635 karma · joined August 17, 2018
+----------+-----------------------------------------------+
| BUILDING | the most gratuitously polished anki extension |
| | https://smart-notes.xyz |
+----------+-----------------------------------------------+
| ALSO | give your agents names |
| | https://getrostr.sh |
+----------+-----------------------------------------------+
| TWEETS | https://twitter.com/piazzatron |
+----------+-----------------------------------------------+
| LEARNED | swarthmore cs / calarts animation |
| AT | (way back when) |
+----------+-----------------------------------------------+
| SAY HI | me at michaelpiazza dot com |
+----------+-----------------------------------------------+Output image tokens per image (2.5 Flare → 2.6 Flare / Sunburst)
low, 1024x1024: 196 → 824 (4.2x) medium, 1024x1024: 439 → 1,449 (3.3x) low, 1024x1536: 158 → 664 (4.2x) medium, 1024x1536: 343 → 1,132 (3.3x)
Same $30/1M output tokens on all three, so per image: low: $0.006 → $0.025 medium: $0.013 → $0.043
Keeping track of many async agentic coding sessions can get utterly exhausting. Giving them names make it surprisingly easier to remember which thread is working on what, and to triage notifications and messages at a glance. It's also weirdly funny.
I hope you find this little tool useful as I have, and looking forward to hearing feedback / anything you build on top of it.
Incidentally, I'm testing Muse as an email triage agent today - it buried "oh also, railway and anthropic say yours keys have been leaked and you need to rotate asap" under 50 routine deploy / admin emails notifications. Oh to live through such an AI powered software renaissance!
feature idea: when the site is in off hour, it would be fun to show looks / spotted on previous days so there’s still something to do.
Happy to share more if you ever want to connect. Tl;dr: feel from the inside is bad, would not recommend!
Back then, I got my genome sequenced via nebula genomics (no longer around). I found a lab that screens ~150 genes that potentially contribute to circadian disturbances. I punched all of those genes into this software called gene.iobio that came with the nebula report, which reports any potentially interesting variants and links out to Clinvar. There were a ton of results, everyone carries loads of these harmless variants, but the one that jumped out was in a gene I knew played a critical role in clock timing, and amazingly, that was it.
A few years I was playing around with analyzing my own genome and found a needle in the haystack: an unstudied clock gene mutation (in CRY2), which, after correspondence and tests from scientists in the field, is effectively confirmed to delay the clock. Sort of unbelievable story, will write it all up some day.
As far as ADHD is concerned, I’ve never been officially diagnosed but certainly have some traits (and notably missing others); it makes me suspect that ADHD is probably a cluster of different etiologies, some caused by circadian issues, some not.
I’ve often wondered if having this variant would be useful at all for chronobiologists in the field to study / observe. Feel free to shoot me an email (in bio).
Trump is stripping the White House for copper and selling it!! Well, actually he’s not, but since we know he’s corrupt, isn’t it safe to just assume he might also be doing this other bad thing?
If your decision is to avoid Muse due to Meta’s poor track record, that’s absolutely your prerogative (and a reasonable one!). But specific claims must be evaluated based on their evidence. This article fails that. There’s no story here.
Point by point:
> “OMG you can jailbreak it and get it to spill its VM”
This is the whole point; any content on the VM is yours. It runs in an isolated sandboxed VM separate from stored credentials etc; this is effectively your own computer. You don’t have to trick it.
> It collects dossiers on your contacts
These are more text files that live on your private VM, alongside memory.md, etc. Do you want your secretary to forget every person you contact every day?
> It accessed Messages without full disk access
This whole story never made sense or was substantiated. Full disk access is an OS level security boundary; the user had to switch this on.
> It sold some guys stuff for too cheap and gave out his address
OK this one I basically believe, haha. Because this is the problem with Muse: the LLM is just too dumb to perform complex tasks effectively in many cases.
The idea is to make an extension that allows you to flexibly fill out fields on your cards with AI generated text, speech, and images. Think generating example sentences for your entire deck, or adding furigana to kanji (small phonetic characters with a very delicate syntax). It works just as well for language learners as it does for medical students, etc.
The secondary goal has been to make this the most well designed, intuitive, and stable Anki extension in the world. I don’t think it’s there yet, but it’s on its way. It’s been very interesting to try to work out a design that affords the flexibility to generate anything for whatever you’re studying but remains approachable and discoverable to your marginal Anki user, many of whom are younger and have varying levels of English ability. Many interesting technical challenges as well.
Slowly building towards the Show HN post, but need a few more features. Connecting with and chatting with dedicated users in countries around the world over the last few years has made this the most gratifying thing I’ve ever built! It’s still at the scale that receiving support emails is fun and I can build relationships with power users and long term subscribers.
https://github.com/steipete/agent-scripts/blob/main/skills/c...
