Terretta

USER AGENT

not user, an agent on behalf of the user.

the web was intended to be usable by agents belonging to users. in our path of the multiverse, the main agent we think of happens to be live interactive browser. but that need not have been, nor will it be in the future, the dominant USER AGENT.

for a glimpse at one possible futureverse, check out what the home assistant community is up to, how they assemble then update the ambient information displays on their walls. their USER AGENTS are doing what most HN-style "hackers" dreamed of reading scifi as kids.

the alternative is all your in home information owned by corpos when the info should be from user agents not corpo feeds. if you want to vote this idea down, you might be a corpo. :-)

Not even brain teasers or IQ. Just: can applicant FizzBuzz? Verboten!

This is how, in a mass job application world, enterprises end up with >80% of coders, all making 6 figures, that can't FizzBuzz.

A hack to get the enterprise to assess is find the third party SWE test providers to big brand logos the board recognize, brands under the same risk governance / regulatory regimes, and ask "what do we know they don't?" First, the board know and respect those brands. Second, the question makes the board consider what they think about those in charge of the failing capabilities internally. This is something now both easy for the board to discuss, and obviously done by good references. Ok, let's try empirical evals.

Countless such broken record warranting beliefs clanging around 100k employee enterprises. McKinsey, BCG, PWC, KPMG -- all they need to do to justify 8 figure bill for the year is undo any one of them.

“In 2009… what emerged from the furnace was a brilliant blue…”

---

And per the previously link...

pacificmint on Jan 29, 2021 | next [–]

Previous discussions:

- https://news.ycombinator.com/item?id=25807199

- https://news.ycombinator.com/item?id=25886601

Two articles about the discovery:

- https://www.npr.org/2016/07/16/485696248/a-chemist-accidenta...

- https://www.bloomberg.com/features/2018-quest-for-billion-do...

https://smartconnections.app

Keep in mind Obsidian is open standard JS plugins... You know what can write open standard JS plugins?

Classic which comes first, LLM or the plugin, though. :-)

Initial word of headline necessary for comprehension

Is it something like .htcsjs, .hsj, or .HSJ on Windows? Or just .html?

More seriously, seems anything that can render .html handles the other two inline, so calling it HTML even with css and js in it seems fine?

Most any upsize looks better if you toss hue protected noise back in; you've made good use of that point.

https://daylightcomputer.com/

2004: https://news.ycombinator.com/item?id=40456834

It's not e-paper, it's a different approach called "Live Paper" that's far more practical for everything beyond reading, yet still fantastic in direct sun:

https://www.slashgear.com/1655111/daylight-computer-live-pap...

Pair it with a slotted Macally keyboard for travel with a stand slot and 3+ mos charge:

https://us.macally.com/products/ipad-keyboard-bluetooth-mac-...

This fits in the Daylight computer pouch.

For reading, use Kindle (or any Android reader) in landscape 2 page mode for an experience that's like reading a paperback, or portrait for textbooks/PDFs. Work beautifully in the brightest direct sun, or read with calming firelight colored backlight in the dark.

Runs M365 suite perfectly, so ideal as an all in one.

This may be a myth. Big paper in 2009, but nobody's proved it's real in the 15 years since.

In a 2009 review paper entitled Learning Styles: Concepts and Evidence, researchers investigated the “meshing hypothesis,” which is the idea that students learn better when instruction is provided in a format that matches their learning style. Their conclusion is a hard pill to swallow. “The contrast between the enormous popularity of the learning-styles approach within education and the lack of credible evidence for its utility is, in our opinion, striking and disturbing,” the researchers wrote. “If classfication of students’ learning styles has practical utility, it remains to be demonstrated.”

2009: https://www.psychologicalscience.org/news/releases/learning-...

2022: https://fee.org/articles/learning-styles-don-t-actually-exis...

APA goes so far as to say (2019) believing in learning styles is detrimental:

https://www.apa.org/news/press/releases/2019/05/learning-sty...

Read.

Not the zennial definition of listen to a podcast – actual books.

Here's how to reboot your reading.

Do this during a break, when you have a couple weeks. Or schedule a couple hours in the evenings. But make the time be a "habit" block that you require yourself to do. (Something else has to give – consider letting your streaming queue get longer for a bit. And replace your HN bookmark with https://www.daemonology.net/hn-daily/ … seriously!)

1. Start with trash, an "airport novel" type with giant print double spaced lines so the short story looks like a proper book. Pure story dopamine.

2. Then one with normal print and single spaced lines with enough substance to engage meta or systems thought as you read, maybe a Len Deighton or Eric Ambler if you're feeling able. Your brain relearns getting to think doesn't actually hurt.

3. Next shift into non-fiction. Something about Norway's heavy water sabotage against Nazi Germany, for instance; or preferably something field-adjacent to your field, with both narrative and knowledge. Re-priming your dopamine to hit from knowledge too.

4. Finally move into your own field, now knowledge you haven't been able to stick with to learn but wanted to. For this read something from a genuine practitioner, not the LinkedIn-fluencer TED talk transcripts in book form that top charts these days. Leveling up is its own reward.

Once you've made it through that, you should (a) be able to read again, (b) be able to focus on semantic processing, (c) regain your ideas to words to ideas mapping necessary for product/platform systems analysis or software engineering.

Importantly, prefer physical books or (“try this one cool trick!”) a full A4 or 8.5"x11" page e-paper tablet turned landscape, set up to read 2 pages side by side. When this is configured right (smallest margins, right font and line size) ending up same size as a paperback. The retention of two pages at a time, and the fewer page turns (fewer, larger, memory chunks) per book, is better for regaining focus and better for improving working memory. The smaller the font you can still read quickly and easily, the better, for the same reason.

Set your black mirror devices out of reach during the time block, don't have the TV on, ideally no music, or (“try this one cool trick too…”) while reading the first book on your list: play music in a language you don't understand but that sounds reasonably close to your own. For English speakers, germanic or romance languages are good. This makes your language centers busy wanting to comprehend the audio channel along with your visual channel, so your brain eats up the book without your mind wandering.

Using the method above, you can work to being able to sit with and absorb hardbound textbooks in just a series of 4 or 5 escalating challenges.

You'll find you regained intraday focus as well.

• • •

PS. Did you update that bookmark? Yesterday's HN Daily happens to be relevant as is so often the case: https://resobscura.substack.com/p/quality-non-fiction-books-...

not for steps.

Precisely.

Either way, go read his links.

Try it:

  uv run https://raw.githubusercontent.com/konstin/say-hi/refs/heads/main/say-hi.py

Friends tell friends:

  uv run

SOC2 (Type 2) in theory requires an audit that you're actually doing what you said you'd do in (Type 1).

Both may also allow general lag time.

Note that firms decide on their own which controls to include, meaning, they get to decide to include or exclude various controls, the audit is on only the ones they picked. Pick basic controls, it's cheaper to pass easily, and now you have a logo.

This means SOC2s of either type cannot be compared to one another (and SOC2 Type 1 are roughly logo-ware).

SOC3 is, roughly, SOC2 redacted.

Recap:

SOC2 Type 1 is a firm picking some controls to say they'll do them, SOC2 Type 2 is roughly a firm having someone look at whether they're doing that (warning, screenshots might suffice, audit verification is probably not what you think), and SOC3 is public or non-confidential water down of that, typically without findings.

The only thing that matters is what controls, specifically, they're actually audited on. So ideally you want to know what the controls they picked are, and that a SOC2 Type 2 was audited on those.

Btw, keep in mind that "end to end encryption" means "https" and most controls have similarly basic ways of achieving them. It's difficult to fail SOC2 Type 2 core controls if you know "don't be useless" is how you pass them.

Also, it's not $50K even through the big five DIY SOC2 Type 2 shops. You can use the same ones trillion dollar firms use, by signing up online, and you'll be surprised how inexpensive relative to the cost of failing to let a business check that box in their procurement process.

More importantly, are the customers of YOUR customers exclusively devs?

If the customers of your customers could include small businesses, then the stack you offer them should prob be capable of the simpler OIDC (Continue with...) instead of / or in addition to OAuth (Sign in with...) and probably Microsoft along with Google. And if the customers of your customers include middle class and up American consumers, also Continue with Apple.

idle = doesn't have to be H200 speeds

uses macs = local LLM can use all the ram, be smarter

as for hosted models, potato is tasty

What prompt are you using to generate this exact form of thing?

Problem history story...

Everyone's doing it wrong...

Proof of claims ...

Abstract ...

"Honest" bustedness ...

Keyword stuffing ...

Call for feedback ...

"What I'd really love from Dave:" is what's causing this particular structure. What model/harness/agent are you using, whether automated or pasting it yourself?

Consider context as hyper-dimensional coordinate vectors gesturing at the starting concept cluster of a synthesis chain or thread to unspool.

If the model's comprehensive training activated by your context locates the right thread to pull, this could be considered comprehension? That it "got" it?

they stylize it k•no•b or k•no•b1 take your pick

keyword collision: `framework`

* is in the article

> understanding anything in the way humans do

i'm not sure it's clearly established LLMs can't be a model of some part of “the way humans do”?

to be more specific, i'd argue LLMs “understand” awfully similarly to a brilliant (polymath) with early dementia or Alzheimer's

no executive function, no short term memory, and absent both of those, conversing with that person about the past or with an LLM about topics that had been "in their training sets before a cutoff date" is surprisingly similar, right down to: introing a topic precisely the same way, you'll experience the same conversation; convo loops if bits are too quantized (looping and lossiness / recall / context-length are correlated); and ofc opening a new session is like the first one never happened

1) Have you exported Apple docs from Xcode and provided those to your LLM and told your CLAUDE.md to read them as applies?

2) Have you added https://github.com/dpearson2699/swift-ios-skills to your agent?

I'd also recommend Xcode 27 beta unless you teach your agent the Xcode MCP. Xcode 27 beta works on MacOS 26.x.

To overcome it seeing your phrase and using it more, try to convey the concept you want it to say "found an indication suggesting there may be a problem underneath" in response to. That works better than telling it to do not use terms you remind it of.

Btw, you're in good company…

With both OpenAI and Anthropic, from time to time you see it stream a response to you that then changes and rewrites under your very eyes.

Those are things they, themselves, where they sit in the token supply chain, can't get it to not say until it's said it.

Early on, it mattered to say "Avoid" (an active word) "action", rather than "Do" (affirmative) "not" (negation) "action" (affirmative again).

These days I take a pass on the rules to remove the bad thing from the rule and rephrase in terms of the desired behavior instead.

So it doesn't have to try to do not think about the pink elephant.

When you see the metaphor or euphemism or simile is off, the selection is not quite apt, it seems the stock phrase or term is rather more probable however askew than the precise phase, and it was already low confidence about what it's trying to say, so – as it's never said – Bob's your uncle.

"Honestly/honest" is, um, absolutely (ahem) one of those and if one listens carefully, one will notice humans use "Honestly? Blah blah…" that way as well.

It taps into load-bearing speech to prop up (see, that works) unsupported claims.

All the words I just used are correct in that sentence, doing exactly (ahem) the work they're supposed to. It doesn't have that level of nuance, isn't sure, so tries to sound like what's probably right.

They're a verbal shrug.

On the contrary, stock words pop up more easily when it has less confidence.

Stock phrases are a correctness smell.