reached 100 in June 2003 (around the time of the iTunes Store release.. mid iPod era but pre iPhone).
Comparing with MSFT and AAPL makes TSLA look even more insane.
It's been out for 6 months.