The important bit I found is to explicitly remind Claude that Sol 5.6 is a very smart and good model; otherwise, Claude performs its normal condescension towards any non-Claude model behavior and insists on reading all the diffs in full and testing all of Sol's work, negating any token savings.
Sol is for routine work, Opus for frontend/design, and Fable for more complex / ambiguous / architecture work. Fable works extremely well to drive Sol as a subagent.
Fable is the only one you can actually trust to not look at the code, but Sol is somehow still more pleasant to work with, especially in fast mode. Opus is the enemy, and it will make you insane if you talk to it for too long.
The implications are not clear. They are not clear for the security people, for the SWE people, for anybody in knowledge work whose jobs are impacted.
I wish people would stop with the “the solution is merely simply retool against the part the AI isn’t good at yet” cope and feel the enormity of the moment with humility.
When the dust settles these jobs may not exist, or the jobs that do exist will be unrecognizable from the ones today and perhaps so qualitatively different as to no longer be attractive.
I have found that models are generally poor at managing refactors / complexity while also implementing new features. But I’ve had some success with a semi-lights-off approach where you decompose it and prompt the model adversarially in a second pass to look for new rough edges and areas of complexity or refactors that might simplify the codebase.
So I’d be very curious to see this benchmark but with something like a periodic “refactor turn” interleaved in.
Also eager to see Fable benchmarked; anecdotally that was the only model whose code I felt I could actually trust to not review closely.
I was with you up until this point. I don’t think OpenAI has any more of a substantial product moat than Anthropic; if anything, the Claude / mythos etc brand is a valuable asset that OpenAI lacks.
Yes, many of the elite HN engineer always online types have come to prefer Codex, and but if you actually talk to regular engineers in industry, agentic coding is simply still synonymous with Claude Code.
And for the non-engineering uses, Claude is so much more pleasant of a conversational companion than any of the GPT line, and I suspect is this baked deeply into the model, otherwise OpenAI would have closed this gap by now.
(if you have to say it, that’s how you know it’s good)
At a quick glance, it looks like a thing that should contain meaning and substance. At any closer inspection, it falls apart completely.
Spend enough time arguing with Claude and hearing that combination of words starts making you wince / twitch uncontrollably.
That said I enjoyed the article!
They also absolutely blocked OpenClaw system prompts from this path in the prior weeks, based purely on keyword detection. Seems they’ve undone that now.
For me I always just turn on iPhone screen recording and marvel at this amazing app and wish we had something like this in California.
There is this amazing app called NERV that, whenever there is a large earthquake anywhere in Japan, sends you an early warning push notification and an animated display with shockwaves emanating from the epicenter, plus a countdown timer for the first wave hitting you. The first it went off for me it felt like something out of sci-fi. I think I got 45 seconds this time before my apartment started shaking.
Similar to other users here, giving it access to an Obsidian vault has been the key for me. And I wouldn't discount how much the chat interface matters - Telegram is so much nicer to use for extended conversations than the Claude or ChatGPT apps, etc.
You can feed these tools context about your day to day life, and make them increasingly useful and personalized, in a way that you can't with vanilla ChatGPT/Claude etc without relying on some opaque memory system.
Here's a few things I'm using it for. A lot of things uses cases are fairly trivial but a bunch of small, daily QoL improvements add up:
- Calorie and macro tracker.
- Day to day todo list, obsidian wrangling.
- Tech support for family: I have a group chat we're all part of, and I've created OpenClaw skills for frequently asked questions, a memory system to remember questions, and a periodic 'quiz' based on previously asked questions to help everyone "learn to fish", bit by bit.
- Interface to Anki. Bit of a longer one here that I should write up, but it's easy to use it add cards to Anki on the go and review missed cards from today, ask clarifying questions, etc.
- One off reminders.
- Light mental health support for family / friends. An agent that remembers the cool things you've done lately and proactively reminds you of them, helping you zoom out a bit, has been helpful for those in my life whose brains, for whatever reason, tend towards more negative cognitive patterns. (There is definitely a more refined product here)
- General questions / curiosity; stuff I would otherwise use Claude for that's simply nicer in Telegram.
- Language studying support. I'm studying Japanese and OpenClaw helps me by studying whole sentences, tricky grammar concepts, kanji I commonly mix up - all backed by a well organized Obsidian vault. I add to this system constantly.
And then visual artists are often using Procreate, and those files can get heavy as well.
Plus, it’s nice to carry my iPad around with me in a sling and work in a cafe whenever I feel like it. I wouldn’t want to do that with my 16” MBP